Musk says SpaceX will build AI infrastructure exclusively on Nvidia, but chip volumes remain undisclosed

wpnews.pro

Elon Musk said on SpaceX’s August 4 earnings call that the company had decided to build exclusively on Nvidia because it considers the Vera Rubin architecture the best available. [1] - SpaceX has not disclosed the number of Nvidia GPUs it plans to buy for its own infrastructure. Its filings instead disclose customer compute agreements covering roughly 325,000 Nvidia GPUs for Anthropic and 110,000 for Google. [2][3] - Nvidia says Rubin-based systems will become available through partners in the second half of 2026. SpaceX’s filing says its orbital AI satellites could begin deployment as early as 2028. [4][5] - Bernstein warned that Nvidia could put SpaceX near the front of the queue for scarce chips, creating a potential supply risk for CoreWeave and Nebius.

[6] SpaceX has committed to building its AI infrastructure exclusively with Nvidia technology, Chief Executive Elon Musk said on the company’s August 4 earnings call, adding a major customer to Nvidia’s rollout of its Vera Rubin platform. [1]

The statement confirms a strategic decision, but it does not establish the commercial terms implied by reports of an Nvidia-SpaceX “deal.” SpaceX has disclosed no purchase volume, delivery timetable, financing from Nvidia or minimum spending commitment tied to the announcement. Its public filings show that SpaceX is already operating a large Nvidia-based compute business, while leaving the company’s own future chip orders unspecified. [2][3]

What SpaceX has actually disclosed #

SpaceX’s filing describes its Colossus and Colossus II facilities as a gigawatt-scale AI training cluster. It says the company entered compute agreements with Anthropic in May 2026 covering approximately 325,000 Nvidia GPUs, along with CPUs, storage and networking. The Anthropic agreements require payments of $1.25 billion per month through May 2029, subject to termination rights after an initial period. [2]

A separate SpaceX free-writing prospectus says a June 5 agreement with Google covers approximately 110,000 Nvidia GPUs and related infrastructure. Google is scheduled to pay $920 million per month from October 2026 through June 2029, with capacity ramping before October. The contract allows Google to terminate if SpaceX fails to provide the committed GPU capacity by September 30, 2026, after a one-month grace period. [3]

Those figures describe capacity SpaceX has promised to customers. They are not evidence of a separate order for 435,000 Nvidia GPUs, nor do they show how many additional chips SpaceX must obtain for Grok, other internal workloads or its planned orbital systems.

Rubin timing is clearer than SpaceX’s order book #

Nvidia introduced Vera Rubin as a rack-scale platform combining Rubin GPUs, Vera CPUs, networking and storage components. The company says Rubin is in full production and that products from cloud partners, including CoreWeave and Nebius, will become available in the second half of 2026. CoreWeave separately said it expected to be among the first cloud providers to deploy Rubin in that period, while Nebius has said Rubin systems are expected from the second half of 2026. [4][7][8]

SpaceX’s own filing gives a longer timetable for its space ambitions. It defines orbital AI compute as satellite constellations functioning as data centers and says deployment could begin as early as 2028. SpaceX’s Starmind website separately says the company plans to produce and deploy thousands of AI satellites starting as soon as late 2027. Neither disclosure provides a satellite-by-satellite chip count or a binding Nvidia delivery schedule. [5][9]

Nvidia has separately announced a Space-1 Vera Rubin module for power- and weight-constrained orbital applications, but it says that product will be available at a later date. The announcement names several space companies using Nvidia platforms; it does not announce a SpaceX purchase contract. [10]

Why analysts see pressure on neoclouds #

Bernstein’s warning, reported by MarketWatch, is about allocation rather than a disclosed contract. The firm said Nvidia could move SpaceX to “the front of the line” for limited chip supply, a possibility that could put pressure on neocloud providers such as CoreWeave and Nebius. [6]

That concern reflects the companies’ overlapping business models. CoreWeave and Nebius provide cloud access to Nvidia-based infrastructure, while SpaceX is selling compute capacity from its own facilities under long-term agreements with AI customers. IDC described the Anthropic arrangement as a concentrated block of power, facilities, networking and Nvidia capacity selected from a market that also includes CoreWeave, Lambda and Nebius. [11]

The effect on CoreWeave and Nebius depends on whether SpaceX receives incremental supply or draws from allocations that would otherwise go to cloud partners. The public record does not establish that Nvidia has diverted chips from either company. Nvidia’s Rubin materials list CoreWeave and Nebius among the cloud partners preparing deployments, which indicates that both remain part of Nvidia’s launch channel. [4]

No Nvidia financing has been announced #

There is no public evidence that Nvidia is financing SpaceX’s buildout through an equity investment, loan or purchase commitment connected to Musk’s August statement. SpaceX announced a senior unsecured notes offering on June 22 and said it intended to use the proceeds primarily to repay a bridge loan, with the remainder for general corporate purposes. [12]

That financing could support SpaceX’s broader capital needs, but SpaceX did not identify Nvidia as a lender or investor in the announcement. The confirmed development is a supplier choice stated by Musk, not a quantified Nvidia-SpaceX chip transaction.