AI & SpaceX – Cash Bonanza

SpaceX cashes in.

• Excellent results from SpaceX have allowed the AI sector to breathe a sigh of relief, as SpaceX has demonstrated in black and white the huge returns on offer for those who have capacity available now to sell on a short-term basis.
• Q3 26 revenues / net profit were $14.6bn / $4.4bn, way ahead of expectations of $12.9bn / $2.0bn, driven exclusively by revenues coming from renting out xAI’s data centre capacity.
• Revenue from the data centre business increased 7x YoY and 3x QoQ due to a decision taken about 6 to 9 months ago to park Grok 5 training and rent the compute capacity instead.
• This was not because Grok 5 was going badly but simply because compute for rent has become so scarce that the returns on offer far exceed anything that Grok 5 is likely to produce.
• RFM has calculated that with the spot price of compute capacity for short-term contracts somewhere between $30bn and $50bn per GW, 5-year returns of 45% to 85% annually can be earned.
• In the grand scheme of things, these are fantastic returns and imply that anyone who has capacity should drop whatever it is that they are doing and sell their capacity to the highest bidder.
• This is pretty much what is going on, explaining the creation of Meta’s new enterprise business as well as everyone else building AI compute capacity as fast as they possibly can.
• However, as always, the cure for high prices is high prices, which means that the current returns that are on offer in the spot market for compute are temporary.
• RFM estimates that in a normalised market, AI compute will sell for something closer to $20bn per GW, which offers both an attractive return to the provider of the compute as well as a more reasonable price to the AI service provider renting it.
• It is worth remembering that very few, if any, of the neo-clouds are in a position to take advantage of this because they need to sell their compute for several years into the future to be able to raise the finance to build the data centre.
• The term market is far more reasonable, and RFM thinks that the longer-term deals are being signed somewhere in the region of $20bn to $25bn per GW.
• This means that at some point (probably during 2027), xAI’s financial performance is going to falter as the market normalises and revenue and profit growth slow considerably as the price it can charge for its capacity drops by 50% or more.
• In the meantime, this has given a much-needed boost to the rest of the AI industry, as this is the first concrete demonstration of the returns that are currently on offer for anyone with capacity to sell into the spot market.
• This is why Elice, a Korean neo-cloud that listed yesterday and cratered 30% as Korean retail panicked, has put back the majority of what it lost.
• Most of the rest of the AI sector is up moderately in pre-opening trading, indicating a mild sigh of relief, but by no means has this settled all of the concerns.
• Consequently, more is needed, which is what I am looking for from Meta, which reports Q3 26 results tonight and from GTC Berlin, where Jensen Huang is due to give a keynote and announce some new products, as well as reassure everyone that the AI boom is still very much on.
• We are still very far from seeing supply and demand in the AI compute market move into balance, and with around half of all new data centre builds facing delays, this might take a bit longer than anyone expects.
• Consequently, there remains a great opportunity for anyone with capacity to sell into the spot market, which remains the reason why I have a position in Meta Platforms.
• The popularity of Muse is the icing on the cake.