SpaceX just dropped a $52 billion AI server order with Foxconn, and the numbers are genuinely absurd. Thirteen thousand racks. 936,000 GPUs. Four million bucks per rack. Shipments kick off late this year and stretch into early next year. Whether Foxconn bags the entire contract or splits it remains unclear, but this shatters the Dell and Supermicro duopoly wide open.
Foxconn wants 40 percent of worldwide AI server revenue by year's end, and this SpaceX order doesn't even factor into that projection. Meanwhile, Colossus 1 already runs over 220,000 GPUs across H100, H200, and GB200 chips, while Colossus 2 flexes over 550,000 units split between GB200 and GB300 accelerators. The compute footprint here is genuinely staggering.
And the cloud deals are wild. Google locked in 110,000 GPU-equivalent capacity for $920 million monthly, while Anthropic grabbed 220,000 GPUs at $1.25 billion per month. The Anthropic contract sits on Colossus 1, where xAI only uses 11 percent of available FLOPs. A separate $6.3 billion Pentagon-linked Reflection deal feeds Colossus 2. SpaceX is basically renting out a GPU empire.
Foxconn wants 40 percent of worldwide AI server revenue by year's end, and this SpaceX order doesn't even factor into that projection. Meanwhile, Colossus 1 already runs over 220,000 GPUs across H100, H200, and GB200 chips, while Colossus 2 flexes over 550,000 units split between GB200 and GB300 accelerators. The compute footprint here is genuinely staggering.
And the cloud deals are wild. Google locked in 110,000 GPU-equivalent capacity for $920 million monthly, while Anthropic grabbed 220,000 GPUs at $1.25 billion per month. The Anthropic contract sits on Colossus 1, where xAI only uses 11 percent of available FLOPs. A separate $6.3 billion Pentagon-linked Reflection deal feeds Colossus 2. SpaceX is basically renting out a GPU empire.
