OpenAI Moves Deeper Into AI Chips — A New Battle With Nvidia?

Nvidia and OpenAI’s relationship has become one of the strangest in tech: Nvidia is simultaneously OpenAI’s biggest supplier, a major investor in the company, and — through the industry’s growing push toward custom silicon — a company OpenAI is quietly working to become less dependent on.

Nvidia’s Massive Financial Backstop

Nvidia said this year it’s providing up to $105 billion for a giant OpenAI data center in Ohio, on top of a $30 billion direct investment in OpenAI made in February. According to CNBC’s reporting on Nvidia’s strategy shift, this reflects a deliberate move by Nvidia to use its capital, not just its chip lead, to keep the AI boom’s biggest compute consumers tied to its ecosystem — Nvidia held $30.2 billion in AI-ecosystem equity investments as of its most recent quarter, more than double the year before.

Why This Isn’t Purely Generous

OpenAI relies heavily on Nvidia’s Vera Rubin training systems, its most advanced hardware, to build its models. By financing OpenAI’s infrastructure directly, Nvidia effectively guarantees continued, massive chip orders — the backstop and the sales pipeline are the same move.

But OpenAI Is Also Building Its Own Path

Even as it leans on Nvidia hardware, OpenAI has been reported to be exploring its own custom AI chip designs — following the same playbook Microsoft, Google, Amazon, and Meta have all pursued with in-house silicon like Maia, TPUs, and Trainium. The logic is the same across the industry: reduce per-token inference costs and reliance on any single supplier, even one that’s also an investor. This mirrors the broader dynamic we’ve covered in Google’s own AI infrastructure investment in Finland, where hyperscalers are increasingly building parallel infrastructure tracks rather than depending entirely on any one partner.

Why Nvidia Isn’t Worried, Yet

Despite every major AI company pursuing custom chips to reduce Nvidia dependency, Nvidia’s revenue keeps growing anyway — demand for AI compute is expanding faster than any single company’s custom-chip capacity can scale to meet it. Nvidia still captures an estimated 90% of AI accelerator spending industry-wide, a dominant position custom silicon has chipped at only modestly so far.

What This Means for the Broader AI Supply Chain

Whether or not OpenAI’s own chip ambitions materialize at scale, the relationship illustrates how tangled AI’s supply chain has become — suppliers investing in customers, customers building alternatives to suppliers, all while overall demand keeps growing for everyone. Businesses relying on [CLIENT LINK PLACEHOLDER] for AI infrastructure planning are watching this dynamic closely, since chip supply and pricing over the next few years will depend heavily on how it resolves.

Frequently Asked Questions

Does OpenAI currently use its own chips instead of Nvidia’s?

No — OpenAI still relies primarily on Nvidia hardware for training and inference today; any custom silicon effort remains at an earlier development stage relative to its current production use of Nvidia GPUs.

Could OpenAI’s chip ambitions actually hurt Nvidia’s business?

Not meaningfully in the near term — overall AI compute demand is growing faster than any single company’s custom chip capacity can absorb, which is part of why Nvidia’s revenue keeps climbing despite the industry-wide push toward alternatives.

The Bottom Line

Nvidia and OpenAI’s relationship captures the AI industry’s central tension in one partnership: everyone wants to reduce dependency on the dominant supplier, while that same supplier keeps financing the very growth that deepens the dependency.