Anthropic's filing that Claude is trained and served on four chip families offers U.S. investors a concrete case of AI computing diversification as economist William Lee questions NVIDIA's long-term margin durability.
An amendment to Anthropic's registration statement filed Nov. 10 with the Securities and Exchange Commission lists AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs among the hardware used for Claude. Anthropic says an in-house orchestration layer assigns work among the platforms based on price and performance. The company says developing that capability took years and significant investment, broadens its computing capacity and lowers reliance on any one supplier. It also says the system is difficult to reproduce at scale.
The filing provides an example of a multi-vendor approach, but does not say Anthropic is moving away from NVIDIA. NDTV Profit reported on Nov. 16 that William Lee, chief economist at Global Economic Advisors, said investors are considering whether NVIDIA can preserve its margins over the next three to five years as rival processors become easier to integrate. The outlet also reported that Lee identified Huawei and other chip suppliers as potential beneficiaries.
NVIDIA describes CUDA as a software platform for accelerated computing. Its toolkit combines GPU libraries with a compiler and developer tools for building and operating applications.
NVIDIA reported $96.2 billion in revenue for its second fiscal 2027 quarter, which ended July 26, including $89 billion from its Data Center business. GAAP and non-GAAP gross margins were both 75%. For the third quarter, the company forecast $108 billion in revenue, with a 2% range in either direction, and margins of 74%, plus or minus 0.5 percentage point. Those figures are forecasts, not reported results.
NVIDIA is scheduled to report its third-quarter results Tuesday, Nov. 17, at about 4:20 p.m. ET (1:20 p.m. PT), followed by a conference call at 5 p.m. ET (2 p.m. PT).



