Key Takeaways
- • Bill Ackman said that if Elon Musk meets SpaceX's 2027 compute buildout objective, SpaceX "might have to borrow a trillion dollars of capital," implying roughly $50 billion per GW across 20 GW of cumulative capacity.
- • Ackman argued the AI compute buildout is sustainable because demand for intelligence-enabling compute will persist, while hyperscalers have the creditworthiness to access capital.
- • Ackman singled out pure-play neocloud providers as highly levered, comparing them to leveraged real estate developers and highlighting their sensitivity to the cost of capital.
- • SpaceX plans to end 2026 with 2.5-3 GW of compute capacity and add roughly 13 GW by end-2027, reaching around 20 GW total, per Wccftech's prior reporting.
- • Wccftech estimates the debt need could be far below $1 trillion, citing reported 2027 compute-related CapEx of $390 billion (about $30 billion per GW for the 13 GW phase).
- • Musk proposed space-based "petawatt/year" AI compute using lunar-manufactured satellites launched by a electromagnetic mass driver, a concept Wccftech notes would require trillions in additional financing.
Summary
Bill Ackman has said that SpaceX may need to borrow as much as $1 trillion to meet the compute buildout targets outlined by Elon Musk. Speaking about SpaceX's earnings call projections, Ackman stated: "If he meets his objective, he might have to borrow a trillion dollars of capital." Wccftech notes that this figure implies a cost of about $50 billion per gigawatt across 20 GW of cumulative capacity.
The context for these projections is SpaceX's stated plan to end 2026 with 2.5-3 GW of compute capacity, including plans for 1.2 million NVIDIA Blackwell GPUs at its Colossus 2 data center, and to add another 13 GW by the end of 2027, bringing total capacity to around 20 GW. Wccftech argues Ackman's per-gigawatt assumption may overstate the required debt, since only the incremental 13 GW is needed to reach the 2027 target. Using reported 2027 compute-related CapEx of $390 billion, the outlet estimates roughly $30 billion per GW for that phase, suggesting SpaceX's actual debt load could be substantially lower than Ackman's figure.
Ackman also gave a broader view of the AI infrastructure cycle, calling the buildout sustainable because demand for the compute that enables intelligence is likely to persist. He said hyperscalers have the financial resources and creditworthiness to access capital, but warned that pure-play neocloud companies, which provide GPU-as-a-service and AI inference capacity, are more exposed. He described them as "a bit like highly levered real estate developers playing a theme" and said they are "much more sensitive to the cost of capital."
Separately, Musk replied to a post by Sequoia's Shaw Maguire on SpaceX economic modeling, writing that launching superintelligence satellites into deep space with a mass driver on the Moon could achieve ">>petawatt/year capability." Wccftech notes that all of humanity currently uses around 20 terawatts, so a petawatt would be about 50 times that level. The proposed architecture would involve manufacturing compute satellites on the Moon and launching them via an electromagnetic track, given the Moon's escape velocity of 2.38 km/s versus 11.2 km/s on Earth. Wccftech adds that realizing this vision would likely require trillions of dollars in additional financing.
The context for these projections is SpaceX's stated plan to end 2026 with 2.5-3 GW of compute capacity, including plans for 1.2 million NVIDIA Blackwell GPUs at its Colossus 2 data center, and to add another 13 GW by the end of 2027, bringing total capacity to around 20 GW. Wccftech argues Ackman's per-gigawatt assumption may overstate the required debt, since only the incremental 13 GW is needed to reach the 2027 target. Using reported 2027 compute-related CapEx of $390 billion, the outlet estimates roughly $30 billion per GW for that phase, suggesting SpaceX's actual debt load could be substantially lower than Ackman's figure.
Ackman also gave a broader view of the AI infrastructure cycle, calling the buildout sustainable because demand for the compute that enables intelligence is likely to persist. He said hyperscalers have the financial resources and creditworthiness to access capital, but warned that pure-play neocloud companies, which provide GPU-as-a-service and AI inference capacity, are more exposed. He described them as "a bit like highly levered real estate developers playing a theme" and said they are "much more sensitive to the cost of capital."
Separately, Musk replied to a post by Sequoia's Shaw Maguire on SpaceX economic modeling, writing that launching superintelligence satellites into deep space with a mass driver on the Moon could achieve ">>petawatt/year capability." Wccftech notes that all of humanity currently uses around 20 terawatts, so a petawatt would be about 50 times that level. The proposed architecture would involve manufacturing compute satellites on the Moon and launching them via an electromagnetic track, given the Moon's escape velocity of 2.38 km/s versus 11.2 km/s on Earth. Wccftech adds that realizing this vision would likely require trillions of dollars in additional financing.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.