A report claiming that SpaceX placed a $52 billion order for almost one million NVIDIA GPUs has been denied by Elon Musk, leaving the alleged deal unconfirmed.
The original claim said SpaceX had awarded Foxconn part of an order covering about 13,000 NVIDIA GB300 server racks. Each rack was reportedly valued at around $4 million, with shipments expected to begin in the fourth quarter of 2026 and continue into the first quarter of 2027.
The reported configuration would include approximately 936,000 GPUs in total. However, it was unclear whether Foxconn would receive the entire order or share production with other server manufacturers.
Musk later described the report as fake news. His denial means the proposed order should be treated as an unverified rumour unless further evidence emerges from SpaceX, Foxconn, NVIDIA, or regulatory filings.
The reported order would have been one of the largest AI hardware purchases ever
The scale of the alleged procurement was extraordinary, even by the standards of the current AI infrastructure market.
A total of 13,000 racks priced at roughly $4 million each produces a combined value of $52 billion. With 72 GPUs per rack, the configuration would reach 936,000 accelerators.
| Reported order detail | Claimed figure |
|---|---|
| NVIDIA server platform | GB300 |
| Number of racks | Around 13,000 |
| GPUs per rack | 72 |
| Total GPUs | Around 936,000 |
| Estimated cost per rack | $4 million |
| Estimated total value | $52 billion |
| Proposed shipping window | Q4 2026 to Q1 2027 |
| Current status | Denied by Elon Musk |
An order of that size would require enormous supporting infrastructure. The GPU racks alone would need extensive power delivery, liquid cooling, networking, storage, buildings, and utility capacity.
The final cost of operating such a system would therefore be much higher than the reported rack purchase price.
Foxconn was said to have secured part of the manufacturing work
The report suggested that Foxconn had joined SpaceX’s server supply chain and could take work previously handled by companies such as Dell, Supermicro, or other rack integrators.
Foxconn has expanded its AI server business as demand for NVIDIA based infrastructure grows. Securing even part of a 13,000 rack order would strengthen its position among the largest manufacturers of complete AI systems.
The company was reportedly targeting around 40 percent of the global AI server market by the end of 2026. The alleged SpaceX order was said to be separate from that forecast, which would have provided additional growth beyond Foxconn’s existing expectations.
However, the denial means no conclusions should be drawn about Foxconn replacing another supplier or gaining a confirmed SpaceX contract.
The supplied information does not include a statement from Foxconn confirming the deal.
SpaceX and xAI already operate large GPU clusters
The rumour appeared plausible partly because companies linked to Elon Musk have already built very large AI computing facilities.
Colossus 1 was reported to contain more than 220,000 NVIDIA GPUs, including H100, H200, and around 30,000 GB200 accelerators. Colossus 2 was said to contain more than 550,000 GPUs using a mixture of GB200 and GB300 hardware.
If accurate, those figures show that the organisation is already comfortable deploying systems at a scale reached by only a small number of technology companies.
Adding another 936,000 GPUs would still represent a major step. It would push the combined infrastructure toward a level requiring several gigawatts of power, depending on the final configuration and utilisation.
| Existing infrastructure claim | Reported scale |
|---|---|
| Colossus 1 | More than 220,000 GPUs |
| GB200 GPUs in Colossus 1 | Around 30,000 |
| Colossus 2 | More than 550,000 GPUs |
| Rumoured additional order | Around 936,000 GPUs |
These figures also involve different accelerator generations, making direct performance comparisons difficult.
Cloud agreements were cited as evidence of growing compute demand
The report also pointed to large compute agreements involving SpaceX infrastructure.
One agreement was said to provide Google with capacity equivalent to 110,000 NVIDIA GPUs, along with CPUs, memory, and related components, for $920 million per month.
Another arrangement reportedly gave Anthropic access to 220,000 NVIDIA GPUs for $1.25 billion per month, equivalent to $15 billion per year.
A separate $6.3 billion compute agreement connected to Colossus 2 was also mentioned.
These claims were used to support the idea that SpaceX was expanding beyond internal AI development and turning its infrastructure into a large compute service.
The reported structure would allow outside companies to rent capacity while xAI continued using part of the same hardware for model training and inference.
However, none of those earlier claims independently confirms the reported Foxconn order. Existing demand does not prove that a new $52 billion procurement has been approved.
The numbers raise important practical questions
Even without Musk’s denial, the reported timeline would have required careful scrutiny.
Shipping 13,000 GB300 racks between late 2026 and early 2027 would place major pressure on NVIDIA’s supply chain. It would require large quantities of advanced GPUs, HBM, networking equipment, power systems, liquid cooling hardware, and server components.
The deployment site would also need to be ready before the racks arrived.
A single modern AI rack can consume a large amount of electricity. Thousands of racks would require dedicated substations, utility agreements, backup generation, and extensive cooling infrastructure.
The reported $52 billion figure only covered the racks. Construction, land, networking, energy, and operations could add many billions more.
Such a purchase would also have significant implications for other customers because it could absorb a substantial portion of NVIDIA’s available GB300 production.
Musk’s denial leaves the report without confirmation
Elon Musk’s response is the most important update.
By calling the report fake news, he directly rejected the claim that SpaceX had placed the described order. Unless supporting documentation appears, the reported agreement should not be presented as established fact.
Executive denials do not always end speculation, especially when companies keep procurement plans confidential. Plans may also change, or negotiations may exist without a final contract.
Still, a denial from the head of the company carries more weight than an unnamed supply chain report.
The safest interpretation is that Foxconn may continue pursuing large AI server contracts, while SpaceX and xAI are likely to keep expanding their computing infrastructure. The specific claim involving 13,000 racks, 936,000 GPUs, and a $52 billion order remains unsupported and has been publicly denied.
Discussion (0)
Be the first to comment.