Breaking point: many people in Silicon Valley are wary of NVIDIA CEO Jensen Huang.
As detailed in this morning's reporting with Leo and Phoebe, that concern is well-founded. The chipmaker holds substantial capital and enormous influence, and Huang has access to multiple senior government officials, including President Trump.
For large model developers like OpenAI, an even more critical point is this: NVIDIA (ASX: NVDA) controls the supply of AI chips needed to develop and run today's most advanced AI models. AI companies already struggle to secure chips, so it makes sense that they would deliberately avoid any move that could anger Huang.
Given this situation, OpenAI has chosen not to take a public position in the current public debate over chip sales to China. It worries about alienating NVIDIA and Huang; NVIDIA earns considerable revenue from Chinese companies and itself wants to maintain access to the Chinese market. (U.S. regulations prohibit NVIDIA from directly selling its most advanced chips to China, but Chinese companies such as ByteDance and DeepSeek can still obtain chips through overseas channels or smuggling routes.)
But how much do AI companies really need to fear? After all, whether buying directly or indirectly, OpenAI and Anthropic are both top customers of NVIDIA. The chipmaker needs them no less than these AI companies depend on NVIDIA hardware.
Anthropic has publicly supported stricter restrictions on sales of NVIDIA chips to Chinese companies, but that does not appear to have hurt its ability to obtain NVIDIA chips. NVIDIA even said late last year that it would invest up to $10 billion in the developer of the Claude large model; recently there have also been reports that NVIDIA intends to participate as a cornerstone investor in Anthropic's IPO later this month.
I mentioned yesterday on the video program that Anthropic has indeed had difficulty securing sufficient computing resources. But that is more because the company was overly conservative in past computing contracts, not because there is a problem in its relationship with NVIDIA. The uncertainty over whether it can obtain enough servers to support business operations is one of the major risk factors Anthropic listed in its IPO prospectus.
Beyond that, if NVIDIA cut the chip quota allocated to OpenAI, the company that ultimately suffers might be NVIDIA itself. OpenAI has been pushing to diversify its computing sources, bringing in chips from AMD, Google, and Cerebras, and also working with NVIDIA competitor Broadcom to develop its own Jalapeño chip for inference. If NVIDIA stopped supplying chips, it would only push OpenAI to accelerate that diversification and reduce its dependence on NVIDIA.
Once OpenAI's business ran into trouble, NVIDIA would also suffer huge losses: NVIDIA is guaranteeing leases for OpenAI's large data center campus in Ohio, involving more than $100 billion. That gives OpenAI stronger bargaining leverage than its silent posture suggests.
But it also explains why OpenAI chooses to stay quiet. Unlike Anthropic, which now has positive cash flow, with cash income exceeding its burn rate, OpenAI relies not only on NVIDIA's chips but also on NVIDIA's balance sheet for financial support.
Comments