DeepSeek's AI Breakthrough Reshapes Global Tech Landscape

DeepSeek: how China's embrace of open-source AI caused a geopolitical earthquake

We are in the early days of a seismic shift in the global AI industry. DeepSeek, a previously little-known Chinese artificial intelligence company, has produced a "game changing"" large language model that promises to reshape the AI landscape almost overnight. But DeepSeek's breakthrough also has wider implications for the technological arms race between the US and China, having apparently caught even the best-known US tech firms off guard. Its launch has been predicted to start a "slow unwinding of the AI bet" in the west, amid a new era of "AI efficiency wars". In fact, industry experts have been speculating for years about China's rapid advancements in AI. While the supposedly free-market US has often prioritised proprietary models, China has built a thriving AI ecosystem by leveraging open-source technology, fostering collaboration between government-backed research institutions and major tech firms. This strategy has enabled China to scale its AI innovation rapidly while the US - despite all the tub-thumping from Silicon Valley - remains limited by restrictive corporate structures. Companies such as Google and Meta, despite promoting open-source initiatives, still rely heavily on closed-source strategies that limit broader access and collaboration. What makes DeepSeek particularly disruptive is its ability to achieve cutting-edge performance while reducing computing costs - an area where US firms have struggled due to their dependence on training models that demand very expensive processing hardware. Where once Silicon Valley was the epicentre of global digital innovation, its corporate behemoths now appear vulnerable to more innovative, "scrappy" startup competitors - albeit ones enabled by major state investment in AI infrastructure. By leveraging China's industrial approach to AI, DeepSeek has crystallised a reality that many in Silicon Valley have long ignored: AI's centre of power is shifting away from the US and the west. It highlights the failure of US attempts to preserve its technological hegemony through tight export controls on cutting-edge AI chips to China. According to research fellow Dean Ball: "You can keep [computing resources] away from China, but you can't export-control the ideas that everyone in the world is hunting for." DeepSeek's success has forced Silicon Valley and large western tech companies to "take stock", realising that their once-unquestioned dominance is suddenly at risk. Even the US president, Donald Trump, has proclaimed that this should be a "wake-up call for our industries that we need to be laser-focused on competing". But this story is not just about technological prowess - it could mark an important shift in global power. Former US secretary of state Mike Pompeo has framed DeepSeek's emergence as a "shot across America's bow", urging US policymakers and tech executives to take immediate action. DeepSeek's rapid rise underscores a growing realisation: globally, we are entering a potentially new AI paradigm, one where China's model of open-source innovation and state-backed development is proving more effective than Silicon Valley's corporate-driven approach. The Insights section is committed to high-quality longform journalism. Our editors work with academics from many different backgrounds who are tackling a wide range of societal and scientific challenges. I've spent much of my career analysing the transformative role of AI on the global digital landscape - examining how AI shapes governance, market structures and public discourse, and exploring its geopolitical and ethical dimensions, now and far in the future. I also have personal connections with China, having lived there while teaching at Jiangsu University, then written my PhD thesis on the country's state-led marketisation programme. Over the years, I have studied China's evolving tech landscape, observing firsthand how its unique blend of state-driven industrial policy and private-sector innovation has fuelled rapid AI development. I believe this moment may come to be seen as a turning point not just for AI, but for the geopolitical order. If China's AI dominance continues, what could this mean for the future of digital governance, democracy, and the global balance of power? China's open-source AI takeover Even in the early days of China's digital transformation, analysts predicted the country's open-source focus could lead to a major AI breakthrough. In 2018, China was integrating open-source collaboration into its broader digitisation strategy, recognising that fostering shared development efforts could accelerate its AI capabilities. Unlike the US, where proprietary AI models dominated, China embraced open-source ecosystems to bypass western gatekeeping, scale innovation faster, and embed itself in global AI collaboration. China's open-source activity surged dramatically in 2020, laying the foundation for the kind of innovation seen today. By actively fostering an open-source culture, China ensured that a broad range of developers had access to AI tools, rather than restricting them to a handful of dominant companies. The trend has continued in recent years, with China even launching its own state-backed open-source operating systems and platforms in 2023, to further reduce its dependence on western technology. This move was widely seen as an effort to cement its AI leadership and create an independent, self-sustaining digital ecosystem. While China has been steadily positioning itself as a leader in open-source AI, Silicon Valley firms remained focused on closed, proprietary models - allowing China to catch up fast. While companies like Google and Meta promoted open-source initiatives in name, they still locked key AI capabilities behind paywalls and restrictive licenses. In contrast, China's government-backed initiatives have treated open-source AI as a national resource, rather than a corporate asset. This has resulted in China becoming one of the world's largest contributors to open-source AI development, surpassing many western firms in collaborative projects. Chinese tech giants such as Huawei, Alibaba and Tencent are driving open-source AI forward with frameworks like PaddlePaddle, X-Deep Learning (X-DL) and MindSpore -- all now core to China's machine learning ecosystem. But they're also making major contributions to global AI projects, from Alibaba's Dragonfly, which streamlines large-scale data distribution, to Baidu's Apollo, an open-source platform accelerating autonomous vehicle development. These efforts don't just strengthen China's AI industry, they embed it deeper into the global AI landscape. Read more: Putting DeepSeek to the test: how its performance compares against other AI tools This shift had been years in the making, as Chinese firms (with state backing) pushed open-source AI forward and made their models publicly available, creating a feedback loop that western companies have also - quietly - tapped into. A year ago, for example, US firm Abicus.AI released Smaug-72B, an AI model designed for enterprises that built directly upon Alibaba's Qwen-72B and outperformed proprietary models like OpenAI's GPT-3.5 and Mistral's Medium. But the potential for US companies to further build on Chinese open-source technology may be limited by political as well as corporate barriers. In 2023, US lawmakers highlighted growing concerns that China's aggressive investment in open-source AI and semiconductor technologies would eventually erode western leadership in AI. Some policymakers called for bans on certain open-source chip technologies, due to fears they could further accelerate China's AI advancements. But by then, China's AI horse had already bolted. AI with Chinese characteristics DeepSeek's rise should have been obvious to anyone familiar with management theory and the history of technological breakthroughs linked to "disruptive innovation". Latecomers to an industry rarely compete by playing the same game as incumbents - they have to be disruptive. China, facing restrictions on cutting-edge western AI chips and lagging behind in proprietary AI infrastructure, had no choice but to innovate differently. Open-source AI provided the perfect vehicle: a way to scale innovation rapidly, lower costs and tap into global research while bypassing Silicon Valley's resource-heavy, closed-source model. From a western and traditional human rights perspective, China's embrace of open-source AI may appear paradoxical, given the country's strict information controls. Its AI development strategy prioritises both technological advancement and strict alignment with the Chinese Communist party's ideological framework, ensuring AI models adhere to "core socialist values" and state-approved narratives. AI research in China has thrived not only despite these constraints but, in many ways, because of them. China's success goes beyond traditional authoritarianism; it embodies what Harvard economist David Yang calls "Autocracy 2.0". Rather than relying solely on fear-based control, it uses economic incentives, bureaucratic efficiency, and technology to manage information and maintain regime stability. The Chinese government has strategically encouraged open-source development while maintaining tight control over AI's domestic applications, particularly in surveillance and censorship. Indeed, authoritarian regimes may have a significant advantage in developing facial-recognition technology due to their extensive surveillance systems. The vast amounts of data collected through these networks enable private AI companies to create advanced algorithms, which can then be adapted for commercial uses, potentially accelerating economic growth. China's AI strategy is built on a dual foundation of state-led initiatives and private-sector innovation. The country's AI roadmap, first outlined in the 2017 new generation artificial intelligence development plan, follows a three-phase timeline: achieving global competitiveness by 2020, making major AI breakthroughs by 2025, and securing world leadership in AI by 2030. In parallel, the government has emphasised data governance, regulatory frameworks and ethical oversight to guide AI development "responsibly". A defining feature of China's AI expansion has been the massive infusion of state-backed investment. Over the past decade, government venture capital funds have injected approximately US$912 billion (£737bn) into early-stage firms, with 23% of that funding directed toward AI-related companies. A significant portion has targeted China's less-developed regions, following local investment mandates. Read more: Three lessons the west can learn from China's economic approach to AI Compared with private venture capital, government-backed firms often lag in software development but demonstrate rapid growth post-investment. Moreover, state funding often serves as a signal for subsequent private-sector investment, reinforcing the country's AI ecosystem. China's AI strategy represents a departure from its traditional industrial policies, which historically emphasised self-sufficiency, support for a handful of national champions, and military-driven research. Instead, the government has embraced a more flexible and collaborative approach that encourages open-source software adoption, a diverse network of AI firms, and public-private partnerships to accelerate innovation. This model prioritises research funding, state-backed AI laboratories, and AI integration across key industries including security, healthcare, and infrastructure. Despite strong state involvement, China's AI boom is equally driven by private-sector innovation. The country is home to an estimated 4,500 AI companies, accounting for 15% of the world's total. As economist Liu Gang told the Chinese Communist Party's Global Times newspaper: "The development of AI is fast in China - for example, for AI-empowered large language models. Aided with government spending, private capital is flowing to the new sector. Increased capital inflow is anticipated to further enhance the sector in 2025." China's tech giants including Baidu, Alibaba, Tencent and SenseTime have all benefited from substantial government support while remaining competitive on the global stage. But unlike in the US, China's AI ecosystem thrives on a complex interplay between state support, corporate investment and academic collaboration. Recognising the potential of open-source AI early on, Tsinghua University in Beijing has emerged as a key innovation hub, producing leading AI startups such as Zhipu AI, Baichuan AI, Moonshot AI and MiniMax -- all founded by its faculty and alumni. The Chinese Academy of Sciences has similarly played a crucial role in advancing research in deep learning and natural language processing. Unlike the west, where companies like Google and Meta promote open-source models for strategic business gains, China sees them as a means of national technological self-sufficiency. To this end, the National AI Team, composed of 23 leading private enterprises, has developed the National AI Open Innovation Platform, which provides open access to AI datasets, toolkits, libraries and other computing resources. DeepSeek is a prime example of China's AI strategy in action. The company's rise embodies the government's push for open-source collaboration while remaining deeply embedded within a state-guided AI ecosystem. Chinese developers have long been major contributors to open-source platforms, ranking as the second-largest group on GitHub by 2021. Founded by Chinese entrepreneur Liang Wenfeng in 2023, DeepSeek has positioned itself as an AI leader while benefiting from China's state-driven AI ecosystem. Liang, who also established the hedge fund High-Flyer, has maintained full ownership of DeepSeek and avoided external venture capital funding. Though there is no direct evidence of government financial backing, DeepSeek has reaped the rewards of China's AI talent pipeline, state-sponsored education programs, and research funding. Liang has engaged with top government officials including China's premier, Li Qiang, reflecting the company's strategic importance to the country's broader AI ambitions. In this way, DeepSeek perfectly encapsulates "AI with Chinese characteristics" - a fusion of state guidance, private-sector ingenuity, and open-source collaboration, all carefully managed to serve the country's long-term technological and geopolitical objectives. Recognising the strategic value of open-source innovation, the government has actively promoted domestic open-source code platforms like Gitee to foster self-reliance and insulate China's AI ecosystem from external disruptions. However, this also exposes the limits of China's open-source ambitions. The government pushes collaboration, but only within a tightly controlled system where state-backed firms and tech giants call the shots. Reports of censorship on Gitee reveal how Beijing carefully manages innovation, ensuring AI advances stay in line with national priorities. Independent developers can contribute, but the real power remains concentrated in companies that operate within the government's strategic framework. The conflicted reactions of US big tech DeepSeek's emergence has sparked intense debate across the AI industry, drawing a range of reactions from leading Silicon Valley executives, policymakers and researchers. While some view it as an expected evolution of open-source AI, others see it as a direct challenge to western AI leadership. Microsoft's CEO, Satya Nadella, emphasised its technical efficiency. "It's super-impressive in terms of both how they have really effectively done an open-source model that does this inference-time compute, and is super-compute efficient," Nadella told CNBC. "We should take the developments out of China very, very seriously". Silicon Valley venture capitalist Marc Andreessen, a prominent advisor to Trump, was similarly effusive. "DeepSeek R1 is one of the most amazing and impressive breakthroughs I've ever seen - and as open source, a profound gift to the world," he wrote on X. For Yann LeCun, Meta's chief AI scientist, DeepSeek is less about China's AI capabilities and more about the broader power of open-source innovation. He argued that the situation should be read not as China's AI surpassing the US, but rather as open-source models surpassing proprietary ones. "DeepSeek has profited from open research and open source (e.g. PyTorch and Llama from Meta)," he wrote on Threads. "They came up with new ideas and built them on top of other people's work. Because their work is published and open source, everyone can profit from it. That is the power of open research and open source." Not all responses were so measured. Alexander Wang, CEO of Scale AI - a US firm specialising in AI data labelling and model training - framed DeepSeek as a competitive threat that demands an aggressive response. He wrote on X: "DeepSeek is a wake-up call for America, but it doesn't change the strategy: USA must out-innovate & race faster, as we have done in the entire history of AI. Tighten export controls on chips so that we can maintain future leads. Every major breakthrough in AI has been American." Elon Musk added fuel to speculation about DeepSeek's hardware access when he responded with a simple "obviously" to Wang's earlier claims on CNBC that DeepSeek had secretly acquired 50,000 Nvidia H100 GPUs, despite US export restrictions. Beyond the tech world, US policymakers have taken a more adversarial stance. House speaker Mike Johnson accused China of leveraging DeepSeek to erode American AI leadership. "They abuse the system, they steal our intellectual property. They're now trying to get a leg up on us in AI." For his part, Trump took a more pragmatic view, seeing DeepSeek's efficiency as a validation of cost-cutting approaches. "I view that as a positive, as an asset ... You won't be spending as much, and you'll get the same result, hopefully." The rise of DeepSeek may have helped jolt the Trump administration into action, leading to sweeping policy shifts aimed at securing US dominance in AI. In his first week back in the White House, the US president announced a series of aggressive measures, including massive federal investments in AI research, closer partnerships between the government and private tech firms, and the rollback of regulations seen as slowing US innovation. The administration's framing of AI as a critical national interest reflects a broader urgency sparked by China's rapid advancements, particularly DeepSeek's ability to produce cutting-edge models at a fraction of the cost traditionally associated with AI development. But this response is not just about national competitiveness - it is also deeply entangled with private industry. Musk's growing closeness to Trump, for example, can be viewed as a calculated move to protect his own dominance at home and abroad. By aligning with the administration, Musk ensures that US policy tilts in favour of his AI ventures, securing access to government backing, computing power, and regulatory control over AI exports. At the same time, Musk's public criticism of Trump's US$500 billion AI infrastructure plan - claiming the companies involved lack the necessary funding - was as much a warning as a dismissal, signalling his intent to shape policy in a way that benefits his empire while keeping potential challengers at bay. Not unrelated, Musk and a group of investors have just launched a US$97.4 billion (£78.7bn) bid for OpenAI's nonprofit arm, a move that escalates his feud with OpenAI CEO Sam Altman and seeks to strengthen his grip on the AI industry. Altman has dismissed the bid as a "desperate power grab", insisting that OpenAI will not be swayed by Musk's attempts to reclaim control. The spat reflects how DeepSeek's emergence has thrown US tech giants into what could be all-out war, fuelling bitter corporate rivalries and reshaping the fight for AI dominance. And while the US and China escalate their AI competition, other global leaders are pushing for a coordinated response. The Paris AI Action Summit, held on February 10 and 11, has become a focal point for efforts to prevent AI from descending into an uncontrolled power struggle. France's president, Emmanuel Macron, warned delegates that without international oversight, AI risks becoming "the wild west", where unchecked technological development creates instability rather than progress. But at the end of the two-day summit, the UK and US refused to sign an international commitment to "ensuring AI is open, inclusive, transparent, ethical, safe, secure and trustworthy ... making AI sustainable for people and the planet". China was among the 61 countries to sign this declaration. Concerns have also been raised at the summit about how AI-powered surveillance and control are enabling authoritarian regimes to strengthen repression and reshape the citizen-state relationship. This highlights the fast-growing global industry of digital repression, driven by an emerging "authoritarian-financial complex" that may exacerbate China's strategic advancement in AI. Equally, DeepSeek's cost-effective AI solutions have created an opening for European firms to challenge the traditional AI hierarchy. As AI development shifts from being solely about compute power to strategic efficiency and accessibility, European firms now have an opportunity to compete more aggressively against their US and Chinese counterparts. Whether this marks a true rebalancing of the AI landscape remains to be seen. But DeepSeek's emergence has certainly upended traditional assumptions about who will lead the next wave of AI innovation - and how global powers will respond to it. End of the 'Silicon Valley effect'? DeepSeek's emergence has forced US tech leaders to confront an uncomfortable reality: they underestimated China's AI capabilities. Confident in their perceived lead, companies like Google, Meta, and OpenAI prioritised incremental improvements over anticipating disruptive competition, leaving them vulnerable to a rapidly evolving global AI landscape. In response, the US tech giants are now scrambling to defend their dominance, pledging over US$400 billion in AI investment. DeepSeek's rise, fuelled by open-source collaboration, has reignited fierce debates over innovation versus security, while its energy-efficient model has intensified scrutiny on AI's sustainability. Yet Silicon Valley continues to cling to what many view as outdated economic theories such as the Jevons paradox to downplay China's AI surge, insisting that greater efficiency will only fuel demand for computing power and reinforce their dominance. Companies like Meta, OpenAI and Microsoft remain fixated on scaling computational power, betting that expensive hardware will secure their lead. But this assumption blinds them to a shifting reality. DeepSeek's rise as the potential "Walmart of AI" is shaking Silicon Valley's foundation, proving that high-quality AI models can be built at a fraction of the cost. By prioritising efficiency over brute-force computing power, DeepSeek is challenging the US tech industry's reliance on expensive hardware like Nvidia's high-end chips. This shift has already rattled markets, driving down the stock prices of major US firms and forcing a reassessment of AI dominance. Nvidia, whose business depends on supplying high-performance processors, appears particularly vulnerable as DeepSeek's cost-effective approach threatens to reduce demand for premium chips. The growing divide between the US and China in AI, however, is more than just competition - it's a clash of governance models. While US firms remain fixated on protecting market dominance, China is accelerating AI innovation with a model that is proving more adaptable to global competition. If Silicon Valley resists structural change, it risks falling further behind. We may witness the unravelling of the "Silicon Valley effect", through which tech giants have long manipulated AI regulations to entrench their dominance. For years, Google, Meta,and OpenAI shaped policies that favoured proprietary models and costly infrastructure, ensuring AI development remained under their control. DeepSeek is redefining AI with breakthroughs in code intelligence, vision-language models and efficient architectures that challenge Silicon Valley's dominance. By optimising computation and embracing open-source collaboration, DeepSeek shows the potential of China to deliver cutting-edge models at a fraction of the cost, outperforming proprietary alternatives in programming, reasoning and real-world applications. More than a policy-driven rise, China's AI surge reflects a fundamentally different innovation model - fast, collaborative and market-driven - while Silicon Valley holds on to expensive infrastructure and rigid proprietary control. If US firms refuse to adapt, they risk losing the future of AI to a more agile and cost-efficient competitor. A new era of geotechnopolitics But China is not just disrupting Silicon Valley. It is expanding "geotechnopolitics", where AI is a battleground for global power. With AI projected to add US$15.7 trillion to the global economy by 2030, China and the US are racing to control the technology that will define economic, military and political dominance. DeepSeek's advancement has raised national security concerns in the US. Trump's government is considering stricter export controls on AI-related technologies to prevent them from bolstering China's military and intelligence capabilities. As AI-driven defence systems, intelligence operations and cyber warfare redefine national security, governments must confront a new reality: AI leadership is not just about technological superiority, but about who controls the intelligence that will shape the next era of global power. China's AI ambitions extend beyond technology, driving a broader strategy for economic and geopolitical dominance. But with over 50 state-backed companies developing large-scale AI models, its rapid expansion faces growing challenges, including soaring energy demands and US semiconductor restrictions. China's president, Xi Jinping, remains resolute, stating: "Whoever can grasp the opportunities of new economic development such as big data and artificial intelligence will have the pulse of our times." He sees AI driving "new quality productivity" and modernising China's manufacturing base, calling its "head goose effect" a catalyst for broader innovation. To counter western containment, China has embraced a "guerrilla" economic strategy, bypassing restrictions through alternative trade networks, deepening ties with the global south, and exploiting weaknesses in global supply chains. Instead of direct confrontation, this decentralised approach uses economic coercion to weaken adversaries while securing China's own industrial base. China is also leveraging open-source AI as an ideological tool, presenting its model as more collaborative and accessible than western alternatives. This narrative strengthens its global influence, aligning with nations seeking alternatives to western digital control. While strict state oversight remains, China's embrace of open-source AI reinforces its claim to a future where innovation is driven not by corporate interests but through shared collaboration and global cooperation. But while DeepSeek claims to be open access, its secrecy tells a different story. Key details on training data and fine-tuning remain hidden, and its compliance with China's AI laws has sparked global scrutiny. Italy has banned the platform over data-transfer risks, while Belgium and Ireland launched privacy probes. Under Chinese regulations, DeepSeek's outputs must align with state-approved narratives, clashing with the EU's AI Act, which demands transparency and protects political speech. Such "controlled openness" raises many red flags, casting doubt on China's place in markets that value data security and free expression. Many western commentators are seizing on reports of Chinese AI censorship to frame other models as freer and more politically open. The revelation that a leading Chinese chatbot actively modifies or censors responses in real time has fuelled a broader narrative that western AI operates without such restrictions, reinforcing the idea that democratic systems produce more transparent and unbiased technology. This framing serves to bolster the argument that free societies will ultimately lead the global AI race. But at its heart, the "AI arms race" is driven by technological dominance. The US, China, and the EU are charting different paths, weighing security risks against the need for global collaboration. How this competition is framed will shape policy: lock AI behind restrictions, or push for open innovation. DeepSeek, for all its transformational qualities, continues to exemplify a model of AI where innovation prioritises scale, speed and efficiency over societal impact. This drive to optimise computation and expand capabilities overshadows the need to design AI as a truly public good. In doing so, it eclipses this technology's genuine potential to transform governance, public services and social institutions in ways that prioritise collective wellbeing, equity and sustainability over corporate and state control. A truly global AI framework requires more than political or technological openness. It demands structured cooperation that prioritises shared governance, equitable access, and responsible development. Following a workshop in Shanghai hosted by the Chinese government last September, the UN's general secretary, António Guterres, outlined his vision for AI beyond corporate or state control: "We must seize this historic opportunity to lay the foundations for inclusive governance of AI - for the benefit of all humanity. As we build AI capacity, we must also develop shared knowledge and digital public goods." Both the west and China frame their AI ambitions through competing notions of "openness" - each aligning with their strategic interests and reinforcing existing power structures. Western tech giants claim AI drives democratisation, yet they often dominate digital infrastructure in parts of Africa, Asia and Latin America, exporting models based on "corporate imperialism" that extract value while disregarding local needs. China, by contrast, positions itself as a technological partner for the rest of the global south; however, its AI remains tightly controlled, reinforcing state ideology. China's proclaimed view on international AI collaboration emphasises that AI should not be "a game of rich countries"", as President Xi stated during the 2024 G20 summit. By advocating for inclusive global AI development, China positions itself as a leader in shaping international AI governance, especially via initiatives like the UN AI resolution and its AI capacity-building action plan. These efforts help promote a more balanced technological landscape while allowing China to strengthen its influence in global AI standards and frameworks. However, beneath all these narratives, both China and the US share a strategy of AI expansion that relies on exploited human labour, from data annotation to moderation, exposing a system driven less by innovation than by economic and political control. Seeing AI as a connected race for influence highlights the need for ethical deployment, cross-border cooperation, and a balance between security and progress. And this is where China may face its greatest challenge - balancing the power of open-source innovation with the constraints of a tightly controlled, authoritarian system that thrives on restriction, rather than openness. For you: more from our Insights series: To understand the risks posed by AI, follow the money Sex machina: in the wild west world of human-AI relationships, the lonely and vulnerable are most at risk Novelist J.G. Ballard was experimenting with computer-generated poetry 50 years before ChatGPT was invented The brain is the most complicated object in the universe. This is the story of scientists' quest to decode it - and read people's minds To hear about new Insights articles, join the hundreds of thousands of people who value The Conversation's evidence-based news. Subscribe to our newsletter.

DeepSeek: How China's embrace of open-source AI caused a geopolitical earthquake

We are in the early days of a seismic shift in the global AI industry. DeepSeek, a previously little-known Chinese artificial intelligence company, has produced a "game changing"" large language model that promises to reshape the AI landscape almost overnight. But DeepSeek's breakthrough also has wider implications for the technological arms race between the US and China, having apparently caught even the best-known US tech firms off guard. Its launch has been predicted to start a "slow unwinding of the AI bet" in the west, amid a new era of "AI efficiency wars". In fact, industry experts have been speculating for years about China's rapid advancements in AI. While the supposedly free-market US has often prioritized proprietary models, China has built a thriving AI ecosystem by leveraging open-source technology, fostering collaboration between government-backed research institutions and major tech firms. This strategy has enabled China to scale its AI innovation rapidly while the US -- despite all the tub-thumping from Silicon Valley -- remains limited by restrictive corporate structures. Companies such as Google and Meta, despite promoting open-source initiatives, still rely heavily on closed-source strategies that limit broader access and collaboration. What makes DeepSeek particularly disruptive is its ability to achieve cutting-edge performance while reducing computing costs -- an area where US firms have struggled due to their dependence on training models that demand very expensive processing hardware. Where once Silicon Valley was the epicenter of global digital innovation, its corporate behemoths now appear vulnerable to more innovative, "scrappy" startup competitors -- albeit ones enabled by major state investment in AI infrastructure. By leveraging China's industrial approach to AI, DeepSeek has crystallized a reality that many in Silicon Valley have long ignored: AI's center of power is shifting away from the US and the west. It highlights the failure of US attempts to preserve its technological hegemony through tight export controls on cutting-edge AI chips to China. According to research fellow Dean Ball: "You can keep [computing resources] away from China, but you can't export-control the ideas that everyone in the world is hunting for." DeepSeek's success has forced Silicon Valley and large western tech companies to "take stock," realizing that their once-unquestioned dominance is suddenly at risk. Even the US president, Donald Trump, has proclaimed that this should be a "wake-up call for our industries that we need to be laser-focused on competing." But this story is not just about technological prowess -- it could mark an important shift in global power. Former US secretary of state Mike Pompeo has framed DeepSeek's emergence as a "shot across America's bow," urging US policymakers and tech executives to take immediate action. DeepSeek's rapid rise underscores a growing realization: globally, we are entering a potentially new AI paradigm, one where China's model of open-source innovation and state-backed development is proving more effective than Silicon Valley's corporate-driven approach. I've spent much of my career analyzing the transformative role of AI in the global digital landscape -- examining how AI shapes governance, market structures and public discourse, and exploring its geopolitical and ethical dimensions, now and far in the future. I also have personal connections with China, having lived there while teaching at Jiangsu University, then written my Ph.D. thesis on the country's state-led marketization program. Over the years, I have studied China's evolving tech landscape, observing firsthand how its unique blend of state-driven industrial policy and private-sector innovation has fueled rapid AI development. I believe this moment may come to be seen as a turning point not just for AI, but for the geopolitical order. If China's AI dominance continues, what could this mean for the future of digital governance, democracy, and the global balance of power? China's open-source AI takeover Even in the early days of China's digital transformation, analysts predicted the country's open-source focus could lead to a major AI breakthrough. In 2018, China was integrating open-source collaboration into its broader digitization strategy, recognizing that fostering shared development efforts could accelerate its AI capabilities. Unlike the US, where proprietary AI models dominated, China embraced open-source ecosystems to bypass western gatekeeping, scale innovation faster, and embed itself in global AI collaboration. China's open-source activity surged dramatically in 2020, laying the foundation for the kind of innovation seen today. By actively fostering an open-source culture, China ensured that a broad range of developers had access to AI tools, rather than restricting them to a handful of dominant companies. The trend has continued in recent years, with China even launching its own state-backed open-source operating systems and platforms in 2023, to further reduce its dependence on western technology. This move was widely seen as an effort to cement its AI leadership and create an independent, self-sustaining digital ecosystem. While China has been steadily positioning itself as a leader in open-source AI, Silicon Valley firms remained focused on closed, proprietary models -- allowing China to catch up fast. While companies like Google and Meta promoted open-source initiatives in name, they still locked key AI capabilities behind paywalls and restrictive licenses. In contrast, China's government-backed initiatives have treated open-source AI as a national resource, rather than a corporate asset. This has resulted in China becoming one of the world's largest contributors to open-source AI development, surpassing many western firms in collaborative projects. Chinese tech giants such as Huawei, Alibaba and Tencent are driving open-source AI forward with frameworks like PaddlePaddle, X-Deep Learning (X-DL) and MindSpore -- all now core to China's machine learning ecosystem. But they're also making major contributions to global AI projects, from Alibaba's Dragonfly, which streamlines large-scale data distribution, to Baidu's Apollo, an open-source platform accelerating autonomous vehicle development. These efforts don't just strengthen China's AI industry, they embed it deeper into the global AI landscape. This shift had been years in the making, as Chinese firms (with state backing) pushed open-source AI forward and made their models publicly available, creating a feedback loop that western companies have also -- quietly -- tapped into. A year ago, for example, US firm Abicus.AI released Smaug-72B, an AI model designed for enterprises that built directly upon Alibaba's Qwen-72B and outperformed proprietary models like OpenAI's GPT-3.5 and Mistral's Medium. But the potential for US companies to further build on Chinese open-source technology may be limited by political as well as corporate barriers. In 2023, US lawmakers highlighted growing concerns that China's aggressive investment in open-source AI and semiconductor technologies would eventually erode western leadership in AI. Some policymakers called for bans on certain open-source chip technologies, due to fears they could further accelerate China's AI advancements. DeepSeek's rise should have been obvious to anyone familiar with management theory and the history of technological breakthroughs linked to "disruptive innovation." Latecomers to an industry rarely compete by playing the same game as incumbents -- they have to be disruptive. China, facing restrictions on cutting-edge western AI chips and lagging behind in proprietary AI infrastructure, had no choice but to innovate differently. Open-source AI provided the perfect vehicle: a way to scale innovation rapidly, lower costs and tap into global research while bypassing Silicon Valley's resource-heavy, closed-source model. From a western and traditional human rights perspective, China's embrace of open-source AI may appear paradoxical, given the country's strict information controls. Its AI development strategy prioritizes both technological advancement and strict alignment with the Chinese Communist party's ideological framework, ensuring AI models adhere to "core socialist values" and state-approved narratives. AI research in China has thrived not only despite these constraints but, in many ways, because of them. China's success goes beyond traditional authoritarianism; it embodies what Harvard economist David Yang calls "Autocracy 2.0." Rather than relying solely on fear-based control, it uses economic incentives, bureaucratic efficiency, and technology to manage information and maintain regime stability. The Chinese government has strategically encouraged open-source development while maintaining tight control over AI's domestic applications, particularly in surveillance and censorship. Indeed, authoritarian regimes may have a significant advantage in developing facial-recognition technology due to their extensive surveillance systems. The vast amounts of data collected through these networks enable private AI companies to create advanced algorithms, which can then be adapted for commercial uses, potentially accelerating economic growth. China's AI strategy is built on a dual foundation of state-led initiatives and private-sector innovation. The country's AI roadmap, first outlined in the 2017 new generation artificial intelligence development plan, follows a three-phase timeline: achieving global competitiveness by 2020, making major AI breakthroughs by 2025, and securing world leadership in AI by 2030. In parallel, the government has emphasized data governance, regulatory frameworks and ethical oversight to guide AI development "responsibly." A defining feature of China's AI expansion has been the massive infusion of state-backed investment. Over the past decade, government venture capital funds have injected approximately US$912 billion (£737bn) into early-stage firms, with 23% of that funding directed toward AI-related companies. A significant portion has targeted China's less-developed regions, following local investment mandates. Compared with private venture capital, government-backed firms often lag in software development but demonstrate rapid growth post-investment. Moreover, state funding often serves as a signal for subsequent private-sector investment, reinforcing the country's AI ecosystem. China's AI strategy represents a departure from its traditional industrial policies, which historically emphasized self-sufficiency, support for a handful of national champions, and military-driven research. Instead, the government has embraced a more flexible and collaborative approach that encourages open-source software adoption, a diverse network of AI firms, and public-private partnerships to accelerate innovation. This model prioritizes research funding, state-backed AI laboratories, and AI integration across key industries including security, health care, and infrastructure. Despite strong state involvement, China's AI boom is equally driven by private-sector innovation. The country is home to an estimated 4,500 AI companies, accounting for 15% of the world's total. As economist Liu Gang told the Chinese Communist Party's Global Times newspaper: "The development of AI is fast in China -- for example, for AI-empowered large language models. Aided with government spending, private capital is flowing to the new sector. Increased capital inflow is anticipated to further enhance the sector in 2025." China's tech giants, including Baidu, Alibaba, Tencent and SenseTime, have all benefited from substantial government support while remaining competitive on the global stage. But unlike in the US, China's AI ecosystem thrives on a complex interplay between state support, corporate investment and academic collaboration. Recognizing the potential of open-source AI early on, Tsinghua University in Beijing has emerged as a key innovation hub, producing leading AI startups such as Zhipu AI, Baichuan AI, Moonshot AI and MiniMax -- all founded by its faculty and alumni. The Chinese Academy of Sciences has similarly played a crucial role in advancing research in deep learning and natural language processing. Unlike the west, where companies like Google and Meta promote open-source models for strategic business gains, China sees them as a means of national technological self-sufficiency. To this end, the National AI Team, composed of 23 leading private enterprises, has developed the National AI Open Innovation Platform, which provides open access to AI datasets, toolkits, libraries and other computing resources. DeepSeek is a prime example of China's AI strategy in action. The company's rise embodies the government's push for open-source collaboration while remaining deeply embedded within a state-guided AI ecosystem. Chinese developers have long been major contributors to open-source platforms, ranking as the second-largest group on GitHub by 2021. Founded by Chinese entrepreneur Liang Wenfeng in 2023, DeepSeek has positioned itself as an AI leader while benefiting from China's state-driven AI ecosystem. Liang, who also established the hedge fund High-Flyer, has maintained full ownership of DeepSeek and avoided external venture capital funding. Though there is no direct evidence of government financial backing, DeepSeek has reaped the rewards of China's AI talent pipeline, state-sponsored education programs, and research funding. Liang has engaged with top government officials including China's premier, Li Qiang, reflecting the company's strategic importance to the country's broader AI ambitions. In this way, DeepSeek perfectly encapsulates "AI with Chinese characteristics" -- a fusion of state guidance, private-sector ingenuity, and open-source collaboration, all carefully managed to serve the country's long-term technological and geopolitical objectives. Recognizing the strategic value of open-source innovation, the government has actively promoted domestic open-source code platforms like Gitee to foster self-reliance and insulate China's AI ecosystem from external disruptions. However, this also exposes the limits of China's open-source ambitions. The government pushes collaboration, but only within a tightly controlled system where state-backed firms and tech giants call the shots. Reports of censorship on Gitee reveal how Beijing carefully manages innovation, ensuring AI advances stay in line with national priorities. Independent developers can contribute, but the real power remains concentrated in companies that operate within the government's strategic framework. The conflicted reactions of US big tech DeepSeek's emergence has sparked intense debate across the AI industry, drawing a range of reactions from leading Silicon Valley executives, policymakers and researchers. While some view it as an expected evolution of open-source AI, others see it as a direct challenge to western AI leadership. Microsoft's CEO, Satya Nadella, emphasized its technical efficiency. "It's super-impressive in terms of both how they have really effectively done an open-source model that does this inference-time compute, and is super-compute efficient," Nadella told CNBC. "We should take the developments out of China very, very seriously." Silicon Valley venture capitalist Marc Andreessen, a prominent advisor to Trump, was similarly effusive. "DeepSeek R1 is one of the most amazing and impressive breakthroughs I've ever seen -- and as open source, a profound gift to the world," he wrote on X. For Yann LeCun, Meta's chief AI scientist, DeepSeek is less about China's AI capabilities and more about the broader power of open-source innovation. He argued that the situation should be read not as China's AI surpassing the US, but rather as open-source models surpassing proprietary ones. "DeepSeek has profited from open research and open source (e.g. PyTorch and Llama from Meta)," he wrote on Threads. "They came up with new ideas and built them on top of other people's work. Because their work is published and open source, everyone can profit from it. That is the power of open research and open source." Not all responses were so measured. Alexander Wang, CEO of Scale AI -- a US firm specializing in AI data labeling and model training -- framed DeepSeek as a competitive threat that demands an aggressive response. He wrote on X: "DeepSeek is a wake-up call for America, but it doesn't change the strategy: U.S. must out-innovate & race faster, as we have done in the entire history of AI. Tighten export controls on chips so that we can maintain future leads. Every major breakthrough in AI has been American." Elon Musk added fuel to speculation about DeepSeek's hardware access when he responded with a simple "obviously" to Wang's earlier claims on CNBC that DeepSeek had secretly acquired 50,000 Nvidia H100 GPUs, despite US export restrictions. Beyond the tech world, US policymakers have taken a more adversarial stance. House speaker Mike Johnson accused China of leveraging DeepSeek to erode American AI leadership: "They abuse the system, they steal our intellectual property. They're now trying to get a leg up on us in AI." For his part, Trump took a more pragmatic view, seeing DeepSeek's efficiency as a validation of cost-cutting approaches: "I view that as a positive, as an asset ... You won't be spending as much, and you'll get the same result, hopefully." The rise of DeepSeek may have helped jolt the Trump administration into action, leading to sweeping policy shifts aimed at securing US dominance in AI. In his first week back in the White House, the US president announced a series of aggressive measures, including massive federal investments in AI research, closer partnerships between the government and private tech firms, and the rollback of regulations seen as slowing US innovation. The administration's framing of AI as a critical national interest reflects a broader urgency sparked by China's rapid advancements, particularly DeepSeek's ability to produce cutting-edge models at a fraction of the cost traditionally associated with AI development. But this response is not just about national competitiveness -- it is also deeply entangled with private industry. Musk's growing closeness to Trump, for example, can be viewed as a calculated move to protect his own dominance at home and abroad. By aligning with the administration, Musk ensures that US policy tilts in favor of his AI ventures, securing access to government backing, computing power, and regulatory control over AI exports. At the same time, Musk's public criticism of Trump's US$500 billion AI infrastructure plan -- claiming the companies involved lack the necessary funding -- was as much a warning as a dismissal, signaling his intent to shape policy in a way that benefits his empire while keeping potential challengers at bay. Not unrelated, Musk and a group of investors have just launched a US$97.4 billion (£78.7bn) bid for OpenAI's nonprofit arm, a move that escalates his feud with OpenAI CEO Sam Altman and seeks to strengthen his grip on the AI industry. Altman has dismissed the bid as a "desperate power grab," insisting that OpenAI will not be swayed by Musk's attempts to reclaim control. The spat reflects how DeepSeek's emergence has thrown US tech giants into what could be an all-out war, fueling bitter corporate rivalries and reshaping the fight for AI dominance. And while the US and China escalate their AI competition, other global leaders are pushing for a coordinated response. The Paris AI Action Summit, held on February 10 and 11, has become a focal point for efforts to prevent AI from descending into an uncontrolled power struggle. France's president, Emmanuel Macron, warned delegates that without international oversight, AI risks becoming "the wild west," where unchecked technological development creates instability rather than progress. But at the end of the two-day summit, the UK and US refused to sign an international commitment to "ensuring AI is open, inclusive, transparent, ethical, safe, secure and trustworthy ... making AI sustainable for people and the planet." China was among the 61 countries to sign this declaration. Concerns have also been raised at the summit about how AI-powered surveillance and control are enabling authoritarian regimes to strengthen repression and reshape the citizen-state relationship. This highlights the fast-growing global industry of digital repression, driven by an emerging "authoritarian-financial complex" that may exacerbate China's strategic advancement in AI. Equally, DeepSeek's cost-effective AI solutions have created an opening for European firms to challenge the traditional AI hierarchy. As AI development shifts from being solely about compute power to strategic efficiency and accessibility, European firms now have an opportunity to compete more aggressively against their US and Chinese counterparts. Whether this marks a true rebalancing of the AI landscape remains to be seen. But DeepSeek's emergence has certainly upended traditional assumptions about who will lead the next wave of AI innovation -- and how global powers will respond to it. End of the 'Silicon Valley effect?' DeepSeek's emergence has forced US tech leaders to confront an uncomfortable reality: they underestimated China's AI capabilities. Confident in their perceived lead, companies like Google, Meta, and OpenAI prioritized incremental improvements over anticipating disruptive competition, leaving them vulnerable to a rapidly evolving global AI landscape. In response, the US tech giants are now scrambling to defend their dominance, pledging over US$400 billion in AI investment. DeepSeek's rise, fueled by open-source collaboration, has reignited fierce debates over innovation versus security, while its energy-efficient model has intensified scrutiny of AI's sustainability. Yet Silicon Valley continues to cling to what many view as outdated economic theories such as the Jevons paradox to downplay China's AI surge, insisting that greater efficiency will only fuel demand for computing power and reinforce their dominance. Companies like Meta, OpenAI and Microsoft remain fixated on scaling computational power, betting that expensive hardware will secure their lead. But this assumption blinds them to a shifting reality. DeepSeek's rise as the potential "Walmart of AI" is shaking Silicon Valley's foundation, proving that high-quality AI models can be built at a fraction of the cost. By prioritizing efficiency over brute-force computing power, DeepSeek is challenging the US tech industry's reliance on expensive hardware like Nvidia's high-end chips. This shift has already rattled markets, driving down the stock prices of major US firms and forcing a reassessment of AI dominance. Nvidia, whose business depends on supplying high-performance processors, appears particularly vulnerable as DeepSeek's cost-effective approach threatens to reduce demand for premium chips. The growing divide between the US and China in AI, however, is more than just competition -- it's a clash of governance models. While US firms remain fixated on protecting market dominance, China is accelerating AI innovation with a model that is proving more adaptable to global competition. If Silicon Valley resists structural change, it risks falling further behind. We may witness the unraveling of the "Silicon Valley effect," through which tech giants have long manipulated AI regulations to entrench their dominance. For years, Google, Meta,and OpenAI shaped policies that favored proprietary models and costly infrastructure, ensuring that AI development remained under their control. DeepSeek is redefining AI with breakthroughs in code intelligence, vision-language models and efficient architectures that challenge Silicon Valley's dominance. By optimizing computation and embracing open-source collaboration, DeepSeek shows the potential of China to deliver cutting-edge models at a fraction of the cost, outperforming proprietary alternatives in programming, reasoning and real-world applications. More than a policy-driven rise, China's AI surge reflects a fundamentally different innovation model -- fast, collaborative and market-driven -- while Silicon Valley holds on to expensive infrastructure and rigid proprietary control. If US firms refuse to adapt, they risk losing the future of AI to a more agile and cost-efficient competitor. A new era of geotechnopolitics But China is not just disrupting Silicon Valley. It is expanding "geotechnopolitics," where AI is a battleground for global power. With AI projected to add US$15.7 trillion to the global economy by 2030, China and the US are racing to control the technology that will define economic, military and political dominance. DeepSeek's advancement has raised national security concerns in the US. Trump's government is considering stricter export controls on AI-related technologies to prevent them from bolstering China's military and intelligence capabilities. As AI-driven defense systems, intelligence operations and cyber warfare redefine national security, governments must confront a new reality: AI leadership is not just about technological superiority, but about who controls the intelligence that will shape the next era of global power. China's AI ambitions extend beyond technology, driving a broader strategy for economic and geopolitical dominance. But with over 50 state-backed companies developing large-scale AI models, its rapid expansion faces growing challenges, including soaring energy demands and US semiconductor restrictions. China's president, Xi Jinping, remains resolute, stating, "Whoever can grasp the opportunities of new economic development such as big data and artificial intelligence will have the pulse of our times." He sees AI driving "new quality productivity" and modernizing China's manufacturing base, calling its "head goose effect" a catalyst for broader innovation. To counter western containment, China has embraced a "guerrilla" economic strategy, bypassing restrictions through alternative trade networks, deepening ties with the global south, and exploiting weaknesses in global supply chains. Instead of direct confrontation, this decentralized approach uses economic coercion to weaken adversaries while securing China's own industrial base. China is also leveraging open-source AI as an ideological tool, presenting its model as more collaborative and accessible than western alternatives. This narrative strengthens its global influence, aligning with nations seeking alternatives to western digital control. While strict state oversight remains, China's embrace of open-source AI reinforces its claim to a future where innovation is driven not by corporate interests but through shared collaboration and global cooperation. But while DeepSeek claims to be open access, its secrecy tells a different story. Key details on training data and fine-tuning remain hidden, and its compliance with China's AI laws has sparked global scrutiny. Italy has banned the platform over data-transfer risks, while Belgium and Ireland launched privacy probes. Under Chinese regulations, DeepSeek's outputs must align with state-approved narratives, clashing with the EU's AI Act, which demands transparency and protects political speech. Such "controlled openness" raises many red flags, casting doubt on China's place in markets that value data security and free expression. Many western commentators are seizing on reports of Chinese AI censorship to frame other models as freer and more politically open. The revelation that a leading Chinese chatbot actively modifies or censors responses in real time has fueled a broader narrative that western AI operates without such restrictions, reinforcing the idea that democratic systems produce more transparent and unbiased technology. This framing serves to bolster the argument that free societies will ultimately lead the global AI race. But at its heart, the "AI arms race" is driven by technological dominance. The US, China, and the EU are charting different paths, weighing security risks against the need for global collaboration. How this competition is framed will shape policy: lock AI behind restrictions, or push for open innovation. DeepSeek, for all its transformational qualities, continues to exemplify a model of AI where innovation prioritizes scale, speed and efficiency over societal impact. This drive to optimize computation and expand capabilities overshadows the need to design AI as a truly public good. In doing so, it eclipses this technology's genuine potential to transform governance, public services and social institutions in ways that prioritize collective well-being, equity and sustainability over corporate and state control. A truly global AI framework requires more than political or technological openness. It demands structured cooperation that prioritizes shared governance, equitable access, and responsible development. Following a workshop in Shanghai hosted by the Chinese government last September, the UN's general secretary, António Guterres, outlined his vision for AI beyond corporate or state control: "We must seize this historic opportunity to lay the foundations for inclusive governance of AI -- for the benefit of all humanity. As we build AI capacity, we must also develop shared knowledge and digital public goods." Both the west and China frame their AI ambitions through competing notions of "openness" -- each aligning with their strategic interests and reinforcing existing power structures. Western tech giants claim AI drives democratization, yet they often dominate digital infrastructure in parts of Africa, Asia and Latin America, exporting models based on "corporate imperialism" that extract value while disregarding local needs. China, by contrast, positions itself as a technological partner for the rest of the global south; however, its AI remains tightly controlled, reinforcing state ideology. China's proclaimed view on international AI collaboration emphasizes that AI should not be "a game of rich countries"", as President Xi stated during the 2024 G20 summit. By advocating for inclusive global AI development, China positions itself as a leader in shaping international AI governance, especially via initiatives like the UN AI resolution and its AI capacity-building action plan. These efforts help promote a more balanced technological landscape while allowing China to strengthen its influence on global AI standards and frameworks. However, beneath all these narratives, both China and the US share a strategy of AI expansion that relies on exploited human labor, from data annotation to moderation, exposing a system driven less by innovation than by economic and political control. Seeing AI as a connected race for influence highlights the need for ethical deployment, cross-border cooperation, and a balance between security and progress. And this is where China may face its greatest challenge -- balancing the power of open-source innovation with the constraints of a tightly controlled, authoritarian system that thrives on restriction, rather than openness.

Open source and under control: The DeepSeek paradox

Chinese company DeepSeek sits at the intersection of two key AI debates: whether source code should be freely accessible and whether development should occur in open or controlled environments. It promotes open-source AI, with models available for modification, while operating in China, one of the world's most tightly controlled data environments.DeepSeek has emerged on the front line of debates determining the future of AI, but its arrival poses questions over who decides what 'intelligence' we need. Chinese company DeepSeek stands at the crossroads of two major battles shaping artificial intelligence development: whether source code should be freely available and whether development should happen in free or controlled-information environments. That also highlights the DeepSeek paradox. It champions open-source AI - where the source code of the underlying model is available for others to use or modify - while operating in China, one of the world's most-controlled data environments. That means DeepSeek prompts obvious questions about who decides what kind of 'intelligence' we need. Such questions are obviously front of mind for some governments, with several already placing restrictions on the use of DeepSeek. DeepSeek, a Chinese startup, unveiled its AI chatbot late last month. It seemed to equal the performance of US models at a fraction of the cost and the news triggered a massive sell-off of tech company shares on the US share market. It also sparked concerns about data security and censorship. In Australia, DeepSeek has been banned from all federal government devices, the NSW government has reportedly banned it from its devices and systems and other state governments are considering their options. The Australian ban followed similar action by Taiwan, Italy and some US government agencies. The Australian government says the bans are not related to DeepSeek's country of origin, but the issues being raised now are similar to those discussed when Chinese-based social media app TikTok was banned on Australian government devices two years ago. Yet aside from those concerns and DeepSeek's role in reshaping the power dynamics in the US-China AI rivalry, it also gives hope to less well-resourced countries to develop their own large language models using DeepSeek's model as a starting point. For those seeking Chinese-related pop culture references, DeepSeek is a Monkey King moment in the global AI landscape. Monkey King, or Wukong in Chinese, was a character featured in the 16th century novel Journey to the West. The story was popularised in the 1980s television series Monkey and later iterations. In these stories, Wukong was the unpredictable force challenging established power, wreaking havoc in the Heavenly Palace and embodying both defiance and restraint. That's a pretty apt description for where DeepSeek stands in the AI world in 2025. A new benchmark As the author of a recent Forbes piece rightly points out, the real story about DeepSeek is not about geopolitics but "about the growing power of open-source AI and how it's upending the traditional dominance of closed-source models". Author Kolawole Samuel Adebayo says it's a line of thought that Meta chief AI scientist Yann LeCun also shares. The AI industry has long been divided between closed-source titans like OpenAI, Google, Amazon, Microsoft and Baidu and the open-source movement, which includes Meta, Stability AI, Mosaic ML as well as universities and research institutes. DeepSeek's adoption of open-source methodologies - building on Meta's open-source Llama models and the PyTorch ecosystem - places it firmly in the open-source camp. While closed-source large language models prioritise controlled innovation, open-source large language models are built on the principles of collaborative innovation, sharing and transparency. DeepSeek's innovative methods challenge the notion that AI development is backed by vast proprietary datasets and computational power, measured by the number and capacity of chips. It also demonstrates a point made by the Australian Institute for Machine Learning's Deval Shah three months before DeepSeek made global headlines: "The future of LLM [large language model] scaling may lie not just in larger models or more training data, but in more sophisticated approaches to training and inference." The DeepSeek case illustrates that algorithmic ingenuity can compensate for hardware and computing limitations, which is significant in the context of US export controls on high-end AI chips to China. That's a crucial lesson for any nation or company restricted by computational bottlenecks. It suggests that an alternative path exists - one where innovation is driven by smarter algorithms rather than sheer hardware dominance. Just as Wukong defied the gods with his wit and agility, DeepSeek has shown that brute strength, or in this case raw computing power, is not the only determinant of AI success. However, DeepSeek's victory in the open-source battle does not mean it has won the war. It faces the toughest challenges for the road ahead, particularly when it comes to scale, refinement and two of the greatest strengths of US AI companies - data quality and reliability. The Achilles' heel DeepSeek appears to have broken free from the limitations of computing dependence, but it remains bound by China's controlled information environment, which is an even greater constraint. Unlike ChatGPT or Llama, which train on vast, diverse and uncensored global datasets, DeepSeek operates in the palm of the Buddha - the walled garden that is the Chinese government-approved information ecosystem. While China's AI models are technically impressive and perform brilliantly on technical or general questions, they are fundamentally limited by the data they can access, the responses they can generate and the narratives they are allowed to shape. This is particularly so when it comes to freedom of expression and was illustrated by a small test conducted on 29 January 2025. DeepSeek was asked questions about the 1989 Tiananmen Square protests and massacre. In the test, DeepSeek was asked three questions, two in Chinese and one in English. It refused to answer the first and third question and evaded the second question. ChatGPT, on the other hand, gives a thorough analysis to all three questions. The test - among many other queries on sensitive topics - exposes the double bind facing Chinese AI: Can its large language model be truly world-class if it is constrained in what data it can ingest and what output it can generate? Can it be trustworthy if it fails the reliability test? This is not merely a technical issue - it's a political and philosophical dilemma. In contrast to models like GPT-4, which can engage in free-form debate, DeepSeek operates within an internet space where sensitive topics must be avoided. DeepSeek may have championed open-source large language models with its Chinese discourse of efficiency and ingenuity, but it remains imprisoned by a deeper limitation: data and regulatory constraints. While its technical prowess lies in its reliance on and contribution to openness in code, it operates within an information 'greenhouse', where production of and access to critical and diverse datasets are 'protected'. In other words such datasets are restricted. This is where the Monkey King metaphor comes full circle. Just as Wukong believed he had escaped but only to realize he was still inside the Buddha's palm, DeepSeek appears to have achieved independence - yet remains firmly within the grip of the Chinese Communist Party. It embodies the most radical spirit of AI transparency, yet it is fundamentally constrained in what it can see and say. No matter how powerful it becomes, it is hard to evolve beyond the ideological limits imposed upon it. The true disruption in generative AI is not technical; it is philosophical. As we move toward generative AI agency and superintelligent AI, the debate might no longer be about finding our own place in the workforce or cognitive hierarchy, or whether large language models should be open or closed. Instead, we could be asking: What kind of 'intelligence' do we need and - more importantly - who gets to decide?

What is DeepSeek, the AI side project that's upsetting the status quo?

Gemini Advanced: Everything you need to know about Google's premium AI The big AI news of the year was set to be OpenAI's Stargate Project, announced on January 21. The project plans to invest $500 billion in AI infrastructure to "secure American leadership in AI." One day before, a little-known Chinese AI company released its seventh major large language model to little acclaim. However, in the weeks since, the LLM changed the AI landscape (currently dominated by ChatGPT) and forced big players like OpenAI to reevaluate their business strategies. Related Gemini Advanced: Everything you need to know about Google's premium AI Google's premium AI explained Posts 2 What is DeepSeek? Let's start with the basics Source: tv.CCTV.com DeepSeek is a Chinese AI company founded by Liang Wenfang, co-founder of a successful quantitative hedge fund company that uses AI to inform its investment decisions. In 2023, Liang started DeepSeek as a side project to pursue artificial general intelligence, but this is more than the lark of an eccentric millionaire. Liang began building his own data center in 2015 with 100 graphics cards. He opened Fire-Flyer 1 in 2019 with 1,100 cards and a $30 million investment. After dropping $140 million, he launched Fire-Flyer 2 in 2021 using 10,000 Nvidia A100 graphics cards (40GB-80GB, ≥1TB/s, ~$10,000/unit). Then, he decided to get serious about this AI business and created DeepSeek. What did DeepSeek do before 2025? Liang didn't waste time. Less than six months after DeepSeek was a thing, it released DeepSeek-Coder and DeepSeek-LLM in November 2023. DeepSeek-MoE was released in January 2024, using a "mixture-of-experts" architecture that makes its current model popular and powerful. In May last year, the world saw how disruptive this quiet Chinese company could be. DeepSeek released its V2 model with tokens priced so low that it triggered a price war within China, forcing companies like Alibaba, ByteDance, and Tencent to slash prices to keep pace. On the day after Christmas 2024, seven months after the release of V2, DeepSeek released V3. That's where DeepSeek's story in the current news cycle kicks off. Related What is ChatGPT? Learn what ChatGPT is, how it works, what you can do with it, and how much it costs to use OpenAI's most advanced AI chatbot Posts What are DeepSeek-V3 and DeepSeek-R1? And why are they shaking up the industry so much? DeepSeek-V3 is a big, general-purpose language model that performs slightly better than GPT-4o and other leading LLMs on most benchmarks. V3 being slightly better than 4o (which has some neat tricks V3 doesn't) doesn't seem like big news, considering the AI industry has been in an arms race since OpenAI's GPT-3 showed up in 2020. It's the conditions under which V3 outperforms 4o that are noteworthy. First, even though V3 was trained with 671 billion parameters, DeepSeek claims the cost of training this model was around $6 million (based on 2.788 million training hours on H800 GPUs at $2 per GPU hour, which is about the market rate). OpenAI CEO Sam Altman once quipped that GPT-4 (GPT-4o's predecessor) cost over $100 million. The difference in training costs for two similarly powered models is so stark that it has shaken the market. Source: DeepSeek/GitHub Further, V3 uses a mixture-of-experts architecture, meaning it doesn't activate all of its 671 billion parameters for each query. It only uses about 37 billion of them. That equates to faster responses and a lower computational cost per query, allowing DeepSeek to charge less for its tokens. OpenAI charges $2.50 per million input tokens and $10 per million output tokens on its GPT-4o model. DeepSeek charges $0.14 per million input tokens and $0.28 per million output tokens. The price difference is staggering. Price per million tokens Input Output DeepSeek-V3 $0.14 $0.28 GPT-4o $2.50 $10.00 Why is DeepSeek-R1 a big deal? DeepSeek's R1 model is built on the back of the previous V3 model and specializes in reasoning, a sort of internal monologue for LLMs (also known as chain of thought). What makes R1 interesting is that it was initially made exclusively using reinforcement learning with no supervised learning. Supervised learning is a machine learning technique that teaches the model using labeled pairs of inputs and outputs. Reinforcement learning eschews labels and rewards the model as it gets closer to the desired outputs. Source: DeepSeek/GitHub The initial model made from this reinforcement-only technique is R1-Zero, and it developed emergent reasoning capabilities. It didn't need to be trained to have an internal monologue. It just happened. The problem with R1-Zero was that it wasn't very comprehensible despite its high benchmark scores. DeepSeek resolved this issue by fine-tuning the model with limited supervised learning, leading to R1, which could match OpenAI's o1 reasoning model on numerous benchmarks while undercutting OpenAI's prices. Price per million tokens Input Output DeepSeek-R1 $0.14-$0.55 $2.19 OpenAI-o1 $7.50-$15.00 $60.00 DeepSeek is not the end of OpenAI (or Llama, or Gemini, or Anthropic) The AI arms race is only just beginning The market consequences of DeepSeek releasing R1 weren't felt for a few days, but when it reacted, the market reacted big. The Nasdaq index saw a loss of $1 trillion in market capitalization. Nvidia had it worse than anyone, losing nearly $600 billion. How could DeepSeek trigger this kind of reaction? Related 7 best ChatGPT alternatives Try these AI tools when ChatGPT is down Posts 1 DeepSeek matched the best that OpenAI had released, using cheaper hardware (like these awesome budget phones) for less training time, calling into question the necessity for big data centers and expensive GPUs. After all, why invest $100 million in training a new model with expensive hardware when $6 million and cheaper silicon is just as good? DeepSeek's models won't replace the American AI giants nor obviate the need for advanced processors. The markets have nearly recovered from last month's crash, indicating that the AI industry is quickly returning to business as usual. But don't think that there hasn't been a sea change. DeepSeek's new models have proven that change can come from smaller players exploring new techniques. Because DeepSeek open sourced how it made its models, everyone is free to replicate its methods to make their models better. That's already happening. A team from Berkeley applied DeepSeek's reinforcement learning algorithm to train the Qwen 3B model to solve simple math puzzles. For $30 of processing time, the same chain-of-thought reasoning present in the R1-Zero model emerged in their specialized model, indicating that the emergent reasoning reported by DeepSeek isn't just an empty boast. In other words, look for more big players to use these training techniques pioneered by DeepSeek and released to the public. The AI arms race is heating up Rather than steal the thunder of major AI firms, DeepSeek injected new wind into the industry's sails. More than demonstrating that innovation can come from smaller players and older hardware, it's shown how much more its better-funded competitors (like Gemini and ChatGPT) are capable of given these new tricks. Last year was a hot year for AI, but don't expect 2025 to cool down.

India Should Develop Full Stack of AI

A week after the launch of China's DeepSeek AI model wiped nearly a trillion US dollars in market cap off American tech stocks, triggering an AI arms race between the US and China, and drawing inevitable parallels with "the Sputnik moment", Sam Altman, the CEO of OpenAI and the poster boy of the US AI industry, appeared altogether unimpressed with the frenzy. A week after the launch of China's DeepSeek AI model wiped nearly a trillion US dollars in market cap off American tech stocks, triggering an AI arms race between the US and China, and drawing inevitable parallels with "the Sputnik moment", Sam Altman, the CEO of OpenAI and the poster boy of the US AI industry, appeared altogether unimpressed with the frenzy. During an interview in New Delhi, sporting a powder blue sweater, dark trousers and Adidas sneakers -- the established dress code of the Silicon Valley tech workforce -- Altman appeared focused and unperturbed, as he took questions from Samidha Sharma & Sruthijith KK, on a range of themes. He addressed the critical question of whether the previously assumed infra costs for generative AI development were inflated, the scepticism around whether he can find $500 billion for his ambitious Stargate project, and whether he has changed his mind about his infamous comment that it would be "hopeless" for Indian startups to compete with OpenAI. He now says countries like India should focus on the full stack of AI, including building expensive foundational models. Edited excerpts: Has DeepSeek and the market's reaction to it reset expectations around the need for computing power (compute) for large language models (LLMs) and foundational models? I don't think so. There are two different trends at play. One is as we continue to push the boundaries of frontier models, the same exponential curve that we've been seeing for a long time continues. In fact, if anything, I think we know how to do better with more compute now than we did a year ago. And the economic returns on increasing intelligence will be exponential. We also see this amazing thing that approximately every 12 months we can reduce the cost of a given unit of intelligence by 10x... Like 10x per year compounding is incredible. So, there will be lots of abundant cheap intelligence. But you'll also need huge amounts of compute to make these big systems. And as intelligence gets cheaper, people want to use much, much more and you need a bigger inference fleet, too. You've announced a couple of Asia alliances on your current tour like the one with SoftBank in Japan and with Kakao in South Korea. Anything in India? India is very important to us and we hope to have more to share soon. It's a market we want to do way more in, and are working on some large partnerships here. It'll be a core part of our future. We love what the government is doing with the AI policy and we want to invest here. When you were here last in June 2023 you said at an ET event it would be hopeless for Indian startups to compete with you on building foundational models. Have you changed your mind since then? Yeah, that was taken out of context, also it was a very different time almost two years ago. The comment was made for scaling the pre-training paradigm. One of the best developments since then is that the world's gotten much better at small models. These reasoning models you can build -- maybe not for as low as it was reported for DeepSeek -- but for much less money. This is a great time for the next paradigm of AI to diffuse around the world, and I'm sure India and other countries will train incredible models. What are the possibilities that can be explored by India while building LLMs? And how would you contribute? We're happy to help. It's a good time to go after it. I'm happy with the government's plans and we want to be a supporter. One of the approaches taken by the Indian government is to secure GPUs and then offer it to companies and institutions at a low price. Is that a good approach? (Thinks for a moment.) I think so. Can you give us details of the Stargate Project? Your announcement with President Trump spoke about a $500 billion commitment to OpenAI for building what can possibly be the largest AI infrastructure play, but is all of this money in? It's remarkable to see the size and scale of this. On the money, you'll see it as it goes into the ground. I totally get why people would make some of those comments. Obviously, they have their own incentives and competition here. But this is the biggest infrastructure project in US history and we're very excited to get to do it. There is a big gap between $500 billion which you'd announced and what's come through... Obviously, the $500 billion is not all equity -- it includes debt and other sorts of project finance. And some of that debt will come online later into the project. But we wouldn't have announced it if we weren't confident. We're not going to get into the details of explaining the breakout but I'm very confident about the scale of the project. How important is the Stargate infrastructure to the future of OpenAI? The infrastructure is critical to keep scaling and improving these models on this curve. We need this level of infrastructure investment. How has the arrival of DeepSeek and Alibaba's Qwen changed, if at all, the thinking around the product pipeline, investments, and pricing of OpenAI products? We will accelerate and respond where we can but mostly, we have a very clear vision. We're working on executing our product pipeline for this year and next year. And we'll figure out 2027 later. We're just going to keep executing on that. Where do you stand on the Jevons paradox? I agree people will just use way more AI. There's an old Bill Gates (Microsoft cofounder) quote about how one couldn't imagine a computer ever needing more than 64K of memory. I understand what that's like because I understand now why people say you never need more intelligence than GPT-5. I am confident that someday that will seem as silly as the 64K comment. We'll just use, as we drive the cost of this down, the value of it up, we'll just use so, so much more. So that's part of why we want to do Stargate. And therefore, the investments needed will not be lower. They'll just be more AI per dollar. On a Reddit AMA (Ask me Anything) recently, you said you were on the wrong side of history for not picking an open-source strategy. There is a place for open source but we're not going to totally change our plans. I don't think it's right to open source everything but there's an important place for it. What are the most promising advances you've seen in energy in terms of securing clean renewable energy for AI? (Nuclear) Fusion and also next-generation fission. Those are the two most promising. But is fusion anywhere close to reality? I think so. I don't want to give a date. You've also invested in fusion companies. Yeah, that's why I'm biased. You know, I've lost track of the boards I'm on. I try to get off boards as much as I can. But the two boards I really spend time on are Oklo, a fission company, and Helion, a fusion company. I think energy is so important. And I think nuclear is such an incredible path to drive energy costs way down and abundance way up. The two mega trends that I care about are the cost of intelligence trending towards zero and the cost of energy trending towards zero. I feel like that's my personal mission. Given the AI arms race -- the emerging one between the US and China, which is now very apparent -- is there still hope for global framework for regulation of AI? The IAEA (International Atomic Energy Agency) is an example I'd point to as a historical precedent in much more challenging scenarios. So, yeah, if they can do it, then we can do it now. Should Indian startups focus on developing more foundational models or we are better off focusing on applications and vertical AI? I think it's worth focusing on the full stack. And the kind of money that would be needed, do you think that's changed? Well, again, to be at the frontier, the price is $500 billion...in this world of reasoning models, because we've had this other trend of 10x per-year cost reduction, you can do amazing models that are still very capable for much, much less money. And what do you think is an example right now of that? DeepSeek is the one you just mentioned. Again, I don't think it was $5 million, but it was something reasonable. But do you think what they claim is the cost is actually the cost? It's an order of magnitude more than that. But this is all speculation online. I don't have any inside data there. Elon Musk is continuing his attacks on you... Very viciously so... Are you now concerned about the influence on policy he (Musk) has under Trump? Well, we were thrilled to announce Stargate with President Trump on his first full day in office, and I think it's great that he's such a supporter of the project. Was it the right decision for OpenAI to switch from the not-for-profit to the for-profit model? We haven't switched yet. We're looking at switching... Even after that we'd still have our non-profit tenets... He's (Musk) a competitor who wants to win, and you have to look at all of his statements (in that light)... Talk a bit about your relationship with Microsoft, keeping in mind that you want to further diversify your shareholding... Microsoft CEO Satya Nadella said that he only knows about the $80 billion that he's committed to Stargate. He doesn't know about any of the other investors and can only speak for his $80 billion, but not the full $500 billion. We have a special relationship with Microsoft, and we will continue to have a special relationship for a long time to come. They are an incredible partner. Satya (Nadella) is a personal friend and a great ally. We obviously are expanding towards a more diversified investor base, but we're growing up as a company... How many months away is AGI (artificial general intelligence)? I have no idea. Make a prediction for us... I try not to predict timelines, because we're always late. Clearly, the models are getting super capable. People have wildly different definitions of AGI. Even if you told me exactly what the definition was, I would still say, I don't know how long the research takes. But we launched Deep Research on Monday, and a lot of people were like, this is an AGI... I certainly don't think so. But clearly, for someone's definition, it was... And Deep Research has wowed people the most since the launch of the original ChatGPT. I cannot believe this move from chatbots to agents, watching people get to see what it's like to have a model that goes off and thinks and does days of work for you. It's pretty cool!

Decoding OpenAI's Super Bowl ad and Sam Altman's grandiose blog post

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More If you were in one of the nearly 40 million U.S. households that tuned into the NFL Super Bowl LIX this year, in addition to watching the Philadelphia Eagles trounce the Kansas City Chiefs, you may have caught an advertisement for OpenAI. This is the company's first Super Bowl ad, and it cost a reported $14 million -- in keeping with the astronomical sums commanded by ads during the big game, which some come to see instead of the football. As you'll see in a copy embedded below, the OpenAI ad depicts various advancements throughout human history, leading up to ChatGPT today, what OpenAI calls the "Intelligence Age." While reaction to the ad was mixed -- I've seen more praise and defense for it than criticism in my feeds -- it clearly indicates that OpenAI has arrived as a major force in American culture, and quite obviously seeks to connect to a long lineage of invention, discovery and technological progress that's taken place here. On it's own, the OpenAI Super Bowl ad seems to me to be a totally inoffensive and simple message designed to appeal to the widest possible audience -- perfect for the Super Bowl and its large audience across demographics. In a way, it's even so smooth and uncontroversial that it is forgettable. But coupled with a blog post OpenAI CEO Sam Altman published on his personal website earlier on Sunday, entitled "Three Observations," and suddenly OpenAI's assessment of the current moment and the future becomes much more dramatic and stark. Altman begins the blog post with a pronouncement about artificial general intelligence (AGI), the raison d'etre of OpenAI's founding and its ongoing efforts to release more and more powerful AI models such as the latest o3 series. This pronouncement, like OpenAI's Super Bowl ad, also seeks to connect OpenAI's work building these models and approaching this goal of AGI with the history of human innovation more broadly. "Systems that start to point to AGI* are coming into view, and so we think it's important to understand the moment we are in. AGI is a weakly defined term, but generally speaking we mean it to be a system that can tackle increasingly complex problems, at human level, in many fields. People are tool-builders with an inherent drive to understand and create, which leads to the world getting better for all of us. Each new generation builds upon the discoveries of the generations before to create even more capable tools -- electricity, the transistor, the computer, the internet, and soon AGI." A few paragraphs later, he even seems to concede that AI -- as many developers and users of the tech agree -- is simply another new tool. Yet he immediately flips to suggest this may be a much different tool than anyone in the world has ever experienced to date. As he writes: "In some sense, AGI is just another tool in this ever-taller scaffolding of human progress we are building together. In another sense, it is the beginning of something for which it's hard not to say "this time it's different"; the economic growth in front of us looks astonishing, and we can now imagine a world where we cure all diseases, have much more time to enjoy with our families, and can fully realize our creative potential." The idea of "curing all diseases," while certainly appealing -- mirrors something rival tech boss Mark Zuckerberg of Meta also sought out to do with his Chan-Zuckerberg Initiative medical research nonprofit co-founded with his wife, Prisicilla Chan. As of two years ago, the timeline proposed for the Chan-Zuckerberg's initiative to reach this goal was by 2100. Yet now thanks to the progress of AI, Altman seems to believe it's attainable even sooner, writing: "In a decade, perhaps everyone on earth will be capable of accomplishing more than the most impactful person can today." Altman and Zuck are hardly the one two high-profile tech billionaires interested in medicine and longevity science in particular. Google's co-founders, especially Sergey Brin, have put money towards analogous efforts, and in fact, there were (or are) at one point so many leaders in the tech industry interested in prolonging human life and ending disease that back in 2017, The New Yorker magazine ran a feature article entitled: "Silicon Valley's Quest to Live Forever." This utopian notion of ending disease and ultimately death seems patently hubristic to me on the face of it -- how many folklore stories and fairy tales are there about the perils of trying to cheat death? -- but it aligns neatly with the larger techno-utopian beliefs of some in the industry, which have been helpfully grouped by AGI critics and researchers Timnit Gebru and Émile P. Torres under the umbrella term TESCREAL, an acronym for "transhumanism, Extropianism, singularitarianism, (modern) cosmism, Rationalism, Effective Altruism, and longtermism," in their 2023 paper. As these authors elucidate, the veneer of progress sometimes masks uglier beliefs such as in the inherent racial superiority or humanity of those with higher IQs, specific demographics, and ultimately evoking racial science and phrenology of more openly discriminatory and oppressive ages past. There's nothing to suggest in Altman's note that he shares such beliefs, mind you...in fact, rather the opposite. He writes: "Ensuring that the benefits of AGI are broadly distributed is critical. The historical impact of technological progress suggests that most of the metrics we care about (health outcomes, economic prosperity, etc.) get better on average and over the long-term, but increasing equality does not seem technologically determined and getting this right may require new ideas." In other words: he wants to ensure everyone's life gets better with AGI, but is uncertain how to achieve that. It's a laudable notion, and one that maybe AGI itself could help answer, but for one thing, OpenAI's latest and greatest models remain closed and proprietary as opposed to competitors such as Llama's Meta family and DeepSeek's R1, though the latter has apparently caused Altman to re-assess OpenAI's approach to the open source community as he mentioned on a recent separate Reddit AMA thread. Perhaps OpenAI could start by open sourcing more of its technology to ensure it spreads wider to more users, more equally? Meanwhile, speaking of specific timelines, Altman seems to project that while the next few years may not be wholly remade by AI or AGI, he's more confident of a visible impact by the end of the decade 2035. As he puts it: "The world will not change all at once; it never does. Life will go on mostly the same in the short run, and people in 2025 will mostly spend their time in the same way they did in 2024. We will still fall in love, create families, get in fights online, hike in nature, etc. But the future will be coming at us in a way that is impossible to ignore, and the long-term changes to our society and economy will be huge. We will find new things to do, new ways to be useful to each other, and new ways to compete, but they may not look very much like the jobs of today. Anyone in 2035 should be able to marshall [sic] the intellectual capacity equivalent to everyone in 2025; everyone should have access to unlimited genius to direct however they can imagine. There is a great deal of talent right now without the resources to fully express itself, and if we change that, the resulting creative output of the world will lead to tremendous benefits for us all." Where does this leave us? Critics of OpenAI would say it's more empty hype designed to continue placating OpenAI's big-pocketed investors such as Softbank and put off any pressure to have working AGI for a while longer. But having used these tools myself, watched and reported on other users and sene what they've been able to accomplish -- such as writing up complex software within mere minutes without much background in the field -- I'm inclined to believe Altman is serious in his prognostications, and hopeful in his commitment to equal distribution. But keeping all the best models closed up under a subscription bundle clearly is not the way to attain equal access to AGI -- so my biggest question remains on what the company does under his leadership to ensure it moves in this direction he so clearly articulated and that the Super Bowl ad also celebrated.

ET Interview: Look to invest, explore large tieups in India

Sam Altman, cofounder and CEO, OpenAI, said the ChatGPT maker will accelerate and respond if needed with new products and pricing amid rising competition from upstarts like China's DeepSeek, which has shaken up the AI world. Altman told ET in an exclusive interview in New Delhi that smaller reasoning models such as that of DeepSeek can be built at much cheaper cost, giving countries like India a chance to create foundational platforms. The OpenAI chief, who's on a trip through Asia, said he's open to partnering with local startups to facilitate the development of indigenous large language models (LLMs) and is looking to make investments in India. Altman spoke about the ambitious Stargate project, why he needs massive amounts of capital to build the world's largest AI infrastructure and his relationship with Microsoft. You've announced a couple of Asia alliances on your current tour like the one with SoftBank in Japan and with Kakao in South Korea. Anything in India specifically? India is important to us and we hope to have more to share soon. It's amarket we want to do way more in, and are working on some large partnerships here. It'll be a core part of our future. We love what the government is doing with the AI policy and we want to invest here. When you were here last in June 2023 you said at an ET event it would be hopeless for Indian startups to compete with you on building foundational models. Have you changed your mind since then? Yeah, that was taken out of context, also it was a very different time almost two years ago. The comment was made for scaling the pre-training paradigm. One of the best developments since then is that the world's gotten much better at small models. These reasoning models you can build -- maybe not for as low as it was reported for DeepSeek -- but for much less money. This is a great time for the next paradigm of AI to diffuse around the world, and I'm sure India and other countries will train incredible models. Also Read: Sam Altman: India second largest market for OpenAI; users have tripled in the last one year Talk a bit about indigenous LLMs that the Indian government is bullish about, especially after the rise of DeepSeek. What are the possibilities that can be explored by India while building LLMs? And how would you contribute? We're happy to help. It's a good time to go after it. I'm happy with the government's plans and we want to be a supporter. Any specifics? We'll have meetings later today (Wednesday). One of the approaches taken by the Indian government is to secure graphics processing units (GPUs) and then offer it to companies and institutions. Is that a good approach? I think so. DeepSeek has stirred up the language model space, having built at a fraction of the resources used by companies like OpenAI. Has this reset the narrative around the need for massive computing power for foundational models? I don't think so. There's two different trends at play. One is to continue to push the boundaries of frontier model intelligence. The same exponential curve that we've been seeing for a long time continues. In fact, if anything, I think we know how to do better with more compute now than we did a year ago. And the economic returns on increasing intelligence will be exponential. Why do you say that? We see this amazing thing that approximately every 12 months we can reduce the cost of a given unit of intelligence by 10x... Like 10x per year compounding is incredible. So, there will be lots of abundant cheap intelligence. But you'll also need huge amounts of compute to make these big systems. And as intelligence gets cheaper, people want to use much, much more and you need a bigger inference fleet, too. Can you give us details of the Stargate Project? Your announcement with President Trump spoke about a $500 billion commitment to OpenAI for building what can possibly be the largest AI infrastructure play, but is all of this money in? It's remarkable to see the size and scale of this. On the money, you'll see it as it goes into the ground. I totally get why people would make some of those comments. Obviously, they have their own incentives and competition here. But this is the biggest infrastructure project in US history and we're very excited to get to do it. There is a big gap between $500 billion which you'd announced and what's come through... Obviously, the $500 billion is not all equity -- it includes debt and other sorts of project finance. And some of that debt will come online later into the project. But we wouldn't have announced it if we weren't confident. We're not going to get into details of explaining the breakout but I'm very confident in the scale of the project. How important is the Stargate infrastructure to the future of Open AI? The infrastructure is critical to keep scaling and improving these models on this curve. We need this level of infrastructure investment. How has the arrival of DeepSeek and Alibaba's Qwen changed, if at all, the product pipeline, investments, and pricing of OpenAI products? We will accelerate and respond where we can but mostly, we have a very clear vision. We're working on executing our product pipeline for this year and next year. And we'll figure out 2027 later. We're just going to keep executing on that. And where do you stand on the Jevons paradox? I agree people will just use way more AI. There's an old Bill Gates (Microsoft cofounder) quote about how one couldn't imagine a computer ever needing more than 64K of memory. I understand what that's like because I understand now why people say you never need more intelligence than GPT-5. I am confident that someday that will seem as silly as the 64K comment. We'll just use, as we drive the cost of this down, the value of it up, we'll just use so, so much more. So that's part of why we want to do Stargate. And therefore, the investments needed will not be lower. They'll just be more AI per dollar. On a Reddit AMA (Ask me Anything) recently, you said you were on the wrong side of history for not picking an open source strategy. There is a place for open source but we're not going to totally change our plans. I don't think it's right to open source everything but there's an important place for it. And you said that not everybody agrees within the organisation about the open source strategy -- what did you mean? Look, if we had unlimited capacity, I think everybody would agree, but we have to make hard trade-offs about what we do. We're still a relatively small organisation trying to do something very hard. So that's where it comes from. What are the most promising advances you've seen in energy in terms of securing clean renewable power for AI? Fusion and next generation fission -- those are the two most promising. You've a big investor in Helion Energy. Yeah, that's why I'm biased. I've lost track of the boards I am on... I am trying to get off all the boards except the two -- Oklo, a fission company, and Helion... I think energy is so important and nuclear is such an incredible path to drive energy costs way down and abundance way up. The two mega trends that I care about are the cost of intelligence and energy trending towards zero. I feel like that's my personal mission. Given the AI arms race -- the emerging one between the US and China, which is now very apparent -- could there be a global framework for regulation of AI? The IAEA (International Atomic Energy Agency) is an example I'd point to as a historical precedent in much more challenging scenarios. So, yeah, if they can do it, then we can do it now. Should Indian startups focus really on developing more foundational models or we are better off focusing on applications and vertical? I think it's worth focusing on the full stack. And the kind of money that would be needed, do you think that's changed? Well, again, to be at the frontier like we've said, what we think the price is, which is $500 billion. But to be at the... in this world of reasoning models, because we've had this other trend of 10x per-year cost reduction, you can do amazing models that are still very capable for much, much less money. And what do you think is an example right now of that? DeepSeek is the one you just mentioned. Again, I don't think it was $5 million, but it was something reasonable. But do you think what they claim is the cost is actually the cost? It's an order of magnitude more than that. But this is all speculation online. I don't have any inside data there. What do you make of Elon Musk continually attacking you? Very viciously so... Are you now concerned about the influence on policy he (Musk) has under Donald Trump? Well, we were thrilled to announce Stargate with President Trump on his first full day in office, and I think it's great that he's such a supporter of the project. Was it the right decision for OpenAI to switch from the not-for-profit model to the for-profit model? We haven't switched yet. We're looking at switching... Even after that we'd still have our non-profit tenets... He's (Musk) a competitor who wants to win, and you have to look at all of his statements (in that light).. Talk a bit about your relationship with Microsoft, keeping in mind that you want to further diversify your shareholding... Microsoft CEO Satya Nadella said that he only knows about the $80 billion that he's committed to Stargate. He doesn't know about any of the other investors and can only speak for his $80 billion, but not the full $500 billion. We have a special relationship with Microsoft, and we will continue to have a special relationship for a long time to come. They're an incredible partner. Satya (Nadella) is a personal friend and a great ally. We obviously are expanding towards a more diversified investor base, but we're growing up as a company. How many months have we got to get to AGI (artificial general intelligence)? I have no idea. Make a prediction for us... I try not to predict timelines, because we're always late. Clearly, the models are getting super capable. People have wildly different definitions of AGI. Even if you told me exactly what the definition was, I would still say, I don't know how long the research takes. But we launched Deep Research on Monday, and a lot of people were like, this is an AGI... I certainly don't think so. But clearly, for someone's definition, it was... And Deep Research has wowed people the most since the launch of the original ChatGPT. I cannot believe this move from chatbots to agents, watching people get to see what it's like to have a model that goes off and thinks and does days of work for you. It's pretty cool!

How DeepSeek Made OpenAI Take India Seriously

OpenAI CEO Sam Altman believes while models are not cheap, India can still build its own reasoning model and become a leader. Although it's been a while since DeepSeek's release, its impact has been profound, making AI both affordable and widely accessible; so much so that even OpenAI found itself under pressure. Referring to a time when he said it was "hopeless" for India to build its own foundational model, OpenAI CEO Sam Altman clarified his previous statement during his visit to India and said, "That was a very specific time with scaling laws." "But we are now in a world where we have made incredible progress with distillation," he said while talking about the power of small models and reasoning models. As per him, while models are still not cheap, India can still build its own reasoning model and become a leader. Recently, Altman published a blog in which he stated that the cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use. This will, in turn, require more compute. DeepSeek's success has left many wondering how China achieved this with limited resources. At MLDS 2025, Paras Chopra, founder of Lossfunk, shared how DeepSeek pulled it off. He said that one of the major hurdles in scaling large AI models is managing the key-value (KV) cache, which grows quadratically and limits the size of inputs and outputs. The conventional approach involves using inefficient methods like linear attention or group query attention to manage this. DeepSeek, however, found a more efficient solution. "They came out with a low-rank approximation of it, what they called compressed latent KV," Chopra explained. This approach allowed DeepSeek to process longer inputs more efficiently, resulting in improved performance and longer chains of reasoning without the need for excessive computational resources. By addressing the quadratic growth of the KV cache, DeepSeek made it possible to handle larger datasets without the usual computational costs. Besides, DeepSeek's approach to the mixture of expert (MoE) architecture was another key factor. MoE allows different parts of the model to be isolated on different GPUs, saving resources. Chopra said that while others simply routed tasks to the best experts or a fixed number of experts, DeepSeek's innovation was more dynamic. "They thought about intelligence as being comprised of two parts - shared experts and routed experts." To further cut costs, Chopra said DeepSeek innovated at the hardware level as well. He shared that DeepSeek was the first to push the boundaries of Compute Unified Device Architecture (CUDA) and Parallel Thread Execution (PTX), NVIDIA's intermediate language, to address memory bandwidth bottlenecks. "They were the first ones to also do FP8 precision training," he said. Using FP8 precision allowed DeepSeek to run its models on smaller, less expensive GPUs. Training with FP8 precision significantly lowered the memory requirements compared to traditional FP16 or FP32 training, which, in turn, reduced the costs associated with both training and inference. Chopra argues that for India to develop a state-of-the-art foundation model sheer compute power might not be the most effective solution. "The human brain is an incredibly efficient AGI. It runs on potatoes. You don't need a nuclear-powered data centre to operate an AGI," he said. Comparing ISRO's accomplishments in several missions at a lower cost than NASA's, he added that India can do the same in AI. "As a nation, we don't have to look too far to see the amazing things we've already accomplished. We've done it in areas like space, and there's no reason why we can't do the same in AI." Chopra's company Lossfunk is also on a mission to build a state-of-the-art foundational reasoning model from India and is inviting applicants to join the effort. "Creativity is born out of constraints, and DeepSeek's success proves that with the right approach, it's possible to innovate and scale AI models without relying on endless financial resources," Chopra further said. In an interview with AIM, Harneet SN, founder of Rabbitt AI, said, "DeepSeek is the Jambavan moment for India in the sense that, just like in the Ramayana, Jambavan came and reminded Hanuman of his powers, DeepSeek has done the same for India's AI community." The IndiaAI mission recently called for proposals to build India's own foundational model, with finance minister Nirmala Sitharaman allocating ₹2,000 crore for the mission - nearly a fifth of the ₹10,370 crore announced for the scheme last year. Similarly, Gaurav Aggarwal, AI/ML lead at Jio, is inviting exceptional graduate students in the US working on challenging AI problems to join as research interns to build next-generation AI models for India and the world. "India has fallen behind in the race to develop its own cutting-edge LLMs - but we are changing that," he said in a post on X. OpenAI vice president Srinivas Narayanan, during an interaction with IIT Madras professor Balaraman Ravindran, said that the success of DeepSeek is overly exaggerated. "What we have learned from DeepSeek is that they've done some things that are efficient and from which we can learn, but the level of efficiency has been extremely exaggerated," he said, adding that while people talk about the cost of building a single model, it is not the cost of running an entire AI lab. "If you take the cost of a single model that OpenAI would train, maybe our most recent runs would be pretty comparable. But it's much harder to lead -- you have to run...a lot more experiments before you finally decide what model you're going to train," he said. Narayanan added that OpenAI's latest model, o3-mini, is comparably cheaper than the other models in the US on inference. He believes that there will be not much difference between closed-source models and open-source models in terms of pricing in the future. Similarly, Google DeepMind chief Demis Hassabis recently said that DeepSeek can do "extremely good engineering" and that it "changes things on a geopolitical scale". However, from a technology point of view, Hassabis said it was not a big change. "Despite the hype, there's no actual new scientific advance...It's using known techniques [in AI]," he said, adding that the hype around DeepSeek has been "exaggerated a little bit". Meanwhile, Amazon chief Andy Jassy, during the recent earnings call, said that with DeepSeek-like models, inference costs will come "meaningfully down". "I think it will make it much easier for companies to infuse all their applications with inference and with generative AI." He clarified that people thought they would spend less money on infrastructure. However, what happens is that companies will spend a lot less per unit of infrastructure, and that is very useful for their businesses. Notably, AWS was the first cloud to host DeepSeek R1 on AWS Bedrock and Sagemaker.

Sam Altman's AI vision: 5 key takeaways from ChatGPT maker's blog post

One of the most striking things about Sam Altman's new blog post is how confidently he talks about AI's potential to reshape our world, but with an undercurrent of caution. "Our mission is to ensure that AGI (Artificial General Intelligence) benefits all of humanity," he begins, underscoring that the real measure of success isn't just technological achievement but when it benefits society at large. For those who've followed Altman's journey since OpenAI made Generative AI a household term, it's interesting to see Sam Altman evolve from being a scrappy AI pioneer to a tech visionary, tasked with shepherding advanced AI in a way that doesn't unravel society. And in keeping with this new avatar, Altman makes three "observations" in his blog post, each one more provocative than the last, about what's driving AI's explosive economics - and how this will affect you and me. Here are five key takeaways that emerge from his blog, capturing both the ambition and the caution in Altman's words. According to Sam Altman's AI trajectory, "Systems that start to point to AGI are coming into view," and we should appreciate that this is an inflection point. The mission, he says, is fundamentally about building a system that can tackle ever more complex problems, mirroring human-level performance. "AGI is a weakly defined term," Altman acknowledges, but in broad strokes, it means something that handles complexity in an all-purpose manner, not just specialized tasks. Also read: Sam Altman on AGI: OpenAI visionary on the future of AI For us, that might mean an AI agent that isn't limited to writing code or editing text - it can do both, plus manage your schedule, interpret complex images, and provide near-instant, context-aware insights across any domain."In another sense, it is the beginning of something for which it's hard not to say 'this time it's different'," he says, noting that the next economic boom could be nothing short of astonishing. Altman's second big point talks about AI's economics. He says that intelligence in these AI systems scales with "the log of the resources used to train and run it" - which translates to bigger spend = bigger payoff, fairly predictably. "The cost to use a given level of AI falls about 10x every 12 months," he continues, which is a dramatic shift that dwarfs the old Moore's law. And finally, "the socioeconomic value of linearly increasing intelligence is super-exponential," meaning that a modest AI improvement can yield staggering value in the real world. Also read: OpenAI o3-mini vs. DeepSeek R1: Which one to choose? Put those together, and you see how rapid the growth could be if organisations keep pouring money into training and inference compute. "It appears that you can spend arbitrary amounts of money and get continuous and predictable gains," Altman says. In other words, no near-term barrier halts the escalation of more advanced models and their ever-rising adoption. If prices keep dropping, we'll see even more usage - a self-reinforcing cycle. A particularly vivid image from the post is Altman's scenario of AI agents, effectively digital co-workers who can do tasks "up to a couple of days long" with the competence of a mid-level professional. "We are now starting to roll out AI agents, which will eventually feel like virtual co-workers," he writes. "Now imagine 1,000 of them. Or 1 million of them. Now imagine such agents in every field of knowledge work." Also read: From AI agents to humanoid robots: Top AI trends for 2025 Altman's point is that these AI agents might not be the prime inventors of the next big idea, but they'll handle grunt work or code reviews or basic research at scale. "In some ways, AI may turn out to be like the transistor economically," he suggests. Just as transistors became so ubiquitous we don't think about how many billions of them are at the heart of every single chip inside our smartphones, laptops, servers, etc, similarly AI agents could become a universal workforce layer - enhancing productivity across sectors, quietly humming in the background without us taking much notice. Yet for all the talk of AI-led seismic shifts across industries, Altman is keen to remind us that "The world will not change all at once; it never does." He predicts that people in 2025 will spend their time roughly the same way they did in 2024. We'll still "fall in love, create families, get in fights online, hike in nature," as he puts it. But the subtle infiltration of AI - like more tasks automated, more convenience in how we do knowledge work - will continue to creep in. Over time, we'll see new ways of being useful, new job types, new industries, even if they don't resemble the old nine-to-five. Also read: Is AI making us think less? The critical thinking dilemma He concedes it's not all going to be rosy. "We expect the impact of AGI to be uneven," meaning some sectors might feel barely a ripple, while others get reimagined overnight. In the same breath, he warns that the rapid cheapening of intelligence might upend economic norms. Throughout his post, Altman circles back to the question of who benefits from AI's disruptive power. "Ensuring that the benefits of AGI are broadly distributed is critical," he writes. History shows that technology tends to raise living standards over time, but not always equitably. Also read: OpenAI launches ChatGPT DeepResearch - 5 things you need to know He proposes "strange-sounding ideas like giving some 'compute budget' to enable everyone on Earth to use a lot of AI," acknowledging that might be naive but also that it might be necessary. Or maybe the relentless drive to reduce AI costs will suffice. If an AI agent is near-free to run, perhaps everyone can tap in. The alternative, he warns, is an authoritarian track where governments harness AI for mass surveillance. That's the darkest path he alludes to, and it's one he hopes we avoid by giving individuals as much empowerment as possible. Altman's final paragraphs resonate with a hopeful tone: AI will supercharge human willfulness, enabling each person to command far more intellectual power than before. "Anyone in 2035 should be able to marshall the intellectual capacity equivalent to everyone in 2025," he writes. Overall, Sam Altman, in his blog post, comes across as an AI optimist, envisioning a future of mass empowerment, not mass subjugation. Wonder how much of what he envisions comes to pass? Only time will tell.