Elon Musk has become one of the most reliable sources of good news for Nvidia shareholders this year. He added another chapter on Sept. 25.
A single post laid out a chip buildout schedule that would make even the most bullish AI infrastructure forecasts look modest.
The update lands at a moment when questions about how long AI spending can keep climbing are getting louder. It gives Nvidia investors a very specific number to point to when they argue the answer is still years away.
Inside Musk’s Colossus 2 update
In a post on X on Friday, Sept. 25, Musk said Colossus 2, SpaceXAI’s computing cluster near Memphis, currently runs 110,000 Nvidia GB200 chips and 440,000 GB300 chips.
The schedule going forward is aggressive. Another 220,000 GB300 chips are set to come online next week, with a second batch of 220,000 following in November. Musk added he hopes to get yet another 220,000 in late December “if we get lucky,” The Motley Fool reported.
More Elon Musk:
- Elon Musk makes bizarre claims about money, future of AI
- Elon Musk sends blunt verdict on the future of humanity and AI
- Elon Musk’s startling claim to SpaceX investors
On schedule, Colossus 2 hits roughly 1.21 million processors by year-end. About 1.1 million of those would be GB300s. The 110,000-chip increments are not arbitrary. Musk said they reflect how many fiber optic cables fit into a central switch.
Add in Colossus 1, which runs 150,000 H100 chips, 50,000 H200 chips, and 30,000 GB200 chips. The full Memphis site would reach roughly 1.44 million GPUs by the end of 2026. That comfortably beats SpaceXAI’s original goal of 1 million, The Motley Fool reported.
How SpaceXAI became Nvidia’s closest partner
This did not happen by accident. During SpaceX’s first earnings call as a public company in early August, Musk told analysts the company would build its AI infrastructure exclusively on Nvidia. He called the Vera Rubin the best architecture and the best AI computer.
The comment carried real market consequences. AMD posted the strongest quarter in its history that same week and still watched its shares fall as much as 10% as investors weighed its outlook against SpaceXAI’s decision to build exclusively on Nvidia.
The relationship extends well beyond chips in a Memphis data center. SpaceX and Nvidia are jointly developing the compute payload for Starmind AI1. It is the first satellite in a planned constellation of orbital AI data centers, built around Nvidia’s Rubin GPUs and Vera CPUs.
Nvidia also has real money at stake beyond selling hardware. The chipmaker holds a significant equity stake in SpaceX, while Musk himself is the company’s largest individual shareholder, as TheStreet has reported.
The numbers behind the spending
The cost of this buildout is enormous. SpaceXAI spent roughly $18 billion on its initial 550,000 GPUs, and the newer batch of GB300 chips could cost even more given how much more expensive that generation is, The Motley Fool reported.
Musk has also laid out the power side of the equation. He said SpaceXAI would end the year with over 2 gigawatts of computing power. The target for 2027 is closer to 10 gigawatts than 5.
Jensen Huang has separately estimated that one gigawatt of compute costs roughly $50 billion to $60 billion, with Nvidia chips and components making up about $35 billion of that total, 24/7 Wall St reported.
Demand from outside customers helps justify the spending. Reflection AI signed a deal in June to pay SpaceXAI $150 million per month starting July 1 for access to GB300 capacity at Colossus 2. Google separately signed a multiyear agreement paying roughly $920 million per month for compute capacity, Benzinga reported.
Musk has also been renting out capacity to rivals. SpaceXAI agreed in May to lease its entire Colossus 1 data center to Anthropic. That is more than 220,000 Nvidia GPUs powering one of Musk’s own competitors.
What it means for Nvidia stock
The spending backdrop is reflected in Nvidia’s own numbers. The company’s second-quarter revenue hit $96.2 billion, up 106% from a year earlier, with $89 billion coming from its data center unit.
The concentration carries a real risk investors should not ignore. If SpaceXAI’s AI ambitions take longer than expected, Nvidia would absorb both the equity loss and the demand shortfall at once, according to TheStreet.
Inference competition is real and growing. CUDA still makes it hard to leave Nvidia. November is when Nvidia’s third-quarter numbers arrive and the next real data point on whether any of this is slowing.
Related: Elon Musk sends strong message to SpaceX and Tesla investors