By Hamza Faisal
Nvidia's biggest customers are also its future rivals. That is the blunt picture painted by chip research firm SemiAnalysis, which named the seven companies it says deploy most of Nvidia's graphics chips — and claimed that six of the seven are now building their own AI chips to replace them.
The list, posted on X on Saturday, November 14, names Google, Microsoft, Meta, Amazon, SpaceX, Anthropic and OpenAI as the companies deploying the vast majority of Nvidia's GPUs. Every one of them except SpaceX is developing its own custom silicon, SemiAnalysis says — a shift that puts the world's most valuable chipmaker in the awkward position of arming its own competition.
The claim that drew the most attention was about Google. SemiAnalysis said its benchmarking found Google's TPU — the search giant's custom AI chip — is "just as good as NVIDIA". Google expanded its eighth-generation TPU line-up on November 9 with the TPU 8tx and TPU 8i Max.
Six customers, six different chip programmes
The custom-chip push is real, but the six programmes are at very different stages. Google and Amazon run custom silicon at scale: Amazon says its Trainium3 chip is serving production workloads, and chief executive Andy Jassy said in October that revenue commitments for its Graviton and Trainium chips each grew more than twentyfold year over year, with customers saving over $2 billion using custom silicon in the third quarter.
Microsoft says its Maia 200 chip already runs production services, including Copilot, with Maia 300 due in 2027. Meta reports hundreds of thousands of its MTIA chips deployed for recommendations and generative-AI inference. OpenAI unveiled an inference processor, codenamed Jalapeño, with Broadcom in June, while Anthropic is still assembling a chip-design team, according to a Reuters report from August.
None of this means the seven have stopped buying Nvidia chips — each still runs enormous fleets of its hardware. But every workload that moves in-house is one fewer Nvidia chip to buy, and that is the slow squeeze on its pricing power.
The neocloud argument — and the caveats
SemiAnalysis also argued that Nvidia's backing of CoreWeave and other neoclouds — newer cloud companies built to rent out AI computing — is what keeps the market competitive. Without that backing, it said, "those industries just would not exist, and the hyperscalers would have full control." It echoes a Wall Street Journal investigation published on November 13, which reported that Nvidia offered billions in financial guarantees to AI startups and small cloud providers that could not otherwise afford its chips.
Some caution is warranted. The seven-name list is SemiAnalysis's assessment, not an audited ranking — Nvidia's filings name no customers, so there is no official confirmation of who its biggest buyers are. The TPU claim is likewise one firm's benchmark result in a social media post, not a peer-reviewed test or a Google disclosure.
Why it lands days before Nvidia's earnings
Nvidia reports third-quarter results on Tuesday, November 17 — its first earnings under interim chief executive Colette Kress, who took over after founder Jensen Huang resigned on November 12. Investors will be listening for how the company answers the customer-turned-rival question, and whether demand from the seven names on the list is still growing.
Nvidia shares closed Friday at $162.85, down 2.01 percent, while Alphabet closed at $295.77, Microsoft at $523.88, Meta at $641.69 and Amazon at $266.66. The market is pricing a weaker Nvidia, not a weaker AI trade — but if the biggest buyers keep building their own chips, the chipmaker's grip on the AI boom will be tested as never before.



