SemiAnalysis names Nvidia's seven biggest GPU customers — and says Google's TPU is 'just as good'
SemiAnalysis says Google, Microsoft, Meta, Amazon, SpaceX, Anthropic and OpenAI deploy most of Nvidia's GPUs — and that all but SpaceX are now building their own chips. These are its views, not Nvidia's disclosures.
The News
In a post on X on Saturday, November 14, SemiAnalysis — a research firm that tracks the chip industry — named the seven companies it sees deploying most of Nvidia's graphics chips: Google, Microsoft, Meta, Amazon, SpaceX, Anthropic and OpenAI. The post said all of them except SpaceX are making their own chips. The claims were relayed by FourWeekMBA. BitcoinVersus analysed the same post and cautioned that it offers 'no independently verified list of seven or consistent definition of a completed chip' — the seven names are SemiAnalysis's assessment, not an audited ranking.
The same post said SemiAnalysis's benchmarking found Google's TPU 'just as good as NVIDIA'. A TPU is Google's own custom chip for AI work, and the claim is striking because Nvidia's graphics chips have long been treated as the benchmark everything else is measured against. SemiAnalysis also argued that without Nvidia's backing of CoreWeave and other neoclouds — newer cloud companies built to rent out AI computing — 'those industries just would not exist, and the hyperscalers would have full control.'
These are SemiAnalysis's views, not Nvidia's disclosures, and that distinction matters. Nvidia's own quarterly filing names no customers, so there is no official confirmation of who its biggest buyers are, how much each buys, or how its chips benchmark against Google's. Nvidia shares closed Friday at $162.85, down 2.01%; Alphabet closed at $295.77, down 0.98%; Microsoft at $523.88, up 0.17%; Meta at $641.69, up 2.90%; and Amazon at $266.66, down 0.29%. The custom-chip theme overlaps with the Wall Street Journal's report on Nvidia financing its own customers — the day's lead story — but the named customer list and the TPU benchmark claim are fresh and distinct.
Nvidia's seven biggest customers are also its future rivals — and one of them, SemiAnalysis says, already has a chip just as good.
Why It Matters
Why the list matters is concentration. If seven companies really do deploy most of Nvidia's GPUs, then Nvidia's fortunes rest on a small number of very large buyers — and every one of those buyers except SpaceX is now building its own chips. That is the classic shape of a supplier problem: today's best customers are tomorrow's competitors, and each custom chip they ship is one fewer Nvidia chip they need to buy.
Why the TPU claim matters is the moat. Nvidia's advantage has never been just the silicon — it is the software, the ecosystem and the sheer performance lead. If Google's TPU is genuinely 'just as good' on the workloads that matter, the performance lead narrows to software and habit, which are easier to overcome than a hardware gap. But note the sourcing: this is SemiAnalysis's benchmark, described in a social media post, not a peer-reviewed test or a Google disclosure. The claim deserves attention, not acceptance.
Why the neocloud argument matters is the buffer. Nvidia's financing of CoreWeave and other neoclouds — the subject of the day's lead story — created a class of buyers that is neither hyperscaler nor startup, and SemiAnalysis argues those buyers only exist because Nvidia backed them. Without them, the hyperscalers would have more pricing power over Nvidia, because there would be fewer alternative buyers competing for the chips. That is a coherent argument, and it connects the two big Nvidia stories of the weekend: the financing kept demand broad, and breadth is what keeps Nvidia's prices high.
Analyst's Take
Our call is Mixed. The customer list itself is plausible, and the custom-chip shift is real — the Wall Street Journal reported the same drift in the day's lead story, with the largest cloud companies quietly moving toward their own designs. A supplier whose biggest customers are all building alternatives is a supplier with a narrowing moat, and that is a genuine risk to Nvidia's pricing power over time.
But the TPU claim needs handling with care. 'Just as good as NVIDIA' is one firm's benchmark result, not an independent test, and SemiAnalysis is an interested observer with its own audience and positions. Google's TPU may well be competitive — Google has spent years and billions on it — but a single benchmark assertion is not proof, and the post gives no detail on what was tested or how. That is why we tag Alphabet as bullish in the per-ticker view while keeping the story itself Mixed: a credible 'as good as Nvidia' verdict is good for Google's chip story, but credibility is exactly what still needs establishing.
Reader poll: which chip wins the next two years? Vote: Nvidia GPUs, Google TPUs, AMD, or Custom chips.
Are you bullish or bearish on this story?
By The Numbers
FAQ
Who are the seven companies?
According to SemiAnalysis, the seven companies deploying most of Nvidia's GPUs are Google, Microsoft, Meta, Amazon, SpaceX, Anthropic and OpenAI. All of them except SpaceX are making their own chips, SemiAnalysis says — though the programs are at very different stages: Google and Amazon run custom silicon at scale, while Anthropic is still assembling a chip-design team.
What is a TPU?
A TPU is a Tensor Processing Unit — Google's own custom chip, designed for AI work. SemiAnalysis said its benchmarking found Google's TPU 'just as good as NVIDIA', meaning it performs comparably to Nvidia's graphics chips on the tests SemiAnalysis ran.
Does Nvidia confirm this list?
No. Nvidia's own quarterly filing names no customers, so there is no official confirmation of who its biggest buyers are. This list is SemiAnalysis's assessment, based on its own research and benchmarking — attribute it to SemiAnalysis, not to Nvidia.
What are neoclouds?
Neoclouds are newer cloud companies built specifically to rent out AI computing — CoreWeave is the best-known example. SemiAnalysis argued that without Nvidia's backing of CoreWeave and other neoclouds, 'those industries just would not exist, and the hyperscalers would have full control.'
👁
Source: FourWeekMBA