SpaceX – Spending Binge

$40bn is just the beginning.

• SpaceX’s plan to raise $40bn to buy Nvidia systems for its data centre expansion is headline news in all of the financial press, but remains a small fraction of what SpaceX will need, highlighting the likelihood of a large issuance of equity in the near future.
• At its Q2 26 results, SpaceX predicted that its year-end 2027 capacity would be closer to 10GW than to 5GW, meaning that it should hit at least 7.5GW by the end of 2027.
• Assuming that it hits the 2GW target by the end of 2026, this means that SpaceX will need to add at least 5.5GW in 2027 alone.
• The capex requirement to build 1GW of Vera Rubin is around $61bn, while Vera Blackwell is around $50bn, meaning that SpaceX needs to spend between $275bn and $336bn on data centre capex in 2027 if this target is to be met.
• Of these figures, Nvidia addresses 68% – 85% of the capex bill, meaning that SpaceX could spend between $187bn and $286bn with Nvidia in 2027 alone.
• SpaceX and Mr Musk in particular are well known to be able to build factories and data centres much faster than anyone else, but the scale of this target requires a suspension of disbelief.
• Against this backdrop, the $40bn being raised by SpaceX to spend on Nvidia equipment covers less than 20% of the 2027 bill, assuming that the raise is only for 2027 investments.
• Given that Nvidia is effectively sold out 12 months in advance, SpaceX will have already had to make commitments to Nvidia for far more if it wants to get filled in 2027, meaning that there are many more and much larger raises on the way.
• At the moment, anyone who has capacity to sell can earn between $30bn and $50bn per GW per year, which, assuming the IT equipment lasts for 5 years, will deliver annual returns of 48% to more than 80%.
• These are stunning returns, which explains why everyone is rushing to build capacity as quickly as they possibly can or to park existing activities and sell the vacated compute to the highest bidder.
• These stunning returns also ensure that the situation will not last, as the cure for high prices is high prices, and whether demand is crimped from compute becoming unaffordable or supply increases, prices are going to fall to normal levels at some point.
• These returns explain how SpaceX can raise this sort of money quite easily, and as there is no sign of the compute shortage being alleviated until at least H2 2027, I expected more raises to come over the next few months.
• SpaceX already has $100bn of cash on its balance sheet, but all of this and another $175bn – $230bn will be needed to meet its targets, which, if it were all in debt, would very likely cause the balance sheet to become unsustainably geared.
• Hence, I suspect that there is an equity issuance on the cards which, given the recovery in the share price over the last 2 months, is now a more viable proposition.
• The main beneficiary here is, of course, Nvidia, where the shares have substantially derated as a result of the market not really believing its 2027 guidance.
• Hence, there is scope for a further rally in Nvidia’s share price, which has already begun since I started thinking about it.
• The risk-reward on Nvidia remains pretty attractive.