Anthropic is hedging its bets by using multiple chips to handle its artificial intelligence work, a strategy that could ultimately undermine Nvidia’s dominance.
In an S-1 filing posted by Anthropic on Monday, the artificial intelligence lab said it manages computing “as a single, fungible pool from multiple strategic partners,” with software directing its work to different wafers based on where they are most efficient. The filing was released on Monday as the company prepares to go public.
Anthropic says it uses hardware from four chip families: Amazon Trainium, Google TPU, Nvidia GPU and AMD GPU.
That flexibility, which Anthropic said “took years and significant investment to build,” could over time make major AI companies less reliant on Nvidia and give them more negotiating leverage, analysts said.
“The real risk may be Nvidia’s pricing power because customers who can move work across four platforms negotiate very differently than customers who can’t,” said Patrick Moorhead, founder and CEO of Moor Insights & Strategy.
IDC analyst Ashish Nadkarni added that the multi-chip approach also makes Anthropic less vulnerable to fluctuations in Nvidia’s chip supply.
Breaking Nvidia’s control won’t be easy
Moorhead said that while Anthropic’s chip portfolio is not new, the company is now using it as a core strength in its pitch to investors.
This threat has its limits. Moorhead said switching between wafers remains difficult, especially when it comes to training AI models rather than running them, something only the largest AI companies have the resources to do.
Anthropic also continues to spend billions on Nvidia computing, Moorhead said.
Gil Luria, director of technology research at DA Davidson, said that while other large companies such as OpenAI may take a multi-die approach, for smaller companies it may not be worth the effort and investment.
Brad Gastwirth, director of global research and market intelligence at Circular Technology, said Anthropic’s strategy does not pose a significant near-term threat. Having more chip options won’t eliminate Nvidia’s advantages in performance, networking and software, he said.
If AI computing becomes more commoditized, Nvidia will face broader risks.
Citadel chief executive Ken Griffin warned that if supply exceeds demand, it could drive down rental prices.
“GPUs, like airlines, have very low marginal operating costs,” he told Bloomberg last week. “Once there is excess capacity, prices will plummet.”
Nvidia did not immediately respond to Business Insider’s request for comment.
Any tips? Contact this reporter via email: gweiss@businessinsider.com Or send a message to @geoffweiss.25. Use personal email addresses, non-work WiFi networks, and non-work devices; here are our guidelines for sharing information safely.