China to restrict US investment in tech companies as AI rivalry intensifies with new controls

Why China's Affordable AI Is a Worry for Silicon Valley

Chinese artificial intelligence companies can't match the financial firepower of their American rivals, and the US government has deprived them of the most cutting-edge chips to train their AI models. Yet China is encroaching on US leadership in the field. Chinese developers such as DeepSeek and Alibaba Group Holding Ltd. have focused on devising systems that perform almost on a par with the top-performing AI models without needing the most powerful hardware. And China is making a bet on "open-weight" AI software -- where the internal parameters are made available for developers to share, study and tweak -- as a way to foster rapid adoption of AI across the national economy. All this challenges the dominant US business model that's predicated on investing billions and getting users paying top-dollar for the most powerful, proprietary AI technology. Here's what to know about Chinese AI and how much of a threat it poses to OpenAI Inc., Anthropic and other US giants. Efficiency Export controls imposed by Washington restrict China's access to US-designed chips that are around 20% faster and consume as much as 30% less power than their Chinese competitors as they squeeze more transistors onto each sheet of silicon. China's tech community has sought to overcome this handicap by developing AI software that is more "compressed." This means fewer steps of calculation are required to achieve outputs comparable to those of the best-performing US AI models. One way this is achieved is by employing a technique known as mixture of experts. When a user prompts a Chinese chatbot such as DeepSeek or Alibaba's Qwen, the model doesn't need to mobilize its entire "neural network" to generate a response (the process known as inference). Instead, it activates specialized sub-networks known as experts that engage only a fraction of the software's available computational capacity. The models of US labs such as OpenAI and Anthropic often have the edge when it comes to the sheer volume of information they can handle in a single interaction and the sophistication of their responses. But to achieve this, they activate far more of a model's available "neurons" and thereby consume a lot more computing power. With DeepSeek's latest model, V4-Pro, less than 3% of its parameters are actually doing work at any given moment, according to performance data released by the company. Global technology stocks tumbled in late January 2025 after DeepSeek demonstrated a new model, R1, that offered comparable performance to US AI chatbots at seemingly a fraction of their development cost. Today, while OpenAI, Anthropic and Google still offer the world's best-performing AI tools, DeepSeek and smaller Chinese rival Moonshot now feature among the top 12, according to LiveBench's large language model rankings. And they are far cheaper: OpenAI's GPT-5.2 cost $14 per million output tokens (the units of data processed by AI models in response to a user prompt) as of February -- around 33 times more than DeepSeek's V3.2-Exp, at 42 US cents per million tokens. Anthropic has accused DeepSeek, Moonshot and another Chinese AI lab, MiniMax, of "industrial-scale distillation attacks" -- illegally extracting capabilities from its proprietary Claude model using 24,000 fraudulent accounts to gain an edge. Since then, Anthropic, OpenAI and Google have been collaborating to try to stamp out such practices. DeepSeek, Moonshot and MiniMax did not respond to emails seeking comment. Open architecture The new crop of AI tools work by ingesting all kinds of information recorded by humans to spot patterns in the data and record the most common relationships between syllables, words, sounds or pixels. From this, they build a vast neural network that can break down the components of a written prompt and send back the most probable useful response based on the training data. The leading US platforms don't publish the parameters of AI models such as ChatGPT, Claude or Gemini. So third-party AI developers are either unable to replicate them or they need to pay for that privilege. While Chinese AI companies can be just as secretive about what data they've used to train their systems, many have made the parameters assigned as numerical values and tweaked during the training of their AI models -- known as weights -- available for download. This allows universities, startups and smaller tech firms to build upon and adjust the models so they can better perform specific tasks at no extra cost. This approach effectively crowdsources national AI research and development, lowering the cost of maintaining and updating the models. And the added model flexibility eases the adoption of AI across a swath of industries. R1 saw rapid uptake after DeepSeek released the model's weights. Developers have produced versions of R1 for finance, medicine and Chinese language tasks. Alibaba's Qwen family of AI models surpassed 1 billion downloads by January, making it the most widely downloaded open-source AI model family ahead of the Llama platform of Mark Zuckerberg's Meta Platforms Inc. Alibaba said Qwen has spawned more than 200,000 derivative AI models worldwide. Regional governments in China have been funding open-weight models and encouraging the emergence of AI communities that host thousands of models and datasets as communal assets that can be shared for the benefit of local companies. China's government sees this approach, designed to broaden the technology's adoption across the economy even if it cuts into the profits of individual tech firms, as a tool of soft power. It's an active sponsor of initiatives such as the July 2025 Global AI Governance Action Plan that seek to make AI models more accessible to software developers in lower-income nations. Energy As training and operating AI services consumes massive amounts of electricity, their rapid adoption is straining energy networks around the world. In some countries, it's proving a struggle to install the new power stations, solar and wind farms, transmission lines, substations and transformers required to handle the surge in demand. This is proving to be less of a problem in China, which has been adding generation capacity faster than other countries. In some Chinese regions, such as Inner Mongolia, abundant and cheap electricity from renewable sources has led to the construction of hundreds of data centers dedicated to training AI models. China's government is capping electricity bills for its AI companies to give them a competitive advantage over foreign rivals. The state is subsidizing half of the electricity costs of some of the country's biggest data centers -- as long as they use solely Chinese chips. Industry In much of the world, the push for AI adoption is most intense in business services: automating customer relations, accelerating business workflows and accelerating productivity in sectors such as finance, law and healthcare. China's state-directed approach places the emphasis on raising national productivity by embedding AI in commerce, manufacturing and logistics. That's perhaps not a surprise: The US economy is largely services-based. China is the world's dominant producer of manufactured goods, including electric vehicles and other low-carbon technologies, an e-commerce powerhouse and an emerging leader in robotics. China installed more than half of the world's new industrial robots in recent years, according to the International Federation of Robotics, as Chinese manufacturers automated inspection, assembly and logistics. The government has made clear its goal is not to create the biggest, wealthiest AI companies, but to use AI to gain a technological edge to accelerate economic development. US companies and academic institutions lead in the basic, long-term scientific work aimed at advancing the foundational technology behind AI. China's AI community is more focused on accumulating massive data sets from deployed AI that can be fed back to improve models. Carmaker BYD has embedded AI in its EV assembly lines to shorten cycle times and reduce material waste. It's documented a 40% decrease in battery faults and a 20% improvement in battery longevity that it attributed to AI-enabled quality assurance. Apple Inc.'s Chinese iPhone assembly partner, Foxconn, has deployed AI-driven computer vision to detect defects and optimize energy use on production lines. It said this resulted in a double-digit percentage reduction in energy consumption at some sites, while automated quality inspections carried out with AI cut defect rates. E-commerce giant Alibaba's AI systems now handle most routine customer service. The government is encouraging, and even providing funding, for the deployment of AI-enabled robots in manufacturing, logistics and service tasks as a way to overcome growing manpower shortages due to a shrinking working-age population. Officials in Beijing are proceeding with caution, given the risk that a sudden, wholesale shift to automation might lead to mass layoffs and cause a wave of social unrest. National Strategy Even if China's tech entrepreneurs wanted to join the computing-power arms race being pursued by the American AI giants in search of giant profits, it's not certain that they could. Chinese President Xi Jinping already made clear his disapproval of ostentatious wealth when he cracked down on the country's free-wheeling billionaires from 2020 to 2023. He wants the nation's capitalists to serve the Chinese Communist Party's overriding objective -- steady, job-creating economic growth. In exchange, Xi's government has been creating a national ecosystem to supply the country's AI startups with scientific talent, cheap capital and commercial opportunities. A 2017 New Generation AI Development Plan aligned incentives across scientific institutions and the private sector. Universities would drive research and companies would commercialize their breakthroughs. The latest state AI initiative launched in 2025, AI Plus, aims to embed the technology across the economy and society. Primary schoolchildren are learning to code and put together AI models. Local governments have built subsidized AI parks, and firms receive capital, easy access to a wealth of data for model training and guaranteed procurement opportunities. Several Chinese AI startups were founded by professors or alumni of elite universities following scientific breakthroughs and then quickly brought into the government's orbit. AI developer Knowledge Atlas Technology JSC Ltd., or Zhipu AI, was founded by Tsinghua professors Tang Jie and Li Juanzi in 2019 and was encouraged to align with national goals. It secured government customers and backing from state funds such as Beijing Zhongguancun Science City Innovation Development and the Chengdu regional government, and benefited from state-supported computing resources and other policies favoring domestic AI champions. All that state backing has compelled Chinese AI companies to align with national imperatives, so wide-scale deployment of the technology is set to take precedence over profits, according to a Bloomberg Intelligence deep dive reportBloomberg Terminal published in December. Expanding abroad The government in Beijing views Chinese open-weight models that can be embedded across industries as a national asset, akin to roads, power grids or telecommunication networks. This asset can be exported to other countries to foster dependence on Chinese technology. Chinese and US AI platforms are competing in regions of Southeast Asia, the Middle East and Africa where the possibility of Chinese entities accessing government, corporate or personal data stirs fewer concerns than in Western countries. Chinese tech giants Alibaba, Huawei Technologies Co. and Tencent Holdings Ltd. are seeking customers for their own cloud platforms in some of these fast-growing markets. Their plan is to undercut US rivals there by offering everything a customer needs to start using Chinese AI: the platform itself, such as Qwen or DeepSeek, low-cost, China-based cloud computing, an internet connection and the necessary hardware. They've adapted Chinese AI models to handle local languages and adhere to national regulations. And the Chinese government has sought to smooth their path by extending financing to many of those countries under its Belt and Road Initiative and other programs. The approach appears to be working: Chinese generative AI models accounted for about 15% of global market share in November 2025, up from around 1% a year earlier, according to AI model marketplace OpenRouter. Threat to American AI? For now, the big Chinese players' access to the US and European markets is fairly limited. Rules and regulations to do with data privacy and national security mean customers there -- especially government entities and large corporations -- are discouraged from using Chinese cloud computing services. This hasn't stopped big US cloud providers such as Microsoft Corp. and Amazon.com Inc. from offering Chinese AI models to their customers and handling the inference process in data centers outside China. The Chinese platforms are being offered as a more flexible and adaptable alternative to US models because of their open-weight approach. And because they tend to use less processing power in responding to user prompts, they can be a lot cheaper. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Plus Signed UpPlus Sign UpPlus Sign Up By continuing, I agree to the Privacy Policy and Terms of Service. Both Chinese and US AI companies are spending heavily on the computing power needed to run their platforms. They buy that data-crunching capacity as a service from cloud operators such as Alibaba, Google, AWS or Microsoft. The huge investments being made in AI infrastructure are predicated on the expectation that these platforms will gradually become so indispensable to working life that the companies will be able to significantly increase the fees they charge users. For now, no one can say with any certainty when, or even if, those revenues will begin to cover all the spending. That's as much the case for Chinese AI companies such as Minimax -- which is burning through large amounts of cash to pay Alibaba for computing capacity -- as it is for OpenAI, Anthropic and other big-spending American players. Alibaba this year released several closed, proprietary AI platforms alongside its open-weight models, in a possible sign that Chinese players are also feeling the pressure to boost returns on their AI investments. The concern for investors in American AI is that the lower cost base and improving performance of Chinese models will make it harder for the US giants to claim their services justify premium prices, and that significant numbers of users in Western markets will put aside any misgivings over data security and opt for the Chinese alternatives. "US-allied countries are not going to be using DeepSeek in any official way," said Robert Lea, a senior analyst at Bloomberg Intelligence. "But in these cost-conscious times, if DeepSeek offers 90% of the functionality of ChatGPT, then consumers might take a different view to governments."

Meta's Chinese stumble suggests a declining tolerance for shades of grey

Tech-related capital flows have benefited from decades of ambiguity, but AI changes the calculus Chinese social network Weibo published its annual report last week. In it was a warning -- the same one the company has published regularly since it first went public in New York in 2014 -- that its corporate structure is a convoluted workaround to sidestep a Chinese curb on foreign investment in tech. It's one of more than a hundred companies that can say the same. Grey areas feature prominently in the history of US-Chinese flows. Sometimes, though, the grey turns suddenly to black and white. Meta Platforms, owner of Facebook, has been ordered by Chinese regulators to unwind its $2bn acquisition of Manus, a maker of AI software. Manus isn't Chinese, but it used to be: it relocated to Singapore last year, before Meta snapped it up. Legal as that might have been, Beijing sees a "conspiratorial" attempt to spirit valuable technology abroad. Practices such as "Singapore washing", as relocating to the city-state for investment reasons is known, or the "variable interest entities" that enable foreign money to enter forbidden sectors, have brought masses of capital to companies that might otherwise have struggled to get off the ground. While China has plenty of tech talent, it's the US that has the cash. Private investment in AI companies in the US reached $286bn in 2025, according to Stanford University; China mustered just $12.4bn. All sides can benefit from these wheezes, so long as everyone agrees to accept the ambiguity. In the case of the VIE, which facilitated listings including Alibaba, tolerance has waned somewhat. Chinese companies that use them must get permission before listing abroad, which now rarely comes. Meanwhile, US politicians from both parties, including Democrat Elizabeth Warren and Republican Tim Scott, claim those who already use the VIE structure could be inadvertently advancing Chinese government goals. AI has changed the calculus further. Unlike social media, ecommerce and gaming, the competition between Chinese and US AI rivals is more direct. A breakthrough from China's Deepseek just over a year ago sent a justified chill through the US tech sector. China's top models lag America's finest, but only by a few months, according to data from ArenaAI. The US already bans high-end chip sales to Chinese companies; for China to demand the unwinding of Meta's deal is a new twist, but fits the zeitgeist. All this is most irksome for Meta itself, which had already melded Manus's teams with its own. There may be a silver lining, though. Meta, Google, OpenAI and their peers have to date escaped federal regulation inhibiting their ability to develop and release ever more potent -- and potentially dangerous -- models. After all, they can argue, if US companies don't do it, China will. Anything that drives a bigger wedge gives them a reason to argue that the AI arms race must be allowed to continue unchecked.

China to Curb US Investment in Tech Companies After Meta Deal

Chinese regulators plan to restrict technology firms including some of the country's highest-profile AI pioneers from accepting US capital without government approval, part of Beijing's broader response to Meta Platforms Inc.'s controversial acquisition of startup Manus. Agencies including the National Development and Reform Commission have told several private firms in recent weeks they should reject capital of US origin in funding rounds unless explicitly approved, according to people familiar with the matter. Moonshot AI, which is considering an initial public offering, was among those that got the guidance from the powerful state planner, according to a person familiar with the matter. Fellow Chinese startup StepFun received similar instructions, another person said, asking to remain anonymous to discuss a non-public matter. Regulators have also decided on similar restrictions for ByteDance Ltd., the owner of TikTok and the most valuable startup in the country, the people said. They don't want the Beijing-based company, which also operates one of China's most popular AI chatbots, to approve secondary share sales to US investors without government approval, one of the people said. The over-arching intent of the latest restrictions is to prevent US investors from taking stakes in sensitive sectors where national security is a priority, the people said. The previously unreported move stems from the $2 billion Manus buyout earlier this year, which triggered a Beijing probe into illegal foreign investment and tech exports shortly after its December announcement. The deal was initially hailed as a template for startups with global aspirations, but critics have since lamented the loss of valuable AI technology to a geopolitical rival. The commission -- a powerful state planning agency with broad policy-making powers -- is now heading a multi-agency probe that includes the Ministry of Commerce into the deal and its repercussions, the people said. Representatives for the NDRC and Ministry of Commerce didn't respond to faxed requests for comment. Moonshot and Stepfun spokespeople didn't respond to requests to comment. ByteDance representatives also didn't respond to messages seeking comment. The new restrictions risk further isolating China's recovering tech sector from the venture backing that has underpinned it for two decades, much of which was sourced from American pensions and endowments. It follows Beijing's decision to restrict "red chips" -- a type of Chinese company incorporated overseas -- from seeking initial public offerings in Hong Kong, threatening to upend a decades-old playbook that helped Chinese companies tap foreign capital by floating overseas. The twin moves suggest that regulators are worried about a leakage of homegrown technology abroad as Chinese-founded startups and companies explore international opportunities. In the wake of the Manus acquisition, many academics decried the loss of a valuable asset to the US. Many worried that the deal would encourage other startups to follow suit. To be sure, Washington has restricted investments into certain Chinese technology sectors, for fear of helping advance its military or economic might. In 2025, US rules designed to curb investment in Chinese-owned semiconductor, quantum and AI companies took effect. But Beijing had for years encouraged its most ambitious firms to seek business and partnerships abroad, including from US financiers, recognizing the need to foster world-class players in areas from electric vehicles to electronics. And global capital allocators have begun re-evaluating China -- a shift accelerated by the breakout successes of AI phenom DeepSeek in 2025. At the heart of the post-Manus debate was the way the startup restructured to make a sale to a foreign company possible before any regulatory review in Beijing. Manus was a Singaporean-incorporated firm, but its founders hailed from China. Launched in March 2025, Manus is a general AI agent capable of automating complex tasks, ranging from S&P 500 analysis to drafting sales pitches. A month later, its parent Butterfly Effect raised $75 million in a round led by Silicon Valley's Benchmark, valuing it at $500 million. The investment triggered a probe by the US Treasury over potential violations of restrictions on investments in sensitive technologies. In July, Manus relocated its China-based staff to Singapore, cutting dozens of roles in the process. Meta announced its acquisition in December after Manus surpassed $100 million in annualized revenue. It remains unclear what other action Beijing will take following its investigation. Manus co-founders Xiao Hong and Ji Yichao had been barred from leaving China, the Financial Times reported in March. Large language model makers remain some of the most coveted investment targets. Beijing-based Moonshot is seeking to raise as much as $1 billion in an expanded funding round that would value the startup at about $18 billion, Bloomberg News reported last month. Its Shanghai rival StepFun, which is considering a $500 million float in Hong Kong, is in the process of unwinding its overseas entities and onshoring capital to meet regulatory requirements, the people said. The restructuring, which could take months and carry significant tax implications, comes as regulators ramp up scrutiny over so-called red-chip firms -- entities registered offshore that house Chinese businesses and assets. ByteDance is the highest-profile private company in China, in part because of the success of video app TikTok. The parent company had to sell a majority stake in TikTok's US operations after a years-long battle with the US government. ByteDance is much more established than its smaller AI competitors and it's not clear the company would seek to raise additional money from outside investors. The company has long been considered a prime candidate for an eventual IPO.

China plans to block US investment in its top AI firms without government approval

Two parallel moves in 24 hours mark a significant escalation of the US-China AI war from chips and exports into capital and models. China plans to restrict its leading technology companies, including top AI startups, from accepting US capital without first obtaining government approval, Bloomberg News reported on Friday, citing people familiar with the matter. No Chinese government official confirmed the report. The move, if implemented, would represent a significant structural shift in how Chinese AI companies access foreign capital, effectively placing US venture capital into the same approval framework that already governs certain technology exports, data flows, and foreign acquisitions of Chinese assets. The timing is not accidental. On Wednesday, the Trump administration announced it would crack down on foreign technology companies, singling out China, that are "exploiting" US artificial intelligence models, a practice known as model distillation. White House Director of Science and Technology Policy Michael Kratsios framed the move as the first major US government response to complaints from Silicon Valley AI companies that Chinese developers have been using open-source or commercially accessible US AI models as training data to build rival-generation systems, thereby closing the capability gap without having to develop from scratch. Bloomberg characterised the US move as targeting Chinese firms "improperly" using American AI models. Together, the two announcements describe a 24-hour escalation in which both governments moved simultaneously to sever the remaining channels of AI technology and capital transfer. The US is trying to prevent its models from being used to train Chinese competitors; China is trying to prevent American money, which carries intangible benefits including managerial expertise, talent networks, and strategic access, from flowing into its AI national champions without state oversight. Each move is a response to the other's prior actions, and each creates the conditions for the next retaliation. The backdrop to China's reported capital controls is the existing US outbound investment rule that came into effect on 2 January 2025, which prohibits US persons from making equity investments in Chinese companies engaged in advanced semiconductors, quantum computing, or certain AI systems without Treasury Department approval or notification. China's reported plan is, in structural terms, the inbound mirror of that US rule: requiring government approval before Chinese AI companies accept capital from the country that has also been restricting chip exports to China since 2022. The model distillation question is the more technically novel of the two moves. Chinese developers have used DeepSeek-R1, Meta's open-source Llama models, and other accessible US models as training signal for their own systems, a practice that is currently legal under open-source licences but which US AI companies argue gives Chinese labs an unfair structural advantage. DeepSeek V4-Pro, released earlier today and covered separately by TNW, was trained with Huawei chips and claimed near-frontier performance; whether it also incorporated distillation from US models is a question the administration's new framework would directly address. The enforcement mechanism for the distillation crackdown has not been specified publicly; the question of how a government would prevent training data from crossing borders is technically and legally unsettled. The commercial implications for Chinese AI startups are significant but uncertain. Companies like Moonshot AI, Zhipu AI, MiniMax, and the entity formerly known as Manus AI have been navigating a capital environment that was already constrained by US regulatory signals. If the approval requirement is implemented, it would add a formal layer of Chinese government oversight to any US VC investment in those companies, potentially chilling investment further or driving more of China's AI capital formation through domestic channels. The Chinese government has been increasing state investment in AI infrastructure and has made no secret of its preference for domestic AI champions over internationally capitalised ones. A formal approval regime for US investment would be consistent with that preference. What neither measure resolves is the underlying dynamic driving it: China's AI capabilities are improving faster than the export controls are degrading them. DeepSeek V4, released today, claims near-frontier performance on coding and mathematics using Huawei chips, not Nvidia ones. The US controls on chip exports and investment were premised on a widening capability gap; that gap is narrowing. The question both governments are now answering is not "how do we maintain the current technological order" but "how do we shape the terms of a competition that is already fully joined."

China to curb US investment in tech companies: Report - The Economic Times

Chinese regulators, including the National Development and ⁠Reform Commission, have recently instructed several private technology firms ⁠to reject U.S. investment in funding rounds unless explicitly approved,, the report said.China plans to restrict top technology firms, including leading AI startups, from accepting U.S. capital without government approval, Bloomberg ⁠News reported ⁠on Friday, citing people familiar with the matter. Reuters could not immediately verify the report. Chinese regulators, including the National Development and ⁠Reform Commission, have recently instructed several private technology firms ⁠to reject U.S. investment in funding rounds unless explicitly approved,, the report said. AI startups Moonshot AI and StepFun were among the companies that received the guidance, the report said, adding that TikTok owner ⁠ByteDance has also been told it should not allow secondary share sales to U.S. investors without clearance.

Where has Chinese AI gone?

Fifteen months later, history repeats itself. Last Friday, DeepSeek launched V4 in preview. Open source, boasting 1.6 trillion parameters, and trained on Chinese Huawei Ascend 950 and Cambricon chips. Following the news, SMIC rose 10% in Hong Kong, while Hua Hong jumped 15%. However, this time, the Nasdaq didn't flinch. Between coffees in the newsroom, we have been wondering for some time: where exactly does Chinese AI stand? The silence has just ended In a statement DeepSeek claims that V4-Pro "rivals the world's best proprietary models" and outperforms all open-source models in agentic coding. It remains surpassed only by Gemini 3.1 Pro in general knowledge. According to CNN, V4 was trained on Chinese Huawei Ascend 950 and Cambricon chips, not Nvidia. We will return to this. It remains to be seen how the model performs under real-world conditions: the version is still in preview. DeepSeek itself acknowledges a 3-to-6 month lag behind GPT-5.4 and Gemini 3.1 Pro. However, the signal is clear: China has just proven it can deliver cutting-edge performance via open source. And Alibaba is not trailing behind. According to the official technical sheet published on Hugging Face, its Qwen 3.5-9B, released on March 2, scored 81.7 on GPQA Diamond, a PhD-level reasoning test. OpenAI's gpt-oss-120b, which carries 13x more total parameters (the internal settings determining a model's size), peaked at 80.1. Behind these advances lies a paradox. The dollar time lag According to the Stanford AI Index 2026, the US injected $285.9bn in private AI investment in 2025, compared to $12.4bn in China, i.e. a 23-fold difference. And the gap is set to widen: the four American hyperscalers alone plan nearly $700bn in AI infrastructure investment for 2026. US soil already hosts 5,427 data centers, ten times more than any other country. Yet, the more the Americans spend, the more the gap narrows. According to Artificial Analysis, two Chinese models (GLM-5.1 and DeepSeek V3.2) now feature in the global top 10. DeepSeek V3.2 is 33 times cheaper to use than Claude Opus 4.7 (as we detailed in a previous article). V4-Pro, released Friday, drops to $3.48 per million output tokens, compared to $25 at Anthropic, according to Fortune. However, while the focus remains on benchmarks and billions raised, the real game is being played elsewhere. When Beijing locks everything down This is where the landscape becomes complicated. Earlier this week, Beijing officially blocked the $2bn acquisition of Manus -- a Chinese agentic AI gem -- by Meta, according to Bloomberg. This is the culmination of an investigation launched in January, after the two co-founders were banned from leaving the country. The message to Chinese entrepreneurs is crystal clear: you can grow, but you cannot escape. And the lockdown does not stop there. On April 22, US Commerce Secretary Howard Lutnick admitted to the Senate that Nvidia has yet to sell a single H200 chip to China, despite Trump giving the green light in January. The reason? Beijing itself is refusing authorization, "to keep its investments focused on its own domestic industry," according to Reuters. The reversal is striking: it is no longer Washington blocking; it is now Beijing closing the door. And that changes everything. The other signal is the local value chain that China is deploying at an accelerated pace. Take BYD, a global leader in electric vehicles. Its autonomous driving system, standardized across all models including the $9,500 Seagull, relies on DeepSeek R1 AI. Chinese car, Chinese AI, Chinese consumers. Only one link remains missing: the chips, currently a mix of Nvidia and Horizon Robotics. So, where is this Chinese AI? Everywhere. In the Ascend chips that Huawei produces with its Chinese partner SMIC, sometimes using TSMC dies obtained through indirect channels according to the RAND Corporation. In the startups that Beijing now prevents from leaving or being sold. In the models that rival the best American offerings for 33 times less. With one caveat, however: according to an FT investigation reported by CNBC in July 2025, over $1bn worth of export-banned Nvidia chips continued to enter China via parallel channels within a 3-month period. Is Beijing blocking American chips out of pure patriotism, or simply to force its industry forward while relying on the black market to avoid falling behind? Regardless, America has not lost the game yet: three American models still lead the Artificial Analysis rankings (Claude Opus 4.7, Gemini 3.1 Pro, and GPT-5.4). But the era of comfort is over. China is not chasing Silicon Valley. It is building its own ecosystem with its own local stars. And while the Americans are closing their models (Meta launched Muse Spark, its first proprietary model, in early April after the failure of Llama), China is releasing them as open source. DeepSeek, Qwen, GLM-5.1 -- all free, all downloadable. Chinese low-cost is not just about price. It is also a strategy of influence.