The biggest chokepoint in the AI boom is memory, says SpaceX chief executive
As one of the largest incremental buyer of compute through SpaceX, Elon Musk is well-positioned to comment on demand for memory.
The SpaceX earnings call with Elon Musk on Wednesday was entertaining as always and featured some ambitious plans for rockets and robots, starships and satellites. However, for markets that are increasingly fixated on semiconductor stocks and taking their cue from them daily, it was Musk's comments on memory chips that captured the attention. One of the world's largest buyers of compute observed:
"The limiting factor currently is memory."
Given that a debate has raged all summer over the sustainability of the semiconductor supercycle, amid concerns about the longevity of the AI capex boom, the remarks from the SpaceX $(SPCX)$ chief executive, provided a fillip to memory chip stocks. After a damaging July in which the Philadelphia Semiconductor Index SOX and the chip-heavy Kospi KR:180721 index both declined by around a fifth, August has been much better for share price performance.
In Seoul, shares in the two semiconductor plays that comprise half of the Korean benchmark, SK Hynix (KR:000660) and Samsung Electronics (KR:005930), jumped 5.77% and 2.5% respectively. Other leading stocks in the sector like Japan's Kioxia Holdings (JP:285A) and Taiwan Semiconductor Manufacturing Company (TW:2330) $(TSM)$ also bounced.
The SpaceX and Tesla $(TSLA)$ business mix means that Musk has an excellent vantage point on the demand for memory.
"The memory output is increasing by around 20% per year. Now, normally, that would be fantastically fast and amazing for any large mature industry, but ask yourself: Is the demand increasing by 20% a year," said Musk, according to a FactSet-compiled transcript. "No, the demand is increasing by 200% a year, maybe higher. So, if you've got demand increasing much faster than supply, then economics 101 would suggest that the price increases. It does not decrease."
This contradicts recent theories of some analysts that the capacity expansion by the likes of SK Hynix might lead to oversupply and lower prices.
The urgent requirement for memory chips to build out AI infrastructure means that Musk is both diplomatic and grateful to suppliers. In the recent Tesla earnings call, he thanked Micron $(MU)$ for allocating memory to him while on Wednesday, he made a point of expressing gratitude to Nvidia (NVDA):
"And going forward, we've decided to build exclusively on Nvidia because we think the Vera Rubin architecture is the best architecture. We think it's the best AI computer and we greatly value our close cooperation and partnership on many levels with Nvidia. So, we're exclusive to Nvidia."
The announcement of increased capex at SpaceX was interpreted negatively by traders and investors. SpaceX shares have fallen almost 10% in pre-market trading Wednesday as Chief Financial Officer Bret Johnsen warned spending rose sharply to $18 billion last quarter, compared to the corresponding period last year. Johnsen also forecast a similar number in the next couple of quarters.
This should be music to the ears of the memory chip makers, however. They would also be encouraged by the views of Gavin Baker, chief investment officer of Atreides Asset Management and a well-regarded commentator on the AI sector and related technologies. While doubts about the sustainability of the memory chip makers earnings cycle have been abundant this summer, Baker posted some of his predictions on X Wednesday.
He wrote that "Game theory suggests LTAs are likely to be durable." Long-term agreements are one way that the chip makers can prolong or insure their earnings an margins, by tying up clients into deals that last years. Demand is so strong right now, Baker observes that AI companies will be very reluctant to renegotiate terms with the chip makers because the latter can so easily find other buyers any time they like.
-Jules Rimmer
Comments