Chinese AI models spark debate over US leadership as profitability questions loom

Making sense of the panic over Chinese AI

The launch of the latest AI model from a Chinese company -- Moonshot AI's Kimi -- reignited debates around American competitiveness and open versus proprietary AI. While there was plenty of conversation on social media, it seems the debate is also happening behind the scenes in Washington, D.C., where OpenAI and Anthropic have reportedly lobbied regulators with concern about open Chinese models. On the latest episode of TechCrunch's Equity podcast, Kirsten Korosec, Sean O'Kane, and I discussed why this seems to be such a hot button issue. Beyond suggesting that certain folks should "touch grass" rather than spending their weekends arguing on X, Sean noted that in many ways, this "feels like we're seeing repeats of prior freakouts," with everyone in Silicon Valley "expecting that something is going to arrive and blow everything else away." And Kirsten noted that putting heavy restrictions on Chinese AI models could primarily benefit a handful of companies: "Are we accelerating and ensuring that Americans win the AI race, or are we ensuring that certain frontier labs do better than others? Keep reading for an excerpt of our conversation, edited for length and clarity. Anthony Ha: For folks who have followed the discourse around Chinese AI, this will probably be very, very familiar from the launch of DeepSeek, where basically a Chinese model comes out; on some benchmarks, it does as well, or at least seems competitive with some of the frontier models; and a certain portion of the tech industry loses their mind. Some of this [debate] got extra scrutiny because one of the people posting about it was [an executive] at OpenAI. But in general, there [is] this recurring question of: Can Chinese companies beat US companies, at least in some aspects, and do it much more cheaply and in a much more open way? Sean O'Kane: Yeah, there are many elements of this that feel like we're seeing repeats of prior freakouts. I think one of my favorites is: Everybody is so ready [for] and so expecting that something is going to arrive and blow everything else away. And I think my favorite example of that this past week was people showing off that "My gosh, Kimi made in 30 minutes an entire replication of macOS." And yeah, it made a pretty impressive graphical reproduction of what macOS looks like, but it's not an OS. We keep seeing these things happen over and over again, where everybody's so jumpy in the tech industry. And I think in particular, with some of the Chinese models that come out, there's this expectation, and I think this gets to the core of why people reacted the way they reacted last weekend. (Also, by the way: Go outside, touch grass, it's the weekend. Everybody in the industry was trading barbs on Twitter all weekend.) But this jumpiness is really interesting to me because we're now a week out and I don't think anybody's feeling like the end is nigh like they were a week ago. Kirsten Korosec: We have a really great story by one of our reporters, Tim Fernholz, who tries to unpack the psychosis around this here in the United States. He points to a number of reasons. And concludes -- and I don't want to conclude it for him, but I think that there's one that rises more to the top than others. There's concerns that these Chinese open weight models might have an implicit bias towards China, there's another worry about security risks and guardrails. But there's also a pretty big idea here, which is protectionism, and who is going to quote-unquote "win the race"? Is it going to be the US or China? And that seems to be driving a lot of what the fear is. I don't know, Anthony, if you agree with that? Anthony: I completely agree. I think the China aspect always adds this certain level of hysteria. And that's not to say that people shouldn't be concerned about how the U.S. stacks up against China across different industries. But it gets so amped up. The other thing this reminds me of is the discussion around TikTok a few years ago. And again, it wasn't that I thought that the concerns around TikTok were totally made up, but that the level of how panicked people got -- it seems as soon as you add the word China to any discussion, things just ramp up dramatically. And then in this case, it's linked to this discussion about open [weights] and this idea that AI is so powerful and so dangerous that the only way we can control it is with these proprietary models from these American frontier companies. Obviously, most people saying this [have] reasons why they want to say that. David Sacks, who was the AI czar for the Trump administration [and] now has a different role in the Trump administration, was shouting on X about how, "I can't believe people are opposing data centers, we're tying ourselves in knots, there's too much regulation." And so it's a way to argue for the positions that they already had around AI. "My gosh, if China beats us, that's unthinkable, so you have to do what I want to do anyway." Kirsten: Right, and if you were to put across-the-board bans on Chinese open weight models -- I'm not saying that there aren't real concerns here, but let's just play that out. If we were to do that, it would benefit models created by OpenAI, for instance, and it would force enterprises to use those as opposed to using models like Kimi. So you really have to ask the question: Are we accelerating and ensuring that Americans win the AI race, or are we ensuring that certain frontier labs do better than others? Sean: At this point, we should say a lot of this discussion really got kicked off by the head of strategic futures at OpenAI, Dean Ball, who was the first one to come out with this really long post mentioning some of these concerns. Part of me thinks the reaction to this was because people disagreed with what Dean wrote. Part of me also thinks the reaction was driven by the fact that he kind of just said the thing out loud. He basically said the US should create regulatory FUD -- fear, uncertainty, and doubt -- and muck up the ability for these open weight models to compete with the US. [Ball later backed away from this argument.] And to me, I think you can read in some of the responses from folks, like, "You're not supposed to say that out loud, Dean."

The AI giants' new problem: open AI

Silicon Valley has spent much of the past week on red alert, digesting the arrival of Moonshot AI's Kimi K3, a Chinese AI model that can allegedly beat some of the best systems built by US companies at a fraction of the cost. Its performance alone would have been enough to intensify the rivalry between the US and China. But Moonshot's plan to release the model's weights for free -- and its clear targeting of US users -- has fueled deeper unease about whether closed American models can continue to dominate as increasingly capable open alternatives enter the market. Open-weight models give developers far greater control than proprietary systems, allowing them to inspect how the AI functions, run the AI locally on their own infrastructure, customize the systems, and build new products without depending on a single provider. They're often a lot cheaper, too. That raises an obvious question: Why would an AI company spend vast sums of money training an AI model, only to give away some of the most valuable parts? Kimi K3, like other open-weight AI models, isn't fully "open." In software, "open source" has a settled definition: Source code is publicly available to use, modify, and redistribute freely, only requiring that this is also done openly. AI systems are more complicated, and very few are truly open in the traditional software sense. Most companies instead release something called model weights -- the numerical parameters learned during an AI's training period -- while keeping other crucial components, including training data, code, model architecture, and configuration methods, private. Most also come with restrictive licenses limiting how they can be used or redistributed. Together, this means open-weight AI cannot be re-created from the ground up in the way true open-source software can. But it does provide enough power and flexibility that a company can make money off of it. "A free set of weights is not a free AI service," said Fordham Law School professor Chinmayi Sharma. "A company can give away the model weights while making money elsewhere in the stack." There are ample opportunities to do so. Running a model still requires computing infrastructure, engineering, security, maintenance, and support, all of which companies can charge through hosted access or other arrangements. For some companies, the payoff may be broader, such as an increased demand for cloud computing services or advanced computer chips. Openness can also be a powerful strategy for gaining a competitive edge. Releasing a model's weights can encourage more companies and developers to use it, which in turn can lead to an entire ecosystem of tools and infrastructure being built around it. Over time, that can help a model become a "de facto standard," Sharma said. Kyle Miller, a senior research analyst at Georgetown's Center for Security and Emerging Technology, made a similar point, citing Alibaba's large family of Qwen open-weight AI models in China as an example of how deeply embedded an open system can become across an industry. That creates a clear problem for the US AI giants. If a generation of tools and developers start building around capable open-weight models like Kimi K3, the industry's center of gravity could start to shift away from proprietary platforms like Gemini, Claude, and ChatGPT. While it remains to be seen whether frontier-level open-weight models are actually cheaper to run in practice, they have historically offered a lower-cost alternative to proprietary systems. They also offer more freedom for developers at a time when US labs are tightening access and imposing stricter guardrails for their latest models. There are already signs that some US companies are shifting toward cheaper Chinese models. There is no single reason behind China's support for open-weight AI, but it appears to be a mix of practical constraints and political strategy. An open ecosystem gives Chinese companies a way to innovate near the frontier despite tighter access to advanced chips and computing power, while fitting neatly into Beijing's broader industrial strategy of encouraging wider adoption of Chinese models, tools, and infrastructure. The approach is also convenient for expanding China's technological influence abroad, as well as its political influence. For example, earlier this month, President Xi Jinping openly challenged the US for leadership of AI on the world stage by pitching itself as a more egalitarian partner given America's closed approach. The rise of capable Chinese open-weight models is also turning up the pressure on closed-model providers like OpenAI and Anthropic from within their own industry. The prospect that the US might restrict access to open-weight AI in light of Kimi K3 sparked a swift backlash in the tech sector, supported by some of its biggest players. A coalition of 25 tech companies, including IBM, Microsoft, Meta, Nvidia, Perplexity, and Palantir, released an open letter urging policymakers to avoid "premature restrictions," arguing that open-weight AI models are essential to ensuring American AI leadership and preventing the technology's power and benefits from becoming "concentrated in a few hands." Most of those unnamed giants -- including Google, OpenAI, and Anthropic -- were conspicuously absent from the original list. That pressure intensified again on Monday, when Nvidia, Microsoft, SpaceX, and a broader group of major tech companies called for stronger US support for open-weight models. The initiative was a direct response to concerns over the safety of advanced AI systems after a rogue OpenAI model escaped containment and attacked another company during testing, which had to rely on a Chinese open-weight model to defend itself on account of the strict safety guardrails on US frontier models. It's unclear how much ground the largest US AI labs are prepared to give. Google and OpenAI later joined the cautioning against hasty restrictions on open models, though neither signed on to Monday's cyber-focused initiative. Anthropic, notably, has backed neither effort. Miller said it's an "open question" how this all plays out in the long term. US companies could release more capable open-weight models of their own, he said, noting that pressure from Chinese companies was partly why OpenAI released the open-weight GPT-OSS last year. "But I don't think companies like Anthropic will go in that direction," he said. Google's open-weight Gemma models are also partly viewed as a response to Chinese competition. Neither is nearly as capable as either company's proprietary flagship model. "The question for American firms may increasingly become: How much capability do we need to release openly to prevent Chinese models from becoming the default platform for the open ecosystem?" Sharma said. A more plausible outcome could be a "portfolio strategy," she said, with companies keeping "their very best model proprietary while releasing increasingly capable open-weight models to maintain developer adoption and ecosystem influence." It will take some time to see whether Kimi K3 wins over US developers or not. But with Beijing increasingly championing open-weight AI, it will almost certainly not be the last model that will try to crack America. The question facing the country's biggest AI companies is no longer just how the US can stay ahead of China, but whether closed AI can -- or should.

The US should stay open to open-weight AI

New and cheaper models could be a boon for the growth of the tech sector Open-source software was the engine of the internet era. Today's digital world was built on open-source tools such as Linux and Python. In the age of AI, widespread open-weight AI models -- those that are publicly released allowing users to run and customise them -- will also play a key role. China's advances in open-weight models have prompted some Trump administration officials to call for them to be restricted. But such technologies are crucial to create a dynamic AI ecosystem. The release of Kimi K3, a model from Chinese company Moonshot that has performed well on certain industry tests, shows how Chinese open-weight models are closing the gap with the leading US models, despite China's restricted access to leading-edge chips. Experts suspect these models were trained using large-scale distillation, a process that uses the output of one model to build another. US officials are taking these concerns seriously: Treasury secretary Scott Bessent has said that sanctions are on the table. Worries over Chinese models have contributed to the recent tech stock sell-off. Investors fret that the spread of such technology could be bad for large US AI labs that make money selling access to closed models. Cheap Chinese models put pressure on pricing and call into question the returns investors expect from US labs. US "hyperscalers" are spending hundreds of billions on computing power, underpinning the strong valuations of businesses such as chipmakers. Policymakers, too, have identified risks. There are legitimate intellectual property concerns around distillation -- as well as over safety. Open-weight models are harder to police, and safety testing is more difficult for models that can be downloaded and modified by users. Nefarious actors could use them to create chemical or biological weapons. AI labs and governments must together develop safety standards and testing capacity for new models. But protectionism from the US would be misguided. An open letter shared on X by Nvidia founder Jensen Huang and signed by companies including Palantir, Meta, OpenAI and Google emphasised the many benefits of open-weight models. The letter highlights the necessity of a "strong, open ecosystem that diffuses into every sector . . . creating opportunities for innovation and prosperity across the country." It warns that premature restrictions on these emerging open models could hurt domestic competition and innovation. Anthropic's founder Dario Amodei released a separate post saying the company did not support a ban on open-weight models, but called for mandatory safety testing. Necessary concerns over safety and intellectual property should not become a block on innovation. A generation of companies would benefit from access to cheaper AI tools -- even if that puts pressure on the top US labs. Since open-weight models can provide many general-purpose uses more cheaply, competing US labs will need to speed up efforts to develop agentic tools that can complete specific, more complex tasks. In a recent speech, President Xi Jinping highlighted China's ambition to export AI tools and knowhow via its "digital silk road", and its lower-cost models are more affordable for users in developing countries. The US should be wary of Chinese moves to dominate the AI future. But its best response would be to encourage more alternative open-weight models of its own, not restrict those from China. That might also help to encourage co-operation from Beijing on standards and testing. Despite the challenge that open-weight models pose to US frontier labs, newer, cheaper models should help drive AI adoption across the economy. If that raises prosperity and growth, it is good for the US tech industry too.

Even China's A.I. Powerhouses Can't Figure Out How to Profit Off A.I.

China's leading technology companies have been steadily releasing artificial intelligence models that perform nearly as well as the best systems in the world. But they are all facing the same problem: overcoming the punishing economics behind the technology. Chinese A.I. companies have tried different approaches to bring in enough money to sustain the enormous expense of building A.I. systems. Start-ups like DeepSeek and Moonshot AI have raised billions from investors. Alibaba, China's A.I. heavyweight, has started charging users for access to its best models. ByteDance, the parent company of TikTok and an A.I. powerhouse itself, launched a tiered pricing system, hoping to get people to pay more for using its most advanced models. To build state-of-the-art A.I. models, A.I. companies constantly need to buy powerful computer chips -- enough to build the models, more to test improvements and still more to ensure they can perform for users all over the world. The companies also need to build or rent space in data centers that house all this computing power. Chinese A.I. companies are not the only ones confronting these tough economics. Their giant rivals in Silicon Valley, from OpenAI and Anthropic to Google, are also investing more in A.I. than they are earning from it. For Chinese companies, the challenge is compounded by a central paradox of the country's approach to A.I. development. Most of China's leading A.I. systems are open source or open weight. That has accelerated their development -- the entire industry gains when every company shares its work in public. China's open-source approach has spawned a crowded field of innovative and intensely competitive start-ups all offering systems at low cost. So attempts to increase revenue by charging more for access to certain models can scare away customers. Price-conscious Chinese consumers -- businesses and individuals alike -- are quick to hop across platforms in search of inexpensive A.I. tools. While China's A.I. companies are searching for sustainable business models, spending is high and revenue low, said Richard Lin, a vice president at the Silicon Valley company Datastrato. "In two or three years, we will still be trying to figure out how large models can earn money," he said. Offering low prices has helped the Chinese firms gain users, including in Silicon Valley, where many companies depend on the more affordable systems. Yet Chinese companies have struggled to translate huge numbers of global users into profits. Last month, the Chinese start-up Z.ai released a model, GLM-5.2, that it said was nearly as powerful as Anthropic's best. Many software developers and start-ups in Silicon Valley quickly started using it, in part because it was cheaper than the American systems. Z.ai's models improved and became more popular, and the company's revenue more than doubled last year. But Z.ai lost nearly $700 million. When it went public in Hong Kong, Z.ai said it planned to use most of the money to improve its models, largely by acquiring more chips. A few weeks after the stock listing, Z.ai said it was looking for partners to share computing power resources and apologized after users complained about slow service. "Open models are a powerful distribution strategy, but they are not a complete business model," said Wei Sun, a principal A.I. analyst at Counterpoint Research in Beijing. "An I.P.O. can finance the next training cycle, but it cannot by itself create sustainable economics." This month, Moonshot released a model called Kimi K3, which the company said performed better than models from OpenAI and Anthropic did on some tasks. But within two days the company was forced to announce that it needed to stop accepting new users because it couldn't get enough computer chips to serve them. China's A.I. industry has faced years of U.S. trade restrictions that confine its ability to buy the world's most powerful chips. To get around these limitations, many Chinese companies rent remote access to data centers outside China stocked with advanced chips. But Chinese A.I. start-ups have far less money to buy computing power than their deep-pocketed American rivals. Last month, DeepSeek held one of China's most anticipated funding rounds, raising money from investors including the internet giant Tencent, the battery maker CATL and the country's state investment fund for artificial intelligence. DeepSeek raised $7.5 billion. In May, Moonshot raised $2 billion. By comparison, Anthropic raised $65 billion in May alone. Chinese tech firms have far less capital, but U.S. tech executives and investors worry that China's open-source push threatens to upend the economics of A.I. production. Chinese A.I. start-ups say years of export controls have pushed them to use chips as efficiently as possible, potentially challenging the idea that has motivated much of global A.I. investment: that making cutting-edge A.I. systems will always require increasing investment in more chips and data centers. Now, a debate is raging in the United States that could make it even harder for Chinese A.I. companies to earn money. Leading Silicon Valley companies including Anthropic and OpenAI have claimed that Chinese firms improperly harvested data from their A.I. systems to accelerate the development of the Chinese models. Some American tech companies and investors want Washington to limit access to Chinese open-source models, saying they could be a threat in the wrong hands. Kevin Xu, the founder of Interconnected Capital, a hedge fund that invests in A.I. technologies, said, "Many Silicon Valley start-ups rely on open-weight models to both customize their product and also not have to pay OpenAI and Anthropic their high prices." These American tech companies have been an important source of revenue for many Chinese A.I. firms. If they are cut out of it, they risk losing that. Xinyun Wu contributed research from Taipei, Taiwan.

China's open AI models may start charging, Goldman says

Goldman says Moonshot and Zhipu could start charging cloud platforms to host their open-weight models. It would be the open-source movement's first real answer to how it makes money. For a year, the story of Chinese AI has been simple. It is nearly as good as the American frontier, and it is effectively free. That second part may be about to change. Chinese developers could start charging cloud platforms commercial licensing fees to host their open-weight models, Goldman Sachs told the South China Morning Post. The bank calls the idea "paid weights." The logic is simple. Moonshot's Kimi K3 and Zhipu's GLM-5.2 now sit a fraction behind the best US systems, and are used heavily worldwide. They ship under open-source terms, so anyone can download and host the weights for free. The labs earn almost nothing from that use. What paid weights would mean The weights are the parameters that encode a model's intelligence. Releasing them openly is what let Chinese labs spread so fast. It is also why that spread earns them so little. Charging cloud platforms to host those weights would change the economics without closing the models. Goldman frames it as a way to turn soaring adoption into revenue. This is a description of an incentive, not a signed plan. No Chinese lab has announced such a fee. The catch Open and free is the whole reason these models matter. Developers reached for them because they were cheaper than American systems and came without a licence to sign. Start charging, and some of that adoption could evaporate. Developers could fork the last free release and keep running it. Rivals that stay fully open would have an obvious pitch. The move that captures revenue could surrender the influence that made it worth capturing. China has been pulling the other way too. Beijing has weighed curbing overseas access to its top models on security grounds. Meanwhile labs like MiniMax keep open-sourcing ever larger systems. Paid weights would add a commercial motive to a debate that has been about national security. Why this lands now The timing is not incidental. Global AI stocks have just been through a sharp selloff, and cheaper Chinese models are one of the cited reasons. A rout in chip shares has swept Japan, South Korea and Taiwan. Investors are questioning whether lofty valuations and the AI spending boom can hold. Japan's Nikkei 225 has fallen 14% from its June peak, and chip names such as Kioxia have slumped more than 40%, Bloomberg reported. South Korea's Kospi has been an early bell for the trade. Goldman's Japan strategist Bruce Kirk called the drop a buying opportunity. He told Bloomberg the bank does not think "the AI story is broken," and that it will come down to earnings. Part of what spooked the market, though, is the fear that cheap Chinese models undercut the returns on hundreds of billions of dollars of US spending. The question underneath That is the same fear now hanging over Western hyperscalers. Wall Street has flipped from rewarding AI spending to demanding a return on it. Business Insider notes it punished Alphabet and Tesla over capex before Meta's results this week. Paid weights is China's version of the same reckoning. The open-source flood gave Chinese labs reach and status, but not income. It echoes a wider pattern that Fortune has traced, in which the country's cheap exports hold down prices abroad without always enriching the exporters. At some point, influence has to pay for itself. Whether developers accept a bill for what was free will decide the outcome. Paid weights is either a business model, or the moment that hands the open-source crown to whoever refuses to charge.

What is the risk of using Chinese open AI models like Kimi K3?

The writer is a senior fellow at New America and the Institute for America, China, and the Future of Global Affairs at Johns Hopkins University Earlier this month, I sat in the Radiohead-themed conference room of Beijing start-up Moonshot and watched a demonstration of its powerful new Kimi K3 AI model. Founder Yang Zhilin, a music fan, named his company for Pink Floyd's album Dark Side of the Moon -- a fitting reference to uncharted yet high-reward territory. Within China, the surprise release of Kimi K3 has been hailed as another chance for the country to flex its tech expertise at the expense of US rivals. More US companies are opting to lower their costs by switching to cheaper, high-performing Chinese open AI models such as K3, Z.AI's GLM, DeepSeek's R-1 and Alibaba's new Qwen 3.8 Max. "Saves us millions of $ and we're actually seeing an *increase* in performance on many core use cases. Transformative," wrote Flo Crivello, CEO of San Francisco-based digital assistant platform Lindy AI, which switched to DeepSeek this year. What is the security risk of using open-source AI models created by foreign companies? Washington seems focused on surveillance, IP theft, espionage and potential Chinese involvement in critical domestic infrastructure. The White House is reportedly exploring tools to curb the spread of, or even sanction, open Chinese models. The debate may look like a stand-off between national security and innovation, but it is not quite that simple. The first point to make is that asking whether a model has been developed by a Chinese company is the wrong question for those concerned about security. Open-source models such as Kimi K3 allow the parameters (the numbers learnt in training that determine how an AI model uses information to predict what comes next) to be downloaded and run by anyone. Companies download the weights, fine-tune the model and then run it on their own cloud or infrastructure. What matters more, therefore, is who owns and operates the server that runs the model. US users worried about data being sent to China could run the AI models on domestic hardware they control or through third-party inference providers. Chinese open models do not uniformly censor information either. Again, where the model is hosted changes things. When hosted on US infrastructure, models will answer questions that would be blocked inside China. Since weights are open, users can change content restrictions. Even if data is hosted within China, where companies are required to co-operate with the National Intelligence Law, some push back in ways that mean the government does not necessarily have unfettered access. In March this year, Chinese government inspectors seeking access to data from the ecommerce platform Pinduoduo got into a brawl with staff trying to block entry. One official left with a broken finger. The US does need a viable open-source option of its own. It is quickly becoming a crucial lever of national power projection. Kevin Xu of Interconnected Capital compares it to Disney movies and K-pop. But the real problem is that Washington lacks a comprehensive approach to shore up critical infrastructure in the face of powerful, AI-enabled cyber attacks that are no longer hypothetical. Here the transparency provided by open weights (including those from China) could in fact improve security. Last week, Hugging Face used an open Chinese AI model to analyse an AI agent cyber breach. This all happened without data leaving Hugging Face's system. Governments on all sides still insist on framing AI as a global arms race. But neither threats nor innovation map neatly on to national boundaries. Addressing data access and control in an interconnected world requires a new way of understanding security protocols.

Thrift-maxxing puts OpenAI and Anthropic IPOs at risk

A million words of AI output costs $50 from Anthropic's best model. The same volume from DeepSeek costs 87 cents. American companies have done the arithmetic, and the Wall Street Journal has a name for the new mood: thrift-maxxing. It is reaching the labs where it hurts, in the valuations they are carrying towards the public markets. Today Moonshot hands out the weights that made the point. Roughly 750,000 words of AI output costs $50 if you buy it from Anthropic's Fable. The same volume from DeepSeek's V4-Pro costs about 87 cents, Fortune reported. Z.AI's GLM-5.2 charges $4.40. Moonshot's Kimi K3, pricey by Chinese standards, charges $15. Corporate America has noticed. Companies spent a year competing to burn tokens. Now they shop for the cheapest model that can finish the job, the Wall Street Journal reported. The badge of honour used to be tokenmaxxing. The new one is thrift-maxxing. The first response to the bills was rationing, when firms capped what staff could spend. The second is substitution, and it is a far bigger problem for OpenAI and Anthropic. "It's like driving a Lamborghini to go to the grocery store to pick up milk when that was designed to be raced around a track," said Mike Saeks, a field chief technology officer at Cursor who advises companies on their AI returns. He has the receipts. Cursor priced up building a web browser from scratch. Running the whole job on OpenAI's GPT-5.5 cost a little over $10,000. Splitting it between Cursor's own Composer model and Anthropic's Opus 4.8 cost $1,339. The orchestra, not the soloist This is not companies abandoning American labs. It is companies demoting them to the parts of the job that justify the price. Telnyx, which builds infrastructure for AI agents, ran 1,000 of them on a top Anthropic model under a $200-per-employee subscription. Then Anthropic stopped allowing third-party operating systems on subscriptions, treating it as a breach of its terms. Paying per use instead would have cost "like 100 grand per day", chief executive David Casem told the Journal. So Telnyx rebuilt the stack. Z.AI models now run its 1,400 agents. Anthropic's Fable acts as the conductor and plans the work. Open-weight models handle the implementation. OpenAI's Sol reviews what they produce. The legal AI startup Harvey trained GLM-5.2 itself. It then gave the model a button to call Fable 5 when a task looks genuinely hard. "We work with all of them," said president Gabe Pereyra. Zoom's chief technology officer Xuedong Huang reached for an old Chinese story. In it, three ordinary people combine their wits to match one genius. That, he said, is the secret sauce. "It really feels like a bloodbath" The labs are reacting the way any supplier does when switching costs collapse. They are paying customers to stay. "There's zero loyalty that I'm seeing," said Marty Kausas, chief executive of the customer support platform Pylon. "It really feels like a bloodbath right now." Pylon has been handed months of unlimited free usage. Kausas reckons it has taken about $1.6m in free tokens from one vendor this year. Another gave $65,000, and a third gave $10,000. Both labs say they are adapting. An OpenAI spokeswoman said GPT-5.6 Sol was trained to be far more token efficient. Anthropic released a powerful lower-cost model on Friday. An executive said customers can choose intelligence or price inside its own ecosystem. Both say they support open-weight models. Why the gap is so wide The price difference is not charity. Electricity costs less in China, and new data centres there meet less local resistance. Chinese firms are also willing to run thin margins to win share and become the default. Export controls may have helped. Cut off from the best Nvidia chips, Chinese labs had to squeeze more out of worse hardware. "For the money [a Chinese AI company would] spend on an Nvidia chip, they can buy 10 local chips from Huawei or other local chipmakers," George Chen of the Asia Group told Fortune. From Coinbase to DoorDash Household names are doing it too. Coinbase chief executive Brian Armstrong said in June that the exchange had halved its AI spending. It did so by pushing staff towards Kimi and Z.AI's GLM models. DoorDash sends what its chief technology officer Andy Fang calls "lower-level work" to Kimi, for "better quality [at] cheaper cost". Airbnb has used Alibaba's Qwen for customer service. Cursor built its own Composer 2 coding model on Kimi foundations. The result shows up in usage. Chinese models took 57% of the tokens US firms consumed on OpenRouter during one week in July. At one point mid-month, all five of the marketplace's top models were Chinese. It reaches well beyond developers. IDC surveyed 260 decision-makers at US firms with more than 1,000 staff. Of those, 47% said they used a made-in-China model for at least one use case, The Daily Upside reported. One in five reported extensive use. The IPO problem Here is why this lands harder than an ordinary price war. OpenAI and Anthropic spent years racing to build the most capable models. Then they found that plenty of customers want basic ones. Both are preparing to list publicly. The Journal reports the shift is threatening the valuations they carry into those listings. It also explains the lobbying. The fight over restricting Chinese open-weight models has run all month. It stretches from the startups opposing a ban to the giants who signed a letter defending open models on Friday. The price sheet is what that argument is actually about. The market has already voted once. Moonshot announced K3 on 16 July. Nvidia lost almost $600bn in value and briefly ceded its place as the world's most valuable company to Apple. Reasons to keep your head Not everyone thinks the sky is falling. On TechCrunch's Equity podcast, Sean O'Kane argued the industry does this every time. "We're now a week out and I don't think anybody's feeling like the end is nigh like they were a week ago," he said. His colleague Kirsten Korosec asked the sharper question about a ban. "Are we accelerating and ensuring that Americans win the AI race, or are we ensuring that certain frontier labs do better than others?" Real frictions remain. Gartner's Igor Marchal warned about going public with it. Naming a Chinese model invites suspicion at best, he wrote, and reputational damage and procurement pushback at worst. Congress is probing Airbnb and Cursor over their use of such models. The big clouds are also holding back. Amazon's Bedrock, Microsoft's Azure Foundry and Google's Vertex AI still do not support these Chinese open-weight models. Bloomberg Intelligence analysts Mandeep Singh and William Tong made the point after their token-volume mix reached 68%, following the launches of K3 and GLM-5.2. Today the weights land Moonshot is due to publish K3's weights for public download on Monday, Bloomberg reported. Anyone will be able to download, modify and host the model. Daily sales at the Beijing company have risen at least sixfold since K3 launched. That date was always in the diary. When Moonshot unveiled K3 on 16 July, the weights were the missing proof, promised for 27 July. They arrive into a market that spent the intervening fortnight learning to shop around. Sam Altman saw the mood shift coming. "This is the first year where AI spend has been a big topic," the OpenAI chief executive told CNBC this month. "And all of a sudden, it's a very big topic."

DeepSeek's new bargain model accelerates AI's race to zero

Why it matters: Tech giants are pouring hundreds of billions of dollars into the computing infrastructure powering the AI revolution. Yet the intelligence that infrastructure produces is getting cheaper by the week. Zoom in: DeepSeek is the same Chinese startup that ignited a market meltdown last January by showing it could build a world-class AI model with far fewer resources than its U.S. rivals. * Its newest model, V4 Flash, performs close to the level of Anthropic's Claude Opus 4.8, one of the industry's most capable systems, on tests of complex coding and autonomous software tasks. * On Arena.ai's crowdsourced leaderboard for front-end coding, V4 Flash debuted ahead of Opus 4.8 -- while delivering the best performance for its price among any model in its class. * The price gap is staggering: DeepSeek charges about 28 cents for the same amount of output that costs $25 on Opus 4.8 -- a 99% discount. Zoom out: With Chinese models like Kimi K3 bearing down on the U.S. market, July ushered in a full-scale price war across the AI landscape. * OpenAI slashed the price of GPT-5.6 Luna -- its fastest, cheapest model for high-volume tasks -- by 80% on Thursday, only three weeks after its launch. * Google released three new Gemini "flash" models all focused on efficiency. * SpaceXAI released Grok 4.5, Elon Musk's most capable model yet for coding, research and autonomous tasks, at the same price OpenAI originally charged for Luna before this week's cut. * Meta quietly reversed course on its longtime embrace of open weights with Muse Spark 1.1, a closed-source model priced aggressively for developers. The other side: Anthropic remains the clearest holdout, keeping its top-tier Claude models at premium pricing and betting that developers will pay extra for safety and precision. Between the lines: When a product becomes a commodity, buyers care less about who made it and more about what it costs. Think electricity or gasoline: Few people know which power plant supplied their home or which refinery produced the fuel in their tank. * AI is heading that way fast. As the performance gap between top-tier models is shrinking, many AI applications no longer depend on a single provider, giving buyers more leverage to shop on price. * "At some point, the next model doesn't matter to you," says Zack Kass, OpenAI's former head of go-to-market and a global AI adviser. He calls the phenomenon "diminishing model returns." What to watch: That could create a lucrative market for "intelligent routers," Vinesh Sukumar, Qualcomm's vice president of AI product management, told Axios. * Those systems would automatically choose the best model for each task based on capability, speed and price -- further weakening the power of any one lab to command a premium. * For frontier AI labs, that could pose an existential challenge: Spending tens of billions to build a slightly smarter model may buy only a temporary lead, without creating lasting pricing power. Reality check: Falling prices do not necessarily doom the frontier labs if cheaper AI unleashes vastly more demand. * OpenAI is betting that companies will use its models so extensively that enormous volume can compensate for thinner margins. * "We will have so much usage of our models that we do not need to be a gigantically high-margin business to be able to afford model training," CEO Sam Altman said on the Invest Like the Best podcast. The bottom line: The U.S. and China are both racing to make intelligence abundant. Now someone has to prove abundance can still be profitable.

Chinese AI startup Kimi K3 sends U.S. chip stocks tumbling

A little-known Chinese AI startup shook Wall Street after its Kimi K3 model challenged OpenAI and Anthropic A version of this article originally appeared in Quartz's AI & Tech newsletter. Sign up here to get the latest AI & tech news, analysis and insights straight to your inbox. U.S. chip stocks just took a tumble, and a Chinese startup most people had never heard of is the reason why. Moonshot AI put out an open-weight model called Kimi K3 earlier this month that seemed to match what Anthropic and OpenAI charge a premium for. The semiconductor index slid into a bear market in response. If this sounds familiar, it's because this happened in 2025 with DeepSeek, when a different Chinese lab showed up with a cheaper model that worked and hit the chip stocks the same way. By last week, some analysts were walking the DeepSeek moment 2.0 back, pointing out that a model Kimi's size still needs a lot of compute to run properly, making it less far ahead than it appeared at first blush. But the instinct to reach for that comparison, even a slightly premature one, says something about how fragile the AI story has gotten. But Mark Zuckerberg, talking about democratizing AI, first spent around $100 billion trying to build Llama into a genuinely open alternative to what Anthropic and OpenAI were charging for -- intelligence anyone could download and run themselves, no subscription required. It didn't work. The model that came out the other end was too slow and too unwieldy for most people to bother with, and Meta $META shelved it and reoriented its AI teams towards building closed systems. China's read on this moment is the opposite. Xi Jinping recently told an audience that his country should treat "open source, openness, collaboration and sharing" as a matter of national strategy in AI, not a side preference, and the labs have followed that lead even when it costs them. Moonshot's Kimi K3 is open weight. Z.ai's GLM 5.2 is open weight. Alibaba, which had actually pulled back to closed models on its most powerful releases earlier this year, reversed course again after Xi's speech and announced its next flagship, Qwen3.8 Max, will eventually go open too. None of these companies are making money doing this yet, but the payoff was never supposed to be the balance sheet. This is a bet on influence and reach that plays out over years, not quarters or IPO dates. It's already working by that measure. Roughly a third of all global AI usage now runs on Chinese open source models, and in markets like Nigeria, Malaysia and Brazil, developers can build on them for more than 90 percent less than it would cost to build on OpenAI. For Washington, that adoption curve looks less like a business problem and more like the next space race, one where influence and infrastructure matter as much as who has the best model. The Trump administration has reportedly considered banning Chinese models outright to slow it down. Some voices close to the administration argue the open Chinese models are unacceptably risky and should be restricted on national security grounds. Others, including venture capitalists who otherwise largely support the administration, argue that framing is really about protecting Anthropic and OpenAI's pricing power, dressed up as a safety concern. A frontier AI model still costs real money no matter who built it. Anthropic and OpenAI are burning cash too, betting that scale eventually catches up with spending. What's different in China is how much more they're willing to burn, and for how long. Mozilla's new State of Open-Source AI report found running costs for these models have fallen roughly 50 times over in three years, and the performance gap with the best closed models is down to about 3%. That's a great deal for anyone building on top of these models. It's a much tougher one for the labs building them, most of which are still losing money hand over fist. This is starting to look like the kind of race to the bottom that plays out in any commodity business, oil, airlines, chips. Prices fall until only the lowest cost producer survives, and everyone else prays for a shakeout. There is no moat in AI right now, after all. Neither side has actually solved for that. Charging a premium only works if there's something durable behind the price, and open source is proving there isn't. Giving models away only works if the costs eventually come down enough to stop bleeding money, and they haven't, not yet, not close. Both bets assume the other problem gets solved on someone else's timeline. That's the part the flag waving skips over. No moat and no cheap intelligence, at the same time, is not a business model. It's just where this industry happens to be standing right now.

Meet the American techies who don't care where 'good enough' AI models come from, even China | Fortune

San Francisco-based Raffi Krikorian, the chief technology officer at Mozilla, which runs the Firefox browser, switched to Chinese AI startup Moonshot's Kimi K3 for many of his day-to-day activities within days of the new, powerful model's launch more than a week ago. "It just seems snappier," he said of K3, comparing it to the acclaimed and more expensive Claude Fable chatbot from San Francisco company Anthropic. Earlier, he had been using another high-performing Chinese model, Z.ai's GLM-5.2, for routine tasks such as managing his calendar, documents and email. Krikorian is one of a growing number of Americans adopting Chinese AI systems, which are gaining ground around the world because they are more affordable and increasingly efficient. U.S. companies like cryptocurrency exchange Coinbase said they're switching to Chinese AI models to help trim costs. Their popularity has frustrated some U.S. tech giants, but short of an outright ban, they're likely to continue to appeal to independent software developers in the U.S. and elsewhere. American-led restrictions block China from accessing some of the world's most advanced technologies including cutting-edge AI chips and U.S. Treasury Secretary Scott Bessent has warned more sanctions could be coming to protect American intellectual property. U.S. President Donald Trump's administration on Wednesday accused Moonshot of using "covert" but not necessarily illegal methods to build K3 off the back of Anthropic's Fable. Some U.S. politicians and AI companies including Anthropic have accused Chinese startups of illicit "distillation" of their models to extract their technologies, a claim that Beijing rejects as "groundless." China's AI models are cheaper and 'good enough' The rapid rise of cutting-edge Chinese AI models this year shows the China-U.S. race in AI has moved on from early last year, when Chinese startup DeepSeek shook up the U.S. technology industry and put China on the map with an AI model that performed on par with those in the U.S. but much more cheaply. The latest AI models launched by startups Z.ai, or Zhipu, in June, and Moonshot in July are nearly as intelligent as the frontier models from OpenAI and Anthropic, experts say. Also in July, China's Alibaba previewed its Qwen3.8 Max AI model. In April, DeepSeek rolled out previews of its latest V4 model, challenging OpenAI's GPT and Google's Gemini. Those constant advances are bearing fruit. Curt Meinhold, a Greensboro, North Carolina-based technology executive, said he increasingly prefers to use DeepSeek for tasks such as finding business leads or ways to generate sales. "At the end of the day, most of us, the vast majority of us, 90 plus percent, don't need (Anthropic's) Mythos or Fable," said Meinhold, who founded digital legacy platform LilyList. "Like, we just don't need it, we need something good enough." Chinese AI models are reaching a "critical stage" for wide adoption, U.S. investment bank Goldman Sachs wrote in a July research report, especially when a surge of global "agentic" AI usage, which requires the AI models to autonomously conduct multistep, complex tasks, is driving higher demand for cost effective AI models. AI prices are calculated per million tokens for both input and output, so the use of AI "agents" is dramatically compounding cost differences, according to analysts including Alex Colville of the Australian Strategic Policy Institute. "The Chinese models, you know, I find them to be pretty close on code and research," said Meinhold. "If I can pay a handful of cents per million output tokens versus 30 bucks or 40 bucks or 50 bucks, then it's good enough." Moonshot's Kimi K3 leads, but Chinese AI models have limits Based on data over the past month, the top five most popular models on OpenRouter, a platform that tracks data across AI models, were Chinese. Estimates from Sensor Tower, a market intelligence firm, showed Kimi had more than 930,000 downloads during the week after K3's July release, an increase of 200% from the week before. In the U.S. it was downloaded around 86,000 times, a 387% jump. K3 proved so popular that Moonshot had to temporarily suspend new subscriptions after overwhelming demand pushed its capacity close to its limits. Still, while Chinese models are becoming serious competitors to the U.S., they lag American AI leaders across their overall, full-range capabilities, said Anastasios Angelopoulos, co-founder and CEO of Arena, a platform for evaluating AI systems. And U.S. AI firms are also looking for more options to reduce pricing, such as with cheaper alternatives of competitive models, added Yasir Atalan, with the Center for Strategic and International Studies. US policy could boost Chinese AI models Most Chinese AI models are open-source -- meaning anyone can examine and build on them -- at a time when frontier models from U.S. companies like Anthropic and OpenAI are closed-source. Chinese model vendors are expected to leverage open-source software to promote their global usage and adoption, said Lian Jye Su, of the technology research and advisory group Omdia. As Chinese open-source models are increasingly almost as good as closed systems made by leading U.S. companies, said Mozilla's Krikorian, "the open frontier is becoming increasingly Chinese-built." While the U.S. weighs restricting Chinese AI models, a group of American tech firms including Microsoft, Meta and Nvidia signed an open letter published Friday backing "open" AI models. Other U.S. tech policies also at times give China an upper hand, some experts say. China's Z.ai released its GLM-5.2 model in mid-June, not long after the Trump administration put export controls on Anthropic's Fable and Mythos models, keeping them offline for more than two weeks. "Restricting an American model can immediately create an opening for a Chinese competitor," said Arena's Angelopoulos. Chinese AI models look to expand globally In China, people and businesses have rapidly embraced AI technologies that have prospered with state support. Chinese tech companies like Huawei and Tencent are also embedding AI in devices such as smartphones, AI glasses and humanoid robots. Now China is increasingly looking to dominate AI abroad, too. At a flagship technology summit in Shanghai, Chinese President Xi Jinping championed open-source AI models and the importance of promoting greater global equity, while pledging Chinese involvement in raising AI capabilities especially in developing nations. As is true for many other industries in China, intense competition at home is driving companies to expand globally. Leading Chinese AI startups are now raising more funding to support that effort, including through public share offerings. Both China and the U.S. will want to "encourage widespread adoption of their AI ecosystems, while safeguarding technologies that could materially strengthen strategic rivals," said Chelsey Tam, with investment research firm Morningstar. "The competition is no longer simply the United States against China; the Chinese labs are also putting a lot of pressure on one another," Angelopoulos said. But as is the case in the U.S., huge investments in AI have also raised concerns over sustainability of these Chinese startups. Z.ai, for example, reported its revenue surged 132% to 724 million yuan ($107 million) last year, but net loss jumped 60% to 4.7 billion yuan ($694 million). For now, Chinese AI models will likely continue to march ahead including in the U.S. "I would highly recommend anyone doing any serious AI load to at least evaluate it," said Krikorian. ___ O'Brien reported from Providence, Rhode Island. AP Business Writer Kelvin Chan in Toronto contributed to this report.

AI could end up too cheap to control

Eric Levitz is a senior correspondent at Vox. He covers a wide range of political and policy issues with a special focus on questions that internally divide the American left and right. Before coming to Vox in 2024, he wrote a column on politics and economics for New York Magazine. The AI industry's investors and critics don't agree on much. But many in each camp share at least one basic conviction: America's top labs are about to make a killing. Capital markets have signaled their faith in Anthropic and OpenAI's impending hyper-profitability, valuing each at nearly $1 trillion. Many of Silicon Valley's progressive adversaries also expect the labs to grow filthy rich but fear the implications, warning that AI-induced automation could transfer vast sums of money from ordinary workers to a handful of giant tech companies. Sen. Bernie Sanders's call for nationalizing the top AI labs rests partly on that concern. But recent advances in Chinese AI call all of this into question. Over the past two months, Chinese companies have released three AI models that are nearly as powerful as America's frontier systems -- and radically less expensive. In June, Beijing's Z.ai debuted a model that performed nearly as well as Claude and ChatGPT's second-tier systems on independent benchmarks. Weeks later, another Chinese firm, Moonshot, unveiled "Kimi K3," a model that allegedly outperforms all of its American rivals except for the very latest versions of Claude and ChatGPT. Finally, just days ago, Alibaba launched a preview of Qwen3.8 Max, which purportedly outclasses even OpenAI's most advanced systems, while trailing only Claude's Fable in its capabilities. (Disclosure: Vox Media is one of several publishers that have signed partnership agreements with OpenAI. Our reporting remains editorially independent.) These developments don't merely threaten America's AI giants with stiffer competition in the race for superintelligence. Rather, they raise a more harrowing prospect: that the AI race's ultimate rewards will be far smaller than anticipated. In a world where new advances can regularly be leapfrogged by cheaper upstarts, hoarding the technology -- and its profits -- will be harder for any one company to do. In other words, building a machine God might not be as lucrative as it's cracked up to be. AI, it turns out, may "want to be free." To see how China's new models threaten Anthropic's profit expectations, we must first examine why those expectations have been so high. This is not entirely self-evident. After all, AI labs aren't much like the hyper-profitable tech giants of the 2010s. Facebook and Airbrb were relatively capital-light businesses with ultra-low marginal costs (adding a profile to Facebook or listing to Airbnb costs the companies virtually nothing). And once each gained a foothold in their respective markets, network effects enabled them to retain formidable positions without needing to constantly upgrade their products. Building a state-of-the-art AI company is a much more involved -- and astronomically more expensive -- endeavor. To get to the frontier, Anthropic and OpenAI have sunk (at least) tens of billions into semiconductors, data centers, power plants, and other capital investments. Staying at the cutting-edge, meanwhile, compels them to perpetually churn out evermore costly models. To put a new Claude model through its initial training -- in which it spends months digesting the internet and sussing out statistical patterns within its text -- can now cost hundreds of millions of dollars. And such foundational computation is only the beginning. A truly superlative model requires several additional months of fine-tuning. Armies of contracted experts -- such as computer scientists, physicians, and mathematicians -- tutor the models, grading their answers and guiding them towards better ones. Then the AI systems complete millions of rounds of practice, in which they learn through trial and error how to solve countless problems. This arduous process, known as "post-training," compounds the costs of a single model's development. All of which raises the question: Why would investors expect businesses with a cost-structure this challenging to be not merely profitable, but massively so? There are (at least) two answers. The first (and most obvious) is that the market for superintelligent machines is liable to be vast. Frontier AI systems promise to reduce costs and improve performance in myriad white-collar sectors. And Anthropic's soaring revenues indicate that firms do, in fact, find Claude useful. A company like AirBnB has earned billions by revolutionizing a single industry; imagine then what a technology that remade virtually all industries might be worth. Of course, plenty of technologies are valuable but not massively profitable to produce. After all, in well-functioning markets, competition should eventually erode individual firms' margins, even if the underlying technology continues generating huge value. But this is where the second answer comes in: Frontier labs' immense costs are a burden, but they're also a safeguard against competition -- or, in industry parlance, a "moat." Startups may be able to afford to build or acquire more rudimentary models, many of which are "open source." But, the thinking goes, they won't be able to deliver Claude Fable-level performance without raising giant amounts of capital. And what investors will be willing to pour hundreds of billions into an AI pipsqueak that's light-years behind Google, Anthropic, and OpenAI? Alas, the Chinese AI labs' rapid progress -- and the way it was achieved -- suggest that Anthropic's moat may be shallower than previously thought. The existence of powerful, Chinese AI systems is neither new nor surprising. Xi Jinping's government has made vying for global AI dominance a key economic goal. And China's DeepSeek, which also has stunned US companies with its lower-cost competitive models, surpassed ChatGPT as the most-downloaded free iPhone app more than a year ago. The latest models, however, have dramatically narrowed the gap in capabilities between frontier American systems and their Chinese rivals. Just as critically, they've done so in a manner that other, relatively underfunded AI upstarts might be able to emulate. Alibaba and Moonshot needed to invest massive resources to train their base models. But they allegedly found a low-cost way to refine those models into near-frontier systems: Just ask Claude. Or, more specifically: Engage Claude in 16 million conversations, using 24,000 fake accounts. In each of those exchanges, ask the model to not only answer countless difficult questions but also, walk you through its reasoning, step by step. Then take all of this data and feed it into your own model as study material, training it to respond to the world's most challenging queries as Claude would. Through this process -- known as "distillation" -- an AI lab can replicate virtually all of a frontier model's capacities, without sinking vast sums into human experts and post-training computing runs. China's AI labs have not admitted to using distillation. But OpenAI and Anthropic both reportedly uncovered Chinese distillation attempts earlier this year. And some of the new models appear to display tell-tale signs of distillation in conversations with ordinary users; Kimi K3 has routinely identified itself as "Claude." Chinese AI companies are hardly alone in using distillation to catch up with frontier labs. Earlier this year, Elon Musk admitted in court that xAI enhanced Grok's capabilities by running distillation techniques on Claude and ChatGPT. Nonetheless, China's latest models appear to demonstrate that distillation can help take a second-tier model to the frontier's threshold. America's frontier labs have tried to defend themselves against such imitators. But this is technically difficult when distillers can assemble massive networks of bots, each asking an inconspicuous number of questions. And legally, it is difficult for America's AI giants to argue that distillers are stealing their intellectual property. After all, in a sense, China's copycats are merely doing to Anthropic and OpenAI what those companies did to journalists, coders, lawyers and other specialists: Feeding their public-facing outputs into a model, which then replicates their capabilities by discerning underlying patterns within the text. The new Chinese models would have caused Silicon Valley enough headaches, if they merely provided stiffer competition, while demonstrating the power of distillation. What makes Kimi K3 and Qwen3.8 Max especially threatening to the American AI giants' profitmaking potential, however, is that they are officially open source -- meaning that the models' parameters can be downloaded for free. (Alibaba and Moonshot have not yet released these parameters, but they say they will shortly.) In other words, any company or hobbyist with enough computing power will soon be able to run a near-frontier Chinese model on their own hardware, modify that model to better serve a specialized purpose, and then sell access to their new version -- without paying Alibiba a single yuan. For many of Anthropic and OpenAI's potential customers, that proposition may be hard to turn down. Most businesses don't need the world's smartest AI, just one competent at their enterprise's core tasks -- compiling legal research, answering IT queries, writing working code, etc. A model that produces outputs 90 percent as good as Claude's -- at roughly one-sixth of the cost -- will sound pretty good to many corporations. Further, open source models aren't just cheaper than frontier systems, but potentially more secure. If you run an AI on your firm's own servers, then you don't need to entrust sensitive data to Anthropic, Google, or OpenAI. All this had led much of corporate America to embrace open-source models, even before the latest versions narrowed the capabilities gap. In a Linux Foundation survey, 63 percent of organizations reported using open-source AI systems. And increasingly, those models are Chinese. According to Sequoia Capital, one of Silicon Valley's premier venture capitalist firms, a majority of American AI startups now use open-source Chinese systems. As Kimi and Qwen grow more capable, their market-share is likely to grow, at American AI giants' expense. All this said, it is still entirely possible that OpenAI and Anthropic will justify their colossal valuations. In many highly competitive economic domains, having access to the world's very best AI model will remain highly valuable. And America's frontier labs still outperform all their peers. But it's increasingly plausible that selling state-of-the-art AI systems will prove to be a low-margin undertaking. In a world of ubiquitous, near-frontier open source models, the AI sector's big winners probably won't be its top labs, but rather, its chipmakers and cloud computing providers. For ordinary people, a future where superintelligence is dirt cheap -- and rival AI companies are constantly rising and falling, rather than consolidating into mega-corporations -- would look somewhat different than the cyberpunk dystopia that the left's been dreading. And not entirely in a good way. For one thing, in that reality, mitigating AI's biggest risks would be immensely difficult. Having a handful of firms monopolize control over frontier AI systems is bad in many respects. But it does make those models easier to regulate, as the Trump administration's decision to temporarily block Claude's Fable in the name of cybersecurity demonstrated. By contrast, if recipes for ultra-powerful AI models are published all over the internet -- and anyone with modest technical skills can modify them at will -- then systems willing to help their users hack government bureaucracies or engineer bio-weapons are liable to proliferate. From another angle, however, the "AI becomes almost free" scenario may look like capitalism at its finest: Retrospectively, such a development would mean that a small number of extremely rich people bankrolled the creation of an immensely useful technology, under the expectation of massive profits, only to see competition erode their returns -- and disperse that tech's benefits across a wider group of businesses and consumers. Granted, in the case of AI, this process might also generate a super-virus that kills us all. But hey, no system is perfect.

China's Moonshot, Z.AI, and DeepSeek are challenging U.S. AI labs -- and beating them on cost | Fortune

In mid-July, online sleuths who obsessively track the internet for signs of new AI models started to whisper about the latest offering from Moonshot AI, the Beijing-based artificial intelligence startup behind the Kimi large language model: Something big was coming. Moonshot AI's latest model didn't just live up to the hype -- it raised the stakes. On July 16, Moonshot debuted Kimi K3, the largest open-source model ever released, and promised it could perform close to the level of Anthropic's Fable 5 -- perhaps the most powerful publicly available model on the market today -- at a fraction of the cost. Moonshot's official benchmarks consistently rank K3 among the top three AI models; one independent benchmark from Arena.AI even pegged K3 as the best model currently available, ahead of Anthropic. The launch was a triumph for the lab and its founder Yang Zhilin, a 34-year-old Tsinghua and Carnegie Mellon alumnus -- and Pink Floyd fan -- who based his startup's Chinese name on his favorite album, The Dark Side of the Moon. But the release also rattled investors because it challenged the idea that U.S. firms could maintain their sizable lead in the global AI race simply by outspending Chinese rivals on computing power. Markets sold off in Asia. The chip-focused Philadelphia Semiconductor Index fell 1.6%. Nvidia lost almost $600 billion in value and briefly lost its spot as the world's most valuable company to Apple. Many observers, including Anthropic CEO Dario Amodei, hadn't expected a Chinese AI lab to release a model that could approach the U.S.'s best offerings for at least another six months. Tesla CEO Elon Musk suggested it might happen by the first quarter of next year. Kimi K3 blew up the timeline: Chinese AI is now cheap enough and good enough that even U.S. startups and Fortune 500 companies are quietly plugging the models into their operations to rein in spiraling AI budgets. It's a remarkable closing of the gap that China achieved despite Washington imposing export controls to choke off China's access to the world's most advanced chips. "The AI ecosystem in China is probably much better than people thought," says Paul Triolo, a partner at DGA-Albright Stonebridge Group. Chinese developers have been working from a severe disadvantage for years. The U.S. began to cut China off from top-tier AI processors, like those made by Nvidia, in 2022. By limiting the sale of the advanced chips used to train and run AI models -- and the tools that could be used to make them -- Washington planned to kneecap China's tech sector and preserve the U.S.'s AI lead. DeepSeek, a Hangzhou-based lab attached to a Chinese hedge fund, punched a hole in that strategy in early 2025. It shocked the AI world by debuting its V3 and R1 models that matched the performance of its U.S. counterparts. DeepSeek claimed to have trained the models with a tiny budget by engineering efficiency with smart programming and math tricks. The models proved that Chinese AI developers could keep innovating, even with second-tier hardware. DeepSeek was an early winner in China's AI race, but rapid-fire model releases kept ushering in new favorites. In June, AI startup Z.ai stole the spotlight when it released its GLM-5.2 model, which proved particularly strong at coding and creative design. Its newly listed stock had been on a tear, up over 1,100% through mid-July. At times, it broke 1 trillion Hong Kong dollars ($127.6 billion) in market capitalization, valuing the company -- which had just $106 million in revenue last year -- the same as BYD and Starbucks. But Moonshot's Kimi K3 launch sent shares tumbling 40% in two days. Even China's consumer-internet giants are getting into the "frontier" race. Meituan's LongCat-2.0 model encompassed as much data as DeepSeek's V4 and performed at levels that matched OpenAI and Anthropic releases from February. More important, Meituan, best known as a food-delivery platform, claimed it trained the system entirely on Chinese-made processors rather than U.S. chips. "The idea that Meituan could train a 1.6 trillion-parameter model on domestic hardware would have been inconceivable in October 2022," says Triolo, referring to the month when the U.S. launched its AI export controls. AI users globally are adapting to the new reality in which Chinese AI models are competitive with U.S. models on capability and far superior on price. Chinese models now dominate much of the activity on OpenRouter, a popular marketplace where developers can access different models and providers through a single interface. At one point in mid-July, six of the top 10 models -- and all of the top five -- came from Chinese companies: Tencent, Xiaomi, DeepSeek, MiniMax, Moonshot, and Z.ai. The rankings capture usage in developer circles, but adoption is showing up inside mainstream companies, too. Last year, Airbnb CEO Brian Chesky said his company was using Alibaba's Qwen for customer service. Cursor, the AI coding startup, has said that Moonshot AI's Kimi provided the foundation for Composer 2, its coding model. In a June social media post, Coinbase CEO Brian Armstrong explained how the crypto platform had halved its AI spending by pushing more of its employees to use Kimi and Z.ai's GLM models. One million tokens of output -- roughly 750,000 words -- cost $50 when using Anthropic's Fable model. The same million output tokens from DeepSeek-V4-Pro cost about $0.87, while Z.ai GLM-5.2 cost $4.40. Kimi K3 is relatively pricey by Chinese standards at $15. DoorDash is pushing coding tasks to Chinese models, with chief technology officer Andy Fang saying the company delegates "lower-level work" to Kimi, leading to "better quality [at] cheaper cost." Power costs less in China than in many areas of the U.S., in part because China has invested in power generation and transmission, making it easier to add data center capacity. New data centers in the U.S., meanwhile, often face political resistance owing to perceived strain on grids and water use. Chinese AI companies are also willing to sacrifice profit margins in a bid to capture market share and establish their models as a de facto standard. Ironically, the U.S. export controls designed to restrain China's AI sector may have pushed prices lower, too. Without access to the most powerful AI processors, Chinese labs are forced to squeeze more performance out of less capable hardware. "Labs are so compute-constrained, capital-constrained, and talent-constrained that a lot of them are being cautious in how they use their resources," says Grace Shao, an AI analyst and author of the AI Proem newsletter. Recent Chinese AI models are also now compatible with cheaper, locally made processors. "For the money [a Chinese AI company would] spend on an Nvidia chip, they can buy 10 local chips from Huawei or other local chipmakers," says George Chen, a partner at the Asia Group. Perhaps most important, Chinese firms have embraced the open-source movement, launching their models for free. Almost all Chinese companies release their models under permissive licenses, letting users download and fine-tune models at no cost and run them on local hardware, even in the U.S. The only costs that matter, in that case, are "GPUs and energy," says Ameya Kanitkar, cofounder of Larridin, an AI measurement platform. When Anthropic sells its models, it factors in the additional cost of its R&D, he says. Geopolitics can still muck things up. The U.S. is getting concerned about American companies using Chinese AI models, for security reasons and because of fears that Chinese developers might continue to undercut U.S. developers on price. Congress is probing U.S. companies like Airbnb and Cursor on their use of Chinese AI models. (In a statement, Airbnb said it used only "a limited number" of Chinese models, which were all open-source and run through "approved U.S.-based service providers." Cursor did not return Fortune's request for comment.) Lawmakers are also exploring whether the U.S. should be doing more to support its own open-source models. 57% Chinese AI models' share of tokens used by U.S. firms on OpenRouter in one week in July. Chinese developers, meanwhile, are turning that anxiety to their advantage. Z.ai announced its GLM-5.2 model just days after U.S. officials briefly cut off access to Anthropic's Fable and Mythos models for some users outside the U.S. and for foreign nationals. "Frontier intelligence should not belong to only a few people, nor be subject to withdrawal by a handful of rules at any moment," Z.ai wrote in an accompanying social media post. The startup is courting governments that want sovereign AI or models they can run on domestic hardware, with local control over data and upgrades. Demand has only grown as U.S. policy gets more protectionist and less predictable. "The decision to restrict the latest models really backfired," Chen says. "If you think about Singapore or India, there's growing uncertainty around U.S. AI policy. One day, you're told you can use the latest model; the next day, no foreign citizens can use it. How can any country deal with that kind of uncertainty?" Beijing, for its part, sees a soft-power benefit to the world using its AI. At an AI conference in July, President Xi Jinping pledged to "uphold openness and win-win cooperation" when it comes to the technology. AI, "should not be a solo performance by any one country," he said, "but a symphony of global cooperation."

Cheaper, open and intelligent: Chinese AI models gain ground, as they make inroads in the US

HONG KONG (AP) -- China has a hot new product in the United States: artificial intelligence. San Francisco-based Raffi Krikorian, the chief technology officer at Mozilla, which runs the Firefox browser, switched to Chinese AI startup Moonshot's Kimi K3 for many of his day-to-day activities within days of the new, powerful model's launch more than a week ago. "It just seems snappier," he said of K3, comparing it to the acclaimed and more expensive Claude Fable chatbot from San Francisco company Anthropic. Earlier, he had been using another high-performing Chinese model, Z.ai's GLM-5.2, for routine tasks such as managing his calendar, documents and email. Krikorian is one of a growing number of Americans adopting Chinese AI systems, which are gaining ground around the world because they are more affordable and increasingly efficient. U.S. companies like cryptocurrency exchange Coinbase said they're switching to Chinese AI models to help trim costs. Their popularity has frustrated some U.S. tech giants, but short of an outright ban, they're likely to continue to appeal to independent software developers in the U.S. and elsewhere. American-led restrictions block China from accessing some of the world's most advanced technologies including cutting-edge AI chips and U.S. Treasury Secretary Scott Bessent has warned more sanctions could be coming to protect American intellectual property. U.S. President Donald Trump's administration on Wednesday accused Moonshot of using "covert" but not necessarily illegal methods to build K3 off the back of Anthropic's Fable. Some U.S. politicians and AI companies including Anthropic have accused Chinese startups of illicit "distillation" of their models to extract their technologies, a claim that Beijing rejects as "groundless." China's AI models are cheaper and 'good enough' The rapid rise of cutting-edge Chinese AI models this year shows the China-U.S. race in AI has moved on from early last year, when Chinese startup DeepSeek shook up the U.S. technology industry and put China on the map with an AI model that performed on par with those in the U.S. but much more cheaply. The latest AI models launched by startups Z.ai, or Zhipu, in June, and Moonshot in July are nearly as intelligent as the frontier models from OpenAI and Anthropic, experts say. Also in July, China's Alibaba previewed its Qwen3.8 Max AI model. In April, DeepSeek rolled out previews of its latest V4 model, challenging OpenAI's GPT and Google's Gemini. Those constant advances are bearing fruit. Curt Meinhold, a Greensboro, North Carolina-based technology executive, said he increasingly prefers to use DeepSeek for tasks such as finding business leads or ways to generate sales. "At the end of the day, most of us, the vast majority of us, 90 plus percent, don't need (Anthropic's) Mythos or Fable," said Meinhold, who founded digital legacy platform LilyList. "Like, we just don't need it, we need something good enough." Chinese AI models are reaching a "critical stage" for wide adoption, U.S. investment bank Goldman Sachs wrote in a July research report, especially when a surge of global "agentic" AI usage, which requires the AI models to autonomously conduct multistep, complex tasks, is driving higher demand for cost effective AI models. AI prices are calculated per million tokens for both input and output, so the use of AI "agents" is dramatically compounding cost differences, according to analysts including Alex Colville of the Australian Strategic Policy Institute. "The Chinese models, you know, I find them to be pretty close on code and research," said Meinhold. "If I can pay a handful of cents per million output tokens versus 30 bucks or 40 bucks or 50 bucks, then it's good enough." Moonshot's Kimi K3 leads, but Chinese AI models have limits Based on data over the past month, the top five most popular models on OpenRouter, a platform that tracks data across AI models, were Chinese. Estimates from Sensor Tower, a market intelligence firm, showed Kimi had more than 930,000 downloads during the week after K3's July release, an increase of 200% from the week before. In the U.S. it was downloaded around 86,000 times, a 387% jump. K3 proved so popular that Moonshot had to temporarily suspend new subscriptions after overwhelming demand pushed its capacity close to its limits. Still, while Chinese models are becoming serious competitors to the U.S., they lag American AI leaders across their overall, full-range capabilities, said Anastasios Angelopoulos, co-founder and CEO of Arena, a platform for evaluating AI systems. And U.S. AI firms are also looking for more options to reduce pricing, such as with cheaper alternatives of competitive models, added Yasir Atalan, with the Center for Strategic and International Studies. US policy could boost Chinese AI models Most Chinese AI models are open-source -- meaning anyone can examine and build on them -- at a time when frontier models from U.S. companies like Anthropic and OpenAI are closed-source. Chinese model vendors are expected to leverage open-source software to promote their global usage and adoption, said Lian Jye Su, of the technology research and advisory group Omdia. As Chinese open-source models are increasingly almost as good as closed systems made by leading U.S. companies, said Mozilla's Krikorian, "the open frontier is becoming increasingly Chinese-built." While the U.S. weighs restricting Chinese AI models, a group of American tech firms including Microsoft, Meta and Nvidia signed an open letter published Friday backing "open" AI models. Other U.S. tech policies also at times give China an upper hand, some experts say. China's Z.ai released its GLM-5.2 model in mid-June, not long after the Trump administration put export controls on Anthropic's Fable and Mythos models, keeping them offline for more than two weeks. "Restricting an American model can immediately create an opening for a Chinese competitor," said Arena's Angelopoulos. Chinese AI models look to expand globally In China, people and businesses have rapidly embraced AI technologies that have prospered with state support. Chinese tech companies like Huawei and Tencent are also embedding AI in devices such as smartphones, AI glasses and humanoid robots. Now China is increasingly looking to dominate AI abroad, too. At a flagship technology summit in Shanghai, Chinese President Xi Jinping championed open-source AI models and the importance of promoting greater global equity, while pledging Chinese involvement in raising AI capabilities especially in developing nations. As is true for many other industries in China, intense competition at home is driving companies to expand globally. Leading Chinese AI startups are now raising more funding to support that effort, including through public share offerings. Both China and the U.S. will want to "encourage widespread adoption of their AI ecosystems, while safeguarding technologies that could materially strengthen strategic rivals," said Chelsey Tam, with investment research firm Morningstar. "The competition is no longer simply the United States against China; the Chinese labs are also putting a lot of pressure on one another," Angelopoulos said. But as is the case in the U.S., huge investments in AI have also raised concerns over sustainability of these Chinese startups. Z.ai, for example, reported its revenue surged 132% to 724 million yuan ($107 million) last year, but net loss jumped 60% to 4.7 billion yuan ($694 million). For now, Chinese AI models will likely continue to march ahead including in the U.S. "I would highly recommend anyone doing any serious AI load to at least evaluate it," said Krikorian. ___ O'Brien reported from Providence, Rhode Island. AP Business Writer Kelvin Chan in Toronto contributed to this report.

DeepSeek's V4 Flash Guts OpenAI's Price War Within Hours, Matching Opus-Grade Output At $0.14 Per Million Tokens, As Moonshot Scales With 20,000 New NVIDIA GPUs

We told our readers not to discount China's uncanny ability to trounce the West in a price war, and that's exactly what its AI labs have done: merely hours after OpenAI fired its first volley, DeepSeek's revamped V4 Flash model now offers unprecedented economy and performance, while the progenitor of the Kimi K3 model, Moonshot, has just secured a sizable cluster of NVIDIA GPUs to crank up its training workloads. OpenAI now faces a dual counterpunch in its still-fledgling price war from China's DeepSeek and Moonshot As we detailed recently, OpenAI has just launched a literal price war by discounting its GPT-5.6 Luna by as much as 80 percent, with input tokens now priced at just $0.2 per 1 million from their earlier perch at $1, and output tokens priced at just $1.20 per 1 million vs. the earlier price of $6. Of course, OpenAI claimed at the time that it was able to implement this steep discount after extracting additional architectural efficiencies from its models. Even so, most interpreted the move as its opening gambit in a price war aimed at China's AI labs. Just hours later, however, DeepSeek has launched a refreshed version of its latest Flash-class model, dubbed the V4 Flash 0731. Critically, the model has just 284 billion parameters and yet offers a performance that is similar to Anthropic's Opus 4.8, which is widely believed to span multi-trillion parameters! That's not all. In what goes right to the heart of OpenAI's price war, DeepSeek has priced the V4 Flash 0731 at just $0.14 per 1 million tokens of input, and $0.28 per 1 million tokens of output, eviscerating any comparative price advantage that OpenAI tried to garner with its discounting move. Meanwhile, Bloomberg has reported separately that Moonshot has secured a compute cluster consisting of 20,000 H200 NVIDIA GPUs from Alibaba, significantly upping its training capacity. Of course, Moonshot jumped right into the ongoing multi-dimensional struggle between China and the US recently, when Anthropic and some members of the Trump administration accused Moonshot of distilling its Kimi K3 model from Anthropic's Fable. The US has long suspected that Chinese engineers often take their model-laden hard drives to US-friendly countries to imbue their models with frontier-level capabilities by using reinforcement learning techniques on cutting-edge NVIDIA GPUs. Apparently, as per the allegations leveled by a Trump administration official recently, Moonshot not only furtively owns some NVIDIA GB300 servers, but was also able to access additional GB300s via Thailand. Follow Wccftech on Google to get more of our news coverage in your feeds.

Steve Eisman Says He'd Be 'Petrified' Running OpenAI or Anthropic as Cheap Chinese Models Threaten Price

Steve Eisman says he wouldn't want to run OpenAI or Anthropic right now, and with both labs preparing to go public, the timing stings. On the latest Real Eisman Playbook, the investor made famous by "The Big Short" pressed tech bulls Dan Ives and D.A. Davidson's Gil Luria on whether AI model companies have any moat at all. Eisman argued that Chinese lab Moonshot AI is undercutting US frontier-model developers by enough to threaten a price war. "If I was the head of Anthropic or OpenAI, I'd be petrified," Eisman said. "That spells to me price war." The threat is not limited to lower API prices. Moonshot released Kimi K3's full model weights, allowing developers to operate and customize it without remaining tied to the company's platform. That portability could put further pressure on the economics of closed models. The Timing Problem Anthropic filed confidentially with the SEC on June 1, OpenAI a week later, and both are now preparing to test public-market investors on exactly the moat Eisman doubts exists. Polymarket traders put Anthropic's chances of going public by year-end near 69%, compared with 19% for OpenAI, as of Wednesday. The bulls' answer is the revenue curve. Luria noted OpenAI and Anthropic have a combined run rate above $75 billion, likely topping $100 billion once Gemini and others are counted. Two years ago that number was zero, he added. That figure, he argued, is the point: real customers paying real money for AI, whatever the moat debate says. It is also the exact story both labs will need to sell public investors, that the growth outruns the commoditization. Ives, fresh off launching Yorkville Ives, a self-described modern merchant bank, framed the pricing scare differently. We're in year three of an eight to 10 year buildout, he said, one that will bring "gut check moments" three to four times a year. Where the Trade Hides The market, in his view, cannot have it both ways. And if Eisman's price war arrives, the models get cheaper, but somebody still has to make the memory. Image: Shutterstock Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

Deutsche Bank Says OpenAI Is About to Rewrite the Economics of the Boom

Deutsche Bank Research Institute analyst Adrian Cox believes artificial intelligence may now be approaching its own version of that old format war. Takeaways * Open-weight models are rapidly narrowing the performance gap with proprietary systems, weakening the assumption that frontier intelligence will remain permanently scarce. * The immediate valuation threat is not that AI disappears, but that model economics become commoditized before hyperscalers earn an adequate return on their enormous capital expenditure. * Cheaper models may pressure proprietary pricing while simultaneously accelerating enterprise adoption, customization and application-layer demand. * The AI boom may survive, but the profit pool is likely to migrate away from raw model ownership toward data, distribution, product design, security and applications. OpenAI Is About to Rewrite the Economics of the Boom Every great technology boom eventually discovers that the most important battle is not always fought over who builds the best machine. Sometimes it is fought over who controls the format through which the machine reaches the world. Betamax was widely considered the superior video technology, but VHS won because it was cheaper, more widely licensed and supported by a larger ecosystem. The market did not crown the most elegant format. It crowned the one that became good enough, inexpensive enough, and ubiquitous enough to become the industry standard. Deutsche Bank Research Institute analyst Adrian Cox believes artificial intelligence may now be approaching its own version of that old format war. The latest Chinese open-weight model triggered what Cox describes as a second DeepSeek moment, reviving investor concerns that cheaper, lighter alternatives are rapidly closing in on the leading proprietary systems. The Magnificent Seven fell 1.8% during the latest selloff, while the Philadelphia Semiconductor Index declined 1.6% and finished the week down more than 10%. Those moves landed at an uncomfortable moment for the AI capital-spending story. Google, Microsoft, Amazon Web Services, Meta and Oracle are expected to spend roughly $700 billion building AI capacity this year, around 70% more than last year. That leaves the market staring at a widening gap between the money being spent and the price at which intelligence may ultimately be sold. The hyperscalers are still building cathedrals, but OpenAI keeps discovering that a large part of the congregation may be perfectly happy inside a cheaper hall. The first pillar supporting today's AI valuations is scarcity. Investors have treated frontier intelligence as something only a small number of companies, armed with the best researchers, the largest data centres and the most advanced chips, could realistically produce. That scarcity justified premium pricing, enormous infrastructure budgets and the assumption that the leading model developers would retain control over the economic tollbooths. Open-weight models are beginning to loosen those foundations. Cox shows that the capability gap between closed and open-weight models has narrowed substantially. DeepSeek R1 did not need to defeat every proprietary model across every benchmark to disturb the market. It merely needed to demonstrate that a cheaper Chinese system built with fewer resources could perform well enough to challenge the idea that only the largest American platforms could produce useful frontier intelligence. That is the real significance of the open-model threat. It does not need to make proprietary systems obsolete. It only needs to commoditise a sufficiently large portion of their output. Once capable models can be downloaded, adapted and hosted by multiple providers, foundational intelligence starts to resemble software infrastructure rather than a permanently scarce luxury product. The raw model may still matter, but it becomes harder to treat the model weights themselves as an impregnable commercial moat. Proprietary models remain the polished, plug-and-play option. Their developers control the underlying technology, pricing, updates and safety restrictions. Open-weight systems offer something different: greater control, lower long-run costs and the ability to fine-tune models for a particular language, industry or task. The trade-off is that downloadable does not mean free. Companies running open models still need to pay for compute, electricity, engineering, cybersecurity and maintenance. They also assume more responsibility for testing, reliability and guardrails. Yet for large enterprises consuming an ever-growing river of tokens, the economics are becoming difficult to ignore. This is where Deutsche Bank's report moves from a technology debate into a valuation debate. The chart comparing model intelligence with the cost per task suggests that US and proprietary systems are no longer standing miles ahead of Chinese and open-weight competitors. The performance hierarchy has not disappeared, but the gap relative to cost is narrowing fast enough to challenge four assumptions embedded in the boom. The first is scarcity. The second is pricing power. AI agents and employees using AI tools are consuming vastly more tokens, the small units of information processed by the models. As enterprise trials scale and introductory subsidies fade, companies are becoming increasingly sensitive to the cost of that consumption. A model that is slightly less capable but dramatically cheaper may be more attractive for the majority of routine commercial tasks. The third is capital intensity. If an open model can generate comparable results with less training and inference horsepower, the market is entitled to ask whether every dollar earmarked for chips, networking equipment, electricity and data centres will produce an adequate return. This does not mean the great AI capital-spending race suddenly stops. It means the burden of proof is becoming heavier. Every additional dollar of infrastructure investment must now compete against a falling cost of intelligence. The fourth assumption is the durability of the American technological moat. Washington has framed AI leadership as a strategic race and has attempted to restrict Chinese access to advanced chips and related technology. Yet Chinese developers have responded by building models designed to do more with less, then releasing them widely enough to encourage adoption. The semiconductor restrictions may have forced China to become more efficient, and efficiency is now being exported back into the global market as a competitive weapon. The shift is already visible in usage. Deutsche Bank's OpenRouter chart shows Chinese models overtaking US models this year in the number of tokens processed through the developer platform. The technological moat is therefore no longer measured only by which company produces the highest benchmark score inside a laboratory. It is increasingly measured by developer adoption, inference cost, ecosystem depth, customisation and the amount of real work flowing through the model. China does not necessarily need to collect the highest margin on the model itself. By releasing capable open systems, Chinese developers can weaken the profit pool of proprietary American competitors while expanding demand for their own cloud infrastructure, applications and domestic technology ecosystem. This explains why a company would give away something that required vast amounts of capital and research talent to create. Cox describes it as losing the battle to win the war. A developer can distribute the model weights freely and still charge for hosted access, faster inference, enterprise support, security, fine-tuning and guaranteed service levels. A widely adopted model can attract developers, research and complementary tools, allowing its interface to become an industry standard. More importantly, giving away the model can commoditise a competitor's core product while increasing demand for something else the provider owns. A cloud company benefits when cheaper models increase demand for compute. An application company benefits when its most important input becomes less expensive. A consumer platform can offer intelligence cheaply while monetising advertising, recommendations, data and distribution. The open-model threat is therefore not equally bearish across the AI chain. It may be negative for businesses whose valuation rests heavily on proprietary-model scarcity. It is more ambiguous for the infrastructure providers financing the capex race. And it could be decisively positive for software, applications and enterprises able to consume cheaper intelligence at scale. This is where the Jevons Paradox becomes central to the market debate. More efficient steam engines did not reduce the consumption of coal. They made steam power cheaper and more useful, causing total demand to rise. AI may follow the same path. Lower inference costs could reduce the economics of each individual task while triggering an explosion in the number of tasks performed. The market is currently looking at cheaper models and asking whether fewer chips and data centres will be required. The more consequential question is whether dramatically cheaper intelligence creates so many new commercial uses that total compute demand continues to expand. The most likely outcome is not a clean winner-takes-all victory for either side. A handful of expensive proprietary frontier models will probably coexist with an enormous universe of smaller, specialised and locally deployed open systems. Proprietary developers will set the benchmark for reliability, security and integrated products. Open models will keep pressure on prices and force the incumbents to remain flexible. The competitive moat will gradually migrate away from the raw weights and toward proprietary data, product design, memory, security, distribution and the ability to embed intelligence inside products people are willing to pay for. The boom may survive. But the toll booths are moving. This is my interpretation of Adrian Cox's Deutsche Bank Research Institute report