Anthropic built a four-chip strategy that could crack Nvidia's grip

Anthropic's initial public offering filing, released Monday, contains a direct challenge to Nvidia's dominance: the AI lab says it manages its computing as "a single, fungible pool from multiple strategic partners," with software that can route work across four chip families - Amazon Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs. That strategy, which Anthropic says "took years and significant investment to build," is being pitched to investors as a core advantage. Analysts say the move could quietly weaken Nvidia's pricing power. "The real exposure could be Nvidia's pricing power," Patrick Moorhead, founder and CEO of Moor Insights & Strategy, told Business Insider, "because a customer that can shift work across four platforms negotiates very differently than one that can't."

The timing is significant. Anthropic is going public at a moment when the AI industry's infrastructure costs are under intense scrutiny, and its S-1 filing is effectively a statement about how it intends to keep those costs in check. By advertising its ability to move workloads across chips from four different vendors, Anthropic is signaling to investors that it does not need to accept whatever price Nvidia sets. Ashish Nadkarni, an analyst at IDC, added that the multichip approach also makes Anthropic less vulnerable to fluctuations in Nvidia's chip supply, giving the company operational resilience that a single-vendor setup would lack. For Nvidia, the filing is a public reminder that its most sophisticated customers are building escape hatches.

But the threat has limits, and analysts were quick to point them out. Switching between chips remains difficult, especially when it comes to training AI models as opposed to running them. Training requires deep optimization for a specific hardware architecture, and moving that work to a different chip can mean weeks or months of engineering. Moorhead noted that only the biggest AI companies have the resources to pull off a true multichip strategy, and even Anthropic continues to spend billions on Nvidia compute. In other words, the four-chip approach is a hedge, not a clean break. Nvidia still benefits from Anthropic's scale, and Anthropic still relies on Nvidia for a substantial portion of its work.

Other analysts agreed that the near-term impact on Nvidia should not be overstated. Gil Luria, head of technology research at D.A. Davidson, said that while other major AI companies like OpenAI likely take a similar multichip approach, it may not be worth the effort and investment for smaller players. Brad Gastwirth, global head of research and market intelligence at Circular Technology, said Anthropic's strategy does not represent a major near-term threat. Having more chip options does not erase Nvidia's advantages in performance, networking, and software, he said, and those advantages are precisely why Nvidia commands premium prices today.

The longer-term risk, however, is more structural. If AI compute becomes more of a commodity, the entire basis of Nvidia's pricing power could erode. Citadel CEO Ken Griffin warned about this dynamic last week, telling Bloomberg that "GPUs, like airlines, have a very low marginal cost to run." He added: "The moment you have overcapacity, pricing goes into freefall." Anthropic's multichip strategy is a step toward making GPUs less unique, and if other large buyers follow the same playbook, the AI hardware market could start to look more like a standard commodity market, where buyers can play vendors against each other.

For executives and boards watching Anthropic's IPO, the filing is a useful window into how the largest AI players are thinking about infrastructure. The strategy that Anthropic is now touting as an investor selling point was not built overnight; it required years of investment and a software layer capable of abstracting away hardware differences. That puts smaller AI companies at a disadvantage, because they lack the scale to justify building such an integrated system. The result could be a widening gap between the handful of AI labs that can flexibly move work across chip architectures and the rest of the field, which remains locked into whatever hardware their cloud provider offers.

For Nvidia, the competitive picture is more nuanced than a simple threat to its market share. Anthropic still spends billions on Nvidia chips, and its training workloads will likely remain on Nvidia hardware for a long time. But the S-1 filing sends a clear message to investors and to Nvidia itself: even a company that is one of Nvidia's largest customers wants options. If Anthropic's multichip strategy proves to be a genuine advantage, other large AI buyers will have more reason to invest in the same flexibility, and Nvidia's negotiating position could slowly weaken. That is the quiet, structural risk hiding inside Anthropic's IPO paperwork, and it is worth watching long after the offering is done. Nvidia did not immediately respond to a request for comment from Business Insider.