Nvidia (NASDAQ: NVDA) was already changing hands at roughly 15 times next year’s projected earnings before this week, a figure that represents less than half the stock’s five-year average multiple.
That discount signals what Wall Street has quietly concluded: the 75% gross margins that have defined Nvidia’s dominance over the past several years are not expected to hold indefinitely.
The clearest evidence yet of that pressure arrived this week through Anthropic’s IPO filing, which laid out in precise terms where one of the world’s leading AI labs is directing its compute spending.
Anthropic’s largest commitments run to Google and Amazon, with combined contract value exceeding $230 billion, and crucially, both of those cloud giants build their own silicon alongside renting out Nvidia hardware.
Google supplies its TPU chips and Amazon offers Trainium, and Anthropic runs its Claude models across both architectures, making the frontier lab far less dependent on any single supplier.
The filing went further, disclosing that Anthropic plans to spend an additional $125 billion to $150 billion on Google’s chips later this decade, plus more than $20 billion directed toward AMD.
Against that backdrop, Cerebras (NASDAQ: CBRS) shares jumped 8.44% on Friday after the company announced a new partnership with Microsoft (NASDAQ: MSFT), adding another hyperscaler relationship to deals it already holds with Amazon and Oracle.
Cerebras designs its own processors, and Microsoft is itself developing proprietary silicon, meaning the partnership represents two Nvidia alternatives converging around a single agreement.
The conventional defense of Nvidia’s position has always centered on CUDA, the software ecosystem that binds most AI development to its hardware and makes switching suppliers a costly, time-consuming undertaking.
That barrier may be weakening faster than expected, however, after Anthropic disclosed that porting one of its models to AMD chips, a process that once required six months to a year, was accomplished in roughly three days using its newest AI model to handle the migration work.
Nebius, an AI cloud provider that currently runs its infrastructure on Nvidia hardware, told reporter Deirdre Bosa that the shift will not happen overnight, given the scale of capital commitments and lead times involved in billion-dollar infrastructure decisions.
Yet the company acknowledged it is actively evaluating alternatives, and its chief revenue officer, Mark Burditsky, offered a measured warning about how quickly the landscape could evolve, saying “we’ll know when they start asking.”
The practical reality is that no major customer is abandoning Nvidia’s ecosystem today, and the company’s financial results continue to exceed expectations by a wide margin.
What has changed is the direction of investment flows at the frontier, where the largest AI spenders are systematically funding the infrastructure that would allow them to route around Nvidia when contract renewal cycles arrive.
