Top White House advisers this week expressed alarm that China's DeepSeek may have benefited from a method that allegedly piggybacks off the advances of U.S. rivals called "distillation."
The technique, which involves one AI system learning from another AI system, may be difficult to stop, according to executive and investor sources in Silicon Valley.
DeepSeek this month rocked the technology sector with a new AI model that appeared to rival the capabilities of U.S. giants like OpenAI but at a much lower cost. And the China-based company gave away the code for free.
Some technologists believe that DeepSeek's model may have learned from U.S. models to make some of its gains. The distillation technique involves having an older, more established, and powerful AI model evaluate the quality of the answers coming out of a newer model, effectively transferring the older model's learnings.
That means the newer model can reap the benefits of the massive investments of time and computing power that went into building the initial model without the associated costs.
This form of distillation, which is different from how most academic researchers previously used the word, is a common technique used in the AI field. However, it is a violation of the terms of service of some prominent models put out by U.S. tech companies in recent years, including OpenAI.
The ChatGPT maker said that it knows of groups in China actively working to replicate U.S. AI models via distillation and is reviewing whether or not DeepSeek may have distilled its models inappropriately, a spokesperson told Reuters.
Naveen Rao, vice president of AI at San Francisco-based Databricks, which does not use the technique when terms of service prohibit it, said that learning from rivals is "par for the course" in the AI industry. Rao likened this to how automakers will buy and then examine one another's engines.
"To be completely fair, this happens in every scenario. Competition is a real thing, and when it's extractable information, you're going to extract it and try to get a win," Rao said. "We all try to be good citizens, but we're all competing at the same time."
Howard Lutnick, President Donald Trump's nominee for Secretary of Commerce who would oversee future export controls on AI technology, told the U.S. Senate during a confirmation hearing on Wednesday that it appeared DeepSeek had misappropriated U.S. AI technology and vowed to impose restrictions.
"I do not believe that DeepSeek was done all above board. That's nonsense," Lutnick said. "I'm going to be rigorous in our pursuit of restrictions and enforcing those restrictions to keep us in the lead."
David Sacks, the White House's AI and crypto czar, also raised concerns about DeepSeek distillation in a Fox News interview on Tuesday.
DeepSeek did not immediately answer a request for comment on the allegations.
OpenAI added it will work with the U.S. government to protect U.S. technology, though it did not detail how.
"As the leading builder of AI, we engage in countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models," the company said in a statement.
The most recent round of concern in Washington about China's use of U.S. products to advance its tech sector is similar to previous concerns about the semiconductor industry, where the U.S. has imposed restrictions on what chips and manufacturing tools can be shipped to China and is examining restricting work on certain open technologies.
NEEDLE IN A HAYSTACK
Technologists said blocking distillation may be harder than it looks.
One of DeepSeek's innovations was showing that a relatively small number of data samples - fewer than one million - from a larger, more capable model could drastically improve the capabilities of a smaller model.
When popular products like ChatGPT have hundreds of millions of users, such small amounts of traffic could be hard to detect - and some models, such as Meta Platforms' (META.O) Llama and French startup Mistral's offerings, can be downloaded freely and used in private data centers, meaning violations of their terms of service may be hard to spot.
"It's impossible to stop model distillation when you have open-source models like Mistral and Llama. They are available to everybody. They can also find OpenAI's model somewhere through customers," said Umesh Padval, managing director at Thomvest Ventures.
The license for Meta's Llama model requires those using it for distillation to disclose that practice, a Meta spokesperson told Reuters.
DeepSeek in a paper did disclose using Llama for some distilled versions of the models it released this month, but did not address whether it had ever used Meta's model earlier in the process. The Meta spokesperson declined to say whether the company believed DeepSeek had violated its terms of service.
One source familiar with the thinking at a major AI lab said the only way to stop firms like DeepSeek from distilling U.S. models would be stringent know-your-customer requirements similar to how financial companies identify with whom they do business.
But nothing like that is set in stone, the source said. The administration of former President Joe Biden had put forth such requirements, which President Donald Trump may not embrace.
The White House did not immediately respond to a request for comment.
Jonathan Ross, chief executive of Groq, an AI computing company that hosts AI models in its cloud, has taken the step of blocking all Chinese IP addresses from accessing its cloud to block Chinese firms from allegedly piggybacking off the AI models it hosts.
"That's not sufficient, because people can find ways to get around it," Ross said. "We have ideas that would allow us to prevent that, and it's going to be a cat and mouse game ... I don't know what the solution is. If anyone comes up with it, let us know, and we'll implement it."
The artificial intelligence boom is a two-edged sword for $290 billion ASML (ASML.AS). Shares in the Dutch maker of lithograph machines used to make chips soared by 8% on Wednesday thanks to booming demand from semiconductor manufacturers such as TSMC (2330.TW). The arrival of a powerful large language model developed by Chinese startup DeepSeek, however, shows that AI also brings risks.
ASML CEO Christophe Fouquet is starting 2025 on a positive note. The Dutch company on Wednesday reported 7 billion euros of new orders in the fourth quarter of 2024, far higher than the 4 billion euros analysts expected, according to Visible Alpha data. Rising demand for chips used in data centres has temporarily eased concerns that stalling smartphone sales or the troubles of key customer Intel (INTC.O) would hurt ASML's growth. Fouquet now expects demand from data centres and other AI usages to represent more than 40% of $1 trillion semiconductor sales in 2030, when he expects to churn out as much as 60 billion euros of annual revenue, up from 30 billion euros to 35 billion euros this year.
But Wednesday’s news may not reflect Fouquet’s future challenges. Chinese AI startup DeepSeek claims to have built an AI model comparable to OpenAI’s at a fraction of the cost. That may mean that training AI models will require fewer, and less sophisticated chips. It might therefore discourage ASML's chip customers from ordering more of ASML’s most cutting-edge, so called extreme ultraviolet machines, which can cost $400 million apiece. Analysts expect these products will account for 52% of group's sales by 2030, according to Visible Alpha data.
China’s growing clout in AI may also increase political tensions. ASML made some 41% of its revenue in the Middle Kingdom in 2024. The company already expects 2025's China sales to decline to 20% of the group's total, partly due to the impact of export curbs imposed by former U. S. President Joe Biden. But China’s evident strength in AI may encourage current President Donald Trump to impose even tougher curbs.
Cheaper AI may have some benefits too. Fouquet argued on Wednesday that it may encourage more companies to use the technology, which should boost demand for chips, and hence ASML’s kit. But the rapid rate of technological change makes the outlook particularly foggy. After taking account of Wednesday’s jump, ASML trades at 29 times forward earnings, still below the 46 times multiple achieved in March last year, per LSEG data. Even with AI booming, getting back to that sky-high valuations looks increasingly tricky for Fouquet.
Days after Chinese upstart DeepSeek revealed a breakthrough in cheap AI computing that shook the U.S. technology industry, the chief executives of Microsoft and Meta defended massive spending that they said was key to staying competitive in the new field.
DeepSeek's quick progress has stirred doubts about the lead America has in AI with models that it claims can match or even outperform Western rivals at a fraction of the cost, but the U.S. executives said on Wednesday that building huge computer networks was necessary to serve growing corporate needs.
"Investing 'very heavily' in capital expenditure and infrastructure is going to be a strategic advantage over time," Meta (META.O) CEO Mark Zuckerberg said on a post-earnings call.
Satya Nadella, CEO of Microsoft (MSFT.O), said the spending was needed to overcome the capacity constraints that have hampered the technology giant's ability to capitalize on AI.
"As AI becomes more efficient and accessible, we will see exponentially more demand," he said on a call with analysts.
Microsoft has earmarked $80 billion for AI in its current fiscal year, while Meta has pledged as much as $65 billion towards the technology.
That is a far cry from the roughly $6 million DeepSeek said it has spent to develop its AI model. U.S. tech executives and Wall Street analysts say that reflects the amount spent on computing power, rather than all development costs.
Still, some investors seem to be losing patience with the hefty spending and lack of big payoffs.
Shares of Microsoft — widely seen as a front-runner in the AI race because of its tie to industry leader OpenAI - were down 5% in extended trading after the company said that growth in its Azure cloud business in the current quarter would fall short of estimates.
"We really want to start to see a clear road map to what that monetization model looks like for all of the capital that's been invested," said Brian Mulberry, portfolio manager at Zacks Investment Management, which holds shares in Microsoft.
Meta, meanwhile, sent mixed signals about how its bets on AI-powered tools were paying off with a strong fourth quarter but a lackluster sales forecast for the current period.
"With these huge expenses, they need to turn the spigot on in terms of revenue generated, but I think this week was a wake-up call for the U.S.," said Futurum Group analyst Daniel Newman.
"For AI right now, there's too much capital expenditure, not enough consumption."
There are some signs though that executives are moving to change that.
Microsoft CFO Amy Hood said the company's capital spending in the current quarter and the next would remain around the $22.6 billion level seen in the second quarter.
"In fiscal 2026, we expect to continue to invest against strong demand signals. However, the growth rate will be lower than fiscal 2025 (which ends in June)," she said.