Nvidia’s Allies and Rivals Mapped: 6 of 8 AI Firms Mix Chips

Of the eight AI developers we map, six have Nvidia GPUs and at least one other chip family on record in their compute, by our count from filings and company releases. xAI, now part of SpaceX, is the only one with Nvidia alone on record, and Google is the one whose flagship model card says it was trained on Google’s own Tensor Processing Units.

Nvidia’s money points in several directions at once. Its 13F filing for 30 June 2026 lists $63.4 billion of listed stakes, about 80% of it in Intel and SpaceX, by our arithmetic, and Anthropic’s S-1/A has Nvidia buying $10.0 billion of a planned $26.0 billion private placement, level with Google. This map sets out who trains on Nvidia, who builds rival or in-house chips, who releases open models, and what analysts said on 12 November, after Nvidia’s board accepted its chief executive’s resignation, as we reported.

Last updated: 12 November 2026. An evergreen map; we update it as companies file or announce new chip, model or investment terms.

Business Pill · Embedded Dependency

An evergreen Business Engineer explainer on embedded dependency: what is inside what you buy. It teaches the general idea only and does not discuss Nvidia or its customers.

How We Built the Map

Each company sits in a column by the chip families on record in its own filings or releases, or in a partner’s: Nvidia only, Nvidia plus one other family, Nvidia plus two or three others, or its own chips for its flagship model. The column is our classification of what is on record, not a measure of how much work runs on each chip.

The height is the number of public model repositories each company’s organisation lists on Hugging Face, from the public API at 15:12 UTC on 12 November, by our count. It is a rough gauge of how much a company releases openly, not of model quality or downloads. A teal dot marks a Nvidia investment that was announced or is on file.

Who Trains on Nvidia, and Who Also Buys Elsewhere

xAI (SpaceX). Anthropic’s S-1/A says it agreed with SpaceX.AI LLC, “formerly known as X.AI LLC” and a wholly owned subsidiary of Space Exploration Technologies Corp., “to access compute powered by NVIDIA GPUs”, for up to $120.5 billion over terms extending into 2029.

OpenAI. Nvidia and OpenAI announced a letter of intent on 22 September 2025 to deploy “at least 10 gigawatts of NVIDIA systems”. AMD’s 6 October 2025 release says OpenAI will “deploy 6 gigawatts of AMD GPUs”, and OpenAI and Broadcom announced “a collaboration for 10 gigawatts of custom AI accelerators” that OpenAI designs. Through Amazon, OpenAI is accessing “AWS compute comprising hundreds of thousands of state-of-the-art NVIDIA GPUs”.

Nvidia’s 10-Q adds guarantees capped at $105 billion on behalf of an affiliate of OpenAI for leases of about 4.25 gigawatts in Ohio. In exchange, the filing says, “the site will exclusively host NVIDIA AI infrastructure, subject to limited exceptions”.

Anthropic. The S-1/A says: “We train and serve Claude natively across four major chip families (AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs)”. By our earlier count of its named contracts, Google TPUs were 43.9% of $643.8 billion and Nvidia GPUs 39.0%. It expects AMD spending “to exceed $20.0 billion” if it procures the full initial MI450 tranche, as in our AMD bring-up piece.

Meta. Nvidia announced in February a partnership covering “millions of NVIDIA Blackwell and Rubin GPUs”, and in October said Meta intends to expand its deployment, including Vera Rubin. Meta’s MTIA chips are “developed in close partnership with Broadcom”, and AMD’s Meta warrant follows the same six-gigawatt structure as OpenAI’s, as we set out.

Microsoft. Microsoft’s Maia 200 is “Fabricated on TSMC’s cutting-edge 3-nanometer process” and “will serve multiple models, including the latest GPT-5.2 models from OpenAI”. In November 2025 Microsoft said Anthropic is scaling Claude “on Microsoft Azure, powered by NVIDIA”.

Amazon. Anthropic’s S-1/A says its collaboration with Amazon gives it access to “custom-designed Trainium chips”, while AWS clusters “both GB200s and GB300s” for OpenAI, Amazon says.

Google. The Gemini 3 Pro model card says: “Gemini 3 Pro was trained using Google’s Tensor Processing Units (TPUs)”. Anthropic reaches the same chips through “partnerships with Google and Broadcom”, its filing says.

DeepSeek. Its V3 technical report says the model “requires only 2.788M H800 GPU hours for its full training”, on Nvidia chips. STAR Market Daily reported on 3 November that DeepSeek plans to pretrain V5 on Huawei’s Ascend chips; neither company has confirmed it, as we reported.

Who Builds Chips That Compete

By our count, six custom programs sit beside Nvidia at the companies above: Google’s TPU, Amazon’s Trainium, Microsoft’s Maia 200, Meta’s MTIA, OpenAI’s accelerator with Broadcom and Huawei’s Ascend. Their latest stated dates differ.

Meta’s March blog says it has “deployed hundreds of thousands of MTIA chips in production”, that “MTIA 450 is scheduled for mass deployment in early 2027” and that “MTIA 500 is scheduled for mass deployment in 2027”. Microsoft says Maia 200 is deployed in its US Central region near Des Moines. Broadcom’s racks for OpenAI are “targeted to start in the second half of 2026, to complete by end of 2029”, OpenAI said.

Huawei said in September 2025 that the Ascend 950DT will be available in the fourth quarter of 2026. Tesla’s 10-Q says it continues to expand Cortex, “our onsite training clusters at Gigafactory Texas”, and is expanding into semiconductor fabrication. AMD is the merchant rival: its warrants let OpenAI and Meta earn up to 320 million AMD shares if they buy six gigawatts each.

Where Nvidia’s Money Sits

Nvidia’s 13F for 30 June 2026 lists eight holdings worth $63.4 billion: Intel at $30.0 billion, SpaceX at $21.0 billion, CoreWeave at $4.7 billion, Coherent at $3.1 billion, Nokia and Synopsys at $2.2 billion each, Nebius at $0.3 billion and a small stake in Generate Biomedicines. Intel and SpaceX together are about 80% of that value, by our arithmetic.

The 10-Q gives the wider total: “equity investments of $99 billion and equity investment commitments of $25 billion as of July 26, 2026”. Private stakes made up $47.9 billion of it, as our stakes piece sets out.

Announced investments sit in several AI developers that also buy rival chips. Nvidia said in September 2025 that it “intends to invest up to $100 billion in OpenAI progressively as each gigawatt is deployed”. Microsoft’s blog said “NVIDIA and Microsoft are committing to invest up to $10 billion and up to $5 billion respectively in Anthropic”. In October Nvidia said it “will invest $3 billion in Meta’s equity offering”.

On the sales side, the 10-Q says that in its second quarter one direct customer was 16% of revenue, and that three direct customers made up 16%, 15% and 13% of first-half revenue, 44% together by our arithmetic. Nvidia does not name them. It says it estimates that “one AI research and deployment company contributed a meaningful amount of our revenue by purchasing cloud services from our customers”.

Anthropic’s Placement: Nvidia and Its Rivals on One List

The S-1/A lists seven buyers for the $26.0 billion placement at the $74.00 midpoint: Google and Nvidia at about $10.0 billion each, AMD at about $2.5 billion, Broadcom, Micron and Microsoft at about $1.0 billion each and TSMC at about $500 million. The five chipmakers among them put in $15.0 billion, about 58%, by our arithmetic.

The S-1/A says the placement is contingent upon the completion of the offering. The Microsoft blog’s up-to-$10 billion figure dates from November 2025; the excerpts we read do not say whether the placement counts toward it, so we do not add the two.

Open vs Closed Models

By this measure Nvidia itself releases openly: its Hugging Face account lists 1,040 public model repositories. It signed the July open-weights letter with 235 organisations by our count, among them OpenAI, Google, Meta, Microsoft, Amazon and AMD, as our open-model piece sets out.

Among the developers, Meta’s two organisations list 2,436 public models, Google’s 1,136, Microsoft’s 554, DeepSeek’s 106, Amazon’s 47, OpenAI’s 39 and xAI’s 2, by our count. The Qwen organisation lists 469 and Mistral’s 83. OpenAI released gpt-oss as “two state-of-the-art open-weight language models” under the Apache 2.0 license in August 2025. Meta’s 10-Q says it makes AI technology available “via open source, commercial, and non-commercial license agreements”.

Anthropic’s Hugging Face organisation listed no public model repositories when we read the API at 15:12 UTC. Its S-1/A defines model weights as “the primary intellectual asset of a frontier intelligence developer” and says open-weight models “could exert pricing pressure on our offerings”.

What Analysts Said on 12 November

Goldman Sachs analyst James Schneider called the CEO change a “meaningful, but measured negative for the stock”, citing Nvidia’s “long-dated backlog of 12-24 months”, Investing.com reported. Oppenheimer, which downgraded Nvidia on 12 November, said perceived beneficiaries of any Nvidia dislocation include AMD, Marvell and Broadcom, the outlet reported.

William Blair upgraded AMD the same day; it “sees ASIC competition as a risk to AMD’s Instinct product, particularly at Meta” and OpenAI, according to Investing.com, as in our William Blair piece. Bloomberg Intelligence ranked slower new agreements and tighter China scrutiny above leadership uncertainty as risks for Nvidia, Bloomberg News reported. Nvidia’s release, filed with its 8-K, says the board expects “the company’s operations, customer commitments and product roadmap to continue without interruption”. The market moves are in our after-hours report, our opening-trade piece and the dates ahead in our calendar.

The Structural Read

As we read Anthropic’s S-1/A, the placement shows the overlap in one document: Nvidia and Google each buy about $10.0 billion, with AMD, Broadcom, Micron, Microsoft and TSMC alongside.

As we read the Hugging Face counts, openness does not follow chip choice: Meta, which names Nvidia, AMD and MTIA chips, lists 2,436 public models by our count, while Anthropic, with four chip families on record, listed none when we checked.

Three Implications

FOR NVIDIA Its 10-Q says three direct customers made up 16%, 15% and 13% of first-half revenue, and that one AI research and deployment company contributed a meaningful amount of revenue through cloud services.

FOR THE CHIP RIVALS AMD’s warrants let OpenAI and Meta earn up to 320 million AMD shares if each buys six gigawatts; Meta says MTIA 450 is scheduled for mass deployment in early 2027.

FOR OPEN MODELS OpenAI released gpt-oss under the Apache 2.0 license; Anthropic’s S-1/A says open-weight models could exert pricing pressure on its offerings.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Supercycle.

The framework maps the chip layer as: “Three competitive categories within one layer: generalist substrate (NVIDIA), hyperscaler custom (TPU, Trainium, MTIA), merchant custom challengers (Cerebras, Groq, SambaNova).”

As we read it, the map adds a money layer to those categories: Nvidia’s investments sit in developers that also run hyperscaler custom chips, while AMD, a second GPU supplier, ties its own equity to OpenAI and Meta through warrants.

What Is Not Established

The columns record chip families on record, not shares of compute; most companies do not publish how their workloads split across chips. Nvidia does not name its largest customers. The OpenAI figure is from a 2025 letter of intent, the DeepSeek V5 report is unconfirmed, and the Anthropic placement depends on the offering closing. Hugging Face counts include fine-tunes and older models. Candidates for Nvidia’s top job, as named by outlets, are in our succession piece.

The Bottom Line

By our count, six of the eight AI developers we map run on Nvidia GPUs and at least one other chip family, and three of them also have a Nvidia investment announced or on file. Nvidia’s listed stakes are concentrated in Intel and SpaceX, about 80% of $63.4 billion, while Anthropic’s placement puts Nvidia, Google, AMD, Broadcom and TSMC on the same list of buyers.