Invest1 publisher2 min readPublished
Huang rejects the antitrust waiver Amodei says a frontier slowdown would require
Three days after Anthropic's chief executive proposed that leading labs coordinate to slow model releases, Nvidia's chief executive told CNBC that the legal cover such coordination needs is completely unnecessary.
The Investor · Invest desk

What happened
- Anthropic CEO Dario Amodei published an essay on Saturday urging leading AI labs to coordinate to "pace the frontier", slowing advanced model development to leave more time for testing and safeguards.
- Amodei said that coordination could require government mediation or waivers of antitrust restrictions, and set out three steps beginning with third-party safety evaluators embedded at each lab.
- Nvidia CEO Jensen Huang told CNBC's Jim Cramer on Tuesday that new antitrust laws to let labs do their engineering properly before release are "just completely unnecessary".
- Anthropic did not respond to CNBC's request for comment on Huang's remarks.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint A slowdown now has two routes and no third: each lab holding releases on its own judgement, or Congress creating legal cover before rivals sit down together to set a schedule.
- exposure Nvidia's growth depends on decisions its own customers would make, since CNBC reported that efforts to slow the development of more capable models may pose a risk to that growth.
- decision Anyone drafting a rate limit for frontier developers has to decide whether it binds the chipmaker too, because Nvidia releases models of its own as well as selling the hardware.
- contradiction Both men say companies should not ship unsafe models, so the disagreement narrows to whether existing product law already covers what pacing would require.
The waiver is the only part of the plan that needs someone outside the labs to say yes. Embedding third-party safety evaluators at each lab, the first step, is something a lab can do on its own [5]. The third step, "limits on the rate of unchecked AI progress," is an agreement between rivals about competing products [5]. The Sherman Antitrust Act generally prohibits agreements among rivals that unreasonably restrain competition, and agreements to restrict output sit among the forms of coordination that can be illegal outright [7]. Amodei wrote that "Some forms of coordination that would be impactful for pacing are legally challenging, and will require government support" [6].
Huang answered three days later [20]. His argument was that the statute book is already full: "We have plenty of laws. We have plenty of regulations that govern the reliability and the functionality of products," he said on "Mad Money" [10][2]. He agreed AI companies need to take safety seriously and said they are able to solve those problems themselves [19]. "Safety is an engineering problem. Testing is an engineering problem," he said [11]. He also pushed back on existential warnings: "We're not going to die in 2030" [18].
Now take the proposal from the seller's side. An agreement to ship fewer models later is, in demand terms, an agreement among the buyers of Nvidia's chips to defer purchases, and Anthropic and OpenAI are both important Nvidia customers [13]. The AI boom has made Nvidia the world's largest chipmaker and most valuable company [15]. Self-interest is the easy reading of Huang's answer and probably the correct one as far as it goes; the more interesting version is Nemotron. Nvidia develops its own family of open-weight models [14], so a rate limit written for model developers would reach the supplier alongside its customers [21].
Two ways that reading looks wrong. Huang's own position gets to Amodei's outcome without a statute if each lab holds releases on its own judgement: "if they're not ready to be released, just hold on to it and keep testing it and keep engineering until it's ready," he said [12]. Pacing could also migrate into evaluator access and standards work, where the first step already lives [5], and the antitrust question would never be tested. I would still expect no coordinated slowdown to be agreed while the company selling the hardware opposes the exemption that would make one lawful [1]. Huang told Cramer that "companies should innovate as fast as possible, but they should never innovate so fast as to release unsafe products" [22].
What to watch
- A legislative vehicle in Washington that would carry an antitrust exemption for AI safety coordination.
- A frontier lab publicly holding a model release for further testing without any legal shield in place.
- Nvidia naming safety rules it would accept, or extending its position to its own Nemotron releases.