Nvidia, Meta and Microsoft have urged Washington not to impose sweeping restrictions on open-weight artificial intelligence models, warning that broad curbs could weaken US technological leadership at a time of intensifying competition with China.
In an open letter to policymakers, the three firms joined 22 other organisations in calling for narrowly targeted legal and commercial measures to tackle abuse of AI systems, rather than blanket rules that would cover technologies used for legitimate research and product development.
The signatories – which include IBM, Palantir, Mistral, Hugging Face, Mozilla, Andreessen Horowitz and the Linux Foundation – argued that open and closed approaches to AI should be treated as complementary, and that both are needed to keep the US at the forefront of the field.
Case for open-weight models
Open-weight models allow companies, academics and public bodies to download the underlying software, adapt it to their own needs and run it on their own servers. According to the letter, that flexibility makes advanced AI easier to tailor to specific tasks, while giving organisations greater control over sensitive data, security and their computing environment.
The coalition said that, alongside proprietary systems, open models play a key role in sharpening competition, reducing deployment costs and giving developers more freedom to inspect, audit or modify the technology underpinning their applications.
The companies insisted that policymakers should not frame the debate as a choice between openness and control, but should instead focus on “targeted measures addressing concrete harms” and on enforcing existing laws against bad actors.
Huang and Musk weigh in
Nvidia chief executive Jensen Huang highlighted the appeal in his first-ever post on X, sharing the letter and defending what he described as a balanced market in which both open and closed development can thrive.
Huang argued that open models can strengthen cybersecurity and safety, support national oversight and help spread AI tools across a wide range of industries.
Elon Musk, whose firm xAI is developing the Grok chatbot as a rival to OpenAI and Anthropic products, backed the message in a reply to Huang’s post. However, xAI itself was not among the 25 organisations listed as signatories in media reports.
China tensions and US enforcement debate
The intervention comes as the Trump administration weighs action against Chinese AI developers accused of building their systems on top of US technology without permission.
US Treasury Secretary Scott Bessent said this week that officials are examining whether Chinese AI models have been trained using outputs from American-built systems without authorisation, and signalled that sanctions or placement on the Entity List could follow if companies are found to have carried out “industrial-scale distillation” that crosses into intellectual-property theft.
Bessent stressed that the administration supports open-source AI, drawing a distinction between lawful development practices and alleged efforts to copy protected US technology.
Distillation, a widely used technique in AI, involves using one model’s outputs to train or refine another. In their letter, Nvidia and the other signatories described distillation as a commonplace method for improving, testing and validating models, and urged US officials not to treat every instance as evidence of theft.
They drew parallels with the history of open-source software, arguing that developers have long improved their products by learning from existing systems and using shared tools. Regulators, they said, should focus enforcement on clear legal violations rather than on the techniques used by legitimate researchers and businesses.
Chinese progress adds urgency
China’s recent advances in AI have added further urgency to the US policy debate. Moonshot AI’s model Kimi K3 has reached first place on the Frontend Code Arena benchmark, putting a Chinese system ahead of several established US models in that specific category.
Former White House AI and crypto adviser David Sacks has warned that such milestones could threaten the US position in the broader AI race.
US officials have separately accused Moonshot of distilling Kimi K3 from Anthropic’s Fable model. According to Reuters, those allegations remain part of an ongoing policy dispute rather than an established legal finding.
Wider concerns over AI and finance
Questions over how open models should be regulated have emerged alongside broader discussions about how financial markets make use of AI.
As crypto.news reported on 24 July, Gate founder and chief executive Dr. Han has advocated using AI tools to gather information and analyse market signals, while insisting that humans should retain final responsibility for trading decisions.
Speaking on the Gatecast podcast, Dr. Han said automated systems could help users navigate a universe of millions of digital assets and tens of thousands of decentralised applications. But he argued that investors must still scrutinise the output of those tools before acting.
“AI + human intelligence” will become a more effective approach in the future, Dr. Han said, casting AI as an assistant rather than a replacement for human judgment. In his view, machines are best suited to processing large volumes of data at speed, while people remain essential for assessing risk and deciding whether to trade.
Financial risks around the rapid build-out of AI infrastructure have also drawn attention. Earlier in July, former Fidelity fund manager George Noble warned that a collapse in the current AI investment boom could inflict damage up to 17 times greater than the dot-com crash, which wiped about $5tn from the Nasdaq.
Noble linked that danger to the vast sums being poured into data centres, chips and related infrastructure, and cautioned that losses could spread beyond the technology sector if expected returns fail to justify the capital committed to AI.
“The fallout from this could really be much more significant,” he said, referring to rising AI capital expenditure.
While Noble’s comments focus on financial exposure rather than open-weight regulation, they touch on the same central policy question facing US officials: how to contain the risks associated with AI without stifling beneficial innovation.
In their letter, Nvidia and its fellow signatories argued that the answer lies in targeted enforcement – pursuing theft and misuse directly while preserving broad access to open AI technologies that, they say, underpin both US competitiveness and global research.
