SpaceX is making one of the largest hardware investments in AI history. The company has placed a massive order with Foxconn for Nvidia GB300 AI server racks valued at approximately $52 billion, according to a Taiwanese media report. The order covers around 13,000 GB300 racks, each costing roughly $4 million, totaling about 936,000 Nvidia GPUs. Deliveries are expected to begin in the fourth quarter of 2026 and continue through the first quarter of 2027.
This deal underscores SpaceX's aggressive expansion into AI infrastructure, building on its existing data center capabilities. The company already operates two major facilities: Colossus 1 and Colossus 2, which house hundreds of thousands of Nvidia GPUs. The new order will more than triple SpaceX's GPU count, positioning it as one of the largest private operators of AI compute hardware globally.
The order also highlights the growing importance of AI compute for SpaceX's broader business strategy. Beyond powering its own AI initiatives, the company has been leasing GPU capacity to major tech firms, generating substantial recurring revenue. This dual-use approach allows SpaceX to offset infrastructure costs while maintaining a leading position in AI research.
What's New: The Scale of the Order
The order is unprecedented in both size and cost. Each GB300 rack is a high-density server system designed for AI workloads, and the total GPU count approaches one million units. For context, SpaceX's existing Colossus 1 data center housed over 220,000 Nvidia GPUs as of May 2026, including H100, H200, and about 30,000 GB200 accelerators. The newer Colossus 2 facility already contains more than 550,000 GPUs, primarily GB200 and GB300 models. With this new order, SpaceX's total GPU count could exceed 1.7 million, making it one of the largest GPU fleets in the world, rivaling those of major cloud providers.
Key details of the order:
- 13,000 GB300 server racks
- Approximately 936,000 Nvidia GPUs
- Total value: $52 billion
- Per-rack cost: ~$4 million
- Delivery window: Q4 2026 to Q1 2027
It remains unclear whether Foxconn will handle the entire production or if SpaceX will split the order among multiple system integrators. Foxconn aims to control about 40% of the global AI server market by the end of 2026, and this deal significantly boosts its position. The Taiwanese manufacturer has been a key partner for Nvidia's GB series servers, and this order reinforces its dominance in AI hardware assembly. Foxconn's ability to scale production to meet this demand will be critical, as supply chain constraints have historically limited the availability of high-end AI servers.
The GB300 is Nvidia's latest AI accelerator, succeeding the GB200. It offers improved performance for training and inference, particularly for large language models. Each rack integrates multiple GPUs with high-bandwidth memory and networking, designed to handle the most demanding AI workloads. The per-rack cost of $4 million reflects the advanced technology and tight integration required.
Why It Matters
SpaceX's AI ambitions go beyond its own needs. The company owns SpaceXAI (formerly xAI), Elon Musk's AI venture, but it also monetizes its infrastructure by leasing compute capacity. Earlier, SpaceX signed a cloud services deal with Google worth $920 million per month, providing 110,000 Nvidia GPUs worth of compute, along with CPU, memory, and related infrastructure. Additionally, SpaceX has a contract with Anthropic valued at $1.25 billion per month (about $15 billion annually), granting access to 220,000 Nvidia GPUs including H100, H200, and GB200 models.
This new order suggests SpaceX is betting heavily on AI demand, both for internal use and external leasing. The scale of the investment—$52 billion—underscores the company's confidence in the growing AI market and its ability to become a major compute provider. It also highlights the intense competition for AI hardware, as companies like Microsoft, Amazon, and Google are also investing billions in GPU clusters. SpaceX's move could pressure other players to accelerate their own infrastructure spending.
From a geopolitical perspective, the order also underscores the importance of Taiwan's semiconductor ecosystem. Foxconn, headquartered in Taiwan, is a critical link in the global AI supply chain. Any disruption to Taiwan's manufacturing capabilities could impact SpaceX's delivery timeline, though the company may diversify production across multiple sites. The order also reflects the ongoing trend of tech companies securing dedicated supply chains for AI hardware, bypassing traditional cloud providers to build their own infrastructure.
For the AI industry as a whole, this investment signals that demand for compute is far from saturated. As models grow larger and more capable, the need for specialized hardware like the GB300 will only increase. SpaceX's order could also spur further innovation in server design and cooling solutions, as data centers must handle the immense power and heat generated by such dense GPU clusters.
Our Take
This is a staggering number, even by hyperscaler standards. A $52 billion GPU order is roughly equivalent to the entire annual AI chip spending of some large cloud providers. SpaceX's strategy appears to be building a massive AI compute reserve that can serve multiple customers, similar to what AWS, Azure, and Google Cloud do. However, the reliance on a single supplier (Foxconn) and a single GPU generation (GB300) carries risk. Delivery timelines are also aggressive, given the current supply constraints for advanced AI chips. If SpaceX can execute, it could reshape the AI infrastructure landscape, but the sheer scale means any delays or shifts in demand could have outsized consequences.
It is also worth noting that SpaceX is not just an aerospace company anymore; it is becoming a major player in the AI compute market. By leasing capacity to Google and Anthropic, SpaceX is effectively competing with traditional cloud providers. This dual role—as both a consumer and supplier of AI compute—gives SpaceX unique leverage. The company can prioritize its own AI projects while also generating revenue from excess capacity. However, managing such a massive infrastructure buildout will require significant operational expertise, and any missteps could be costly.
Finally, the order raises questions about the broader AI hardware market. Nvidia's GB300 is not yet widely available, and demand already outstrips supply. SpaceX's massive order could exacerbate shortages for other customers, potentially driving up prices. At the same time, it signals to investors that AI infrastructure spending shows no signs of slowing down, which could boost Nvidia's stock and encourage further investment in the sector. The deal also highlights the growing importance of Foxconn as a manufacturing partner for AI hardware, potentially shifting more leverage to the Taiwanese firm in negotiations with other customers.
From an editorial perspective, this development is a clear indicator that the AI arms race is intensifying. Companies are no longer content to rely on cloud providers; they are building their own infrastructure at unprecedented scale. SpaceX's move may prompt other large enterprises to follow suit, further straining GPU supply and accelerating the need for alternative AI accelerators from companies like AMD and Intel. The long-term implications for the AI ecosystem are profound, as control over compute resources becomes a key competitive advantage.
FAQ
How many GPUs is SpaceX ordering?
The order covers approximately 936,000 Nvidia GPUs across 13,000 GB300 server racks, totaling about 1 million GPUs when considering potential additional units.
What is the total cost of the order?
The deal is valued at roughly $52 billion, with each rack costing about $4 million.
When will the GPUs be delivered?
Deliveries are scheduled to begin in the fourth quarter of 2026 and continue through the first quarter of 2027.
Will SpaceX use all these GPUs itself?
No. SpaceX also leases compute capacity to other companies. It already has agreements with Google and Anthropic to provide GPU access, and this new order likely supports both internal AI development and external cloud services.
What is Foxconn's role?
Foxconn is the primary manufacturer for the GB300 server racks. The company aims to capture about 40% of the global AI server market by end of 2026.
