SpaceX's crazy expansion is devouring computing power! 2027 GPU shortage alert resurfaces: Bank France and Pakistan sees CoreWeave (CRWV.US), Nebius (NBIS.US), and Oracle (ORCL.US) as the biggest potential winners

The Zhitong Finance App notes that BNP Paribas said that SpaceX (SPCX.US) management's comments on the expansion of artificial intelligence computing power and monetization of cloud services have a broad impact on the entire AI infrastructure ecosystem, including new cloud service providers CoreWeave (CRWV.US) and Nebius (NBIS.US), as well as Nvidia's major customers Microsoft Oracle (MSFT.US) and Oracle Oracle (ORCL.US).

Analyst Stefan Slowinski said, “With SpaceX releasing its first quarterly earnings report as a listed company, Musk has outlined an ambitious infrastructure roadmap with the goal of expanding its AI computing footprint from around 2 gigawatts by the end of 2026 to 'close to 10 gigawatts rather than 5 gigawats' by the end of 2027. Combined with SpaceX's repeated emphasis on the relationship with Nvidia, we are beginning to wonder whether GPU supply will once again become a key constraint on AI infrastructure deployment in the next 12 to 18 months, bringing both risks and opportunities.”

GPU computing power bottlenecks have intensified again, highlighting the value of “strategic priority”

SpaceX's aggressive production expansion (plus 8 giants including Microsoft, Alphabet, Meta, Amazon, Oracle, CoreWeave, and Nebius) is expected to generate 30 GW+ incremental AI demand in 2027. However, the market's revenue expectations for Nvidia only correspond to 19 GW of delivery capacity, and the demand gap is as high as 17%.

This means that GPUs are likely to be in short supply until 2027 and beyond. In this context, leading large customers with “priority deployment rights” of Nvidia hardware or the ability to develop their own chips will receive huge competitive barriers and excessive market dividends.

“Within our coverage area, we continue to see CoreWeave, Nebius, and Oracle as key beneficiaries because of their strategic partnership with Nvidia,” said Slowinsky. We believe that once the market tightens again, this will put them in a good position to get a decent GPU allocation, but there is also a risk of being squeezed out if Nvidia prioritizes an increasingly close partnership with SpaceX,”

Analysts believe that SpaceX's statement further confirms that the shortage of AI computing power is unlikely to be alleviated, even until 2027. SpaceX management said that memory output is growing by about 20% each year, while demand for underlying AI is growing by more than 200%, which means that supply growth continues to lag significantly behind demand. Even if the direction is broadly accurate, this indicates that the current favorable pricing environment is unlikely to return to normal anytime soon.

Analysts believe that, perhaps more importantly, Musk speculates that a system based on the next-generation Vera Rubin architecture can ultimately achieve commercial revenue of between $30 billion and $50 billion per GW, far exceeding the industry's expectations of $15 billion per GW.

According to Slowinsky, SpaceX revealed that its new computing power deployment had a payback period of less than 1 year. This has greatly consolidated the business model where new cloud vendors profit from renting computing power, indicating that the commercial return on computing power infrastructure is still extremely strong.

Pricing pressure coexists with share squeeze

If Nvidia decides to prioritize securing a massive and rapidly growing strategic partnership with SpaceX, even new cloud vendors (such as CoreWeave, Nebius) or traditional cloud giants (such as Oracle), which are also closely related to Nvidia, face the real risk that their chip quota will be overwhelmed by SpaceX.

For small and medium-sized cloud vendors that lack direct strategic cooperation with Nvidia or the ability to develop their own chips, they will face a more serious risk of chip supply cuts and blocked computing power deployment.

Supply-side growth (such as an annual increase in memory production of about 20%) lags seriously behind the demand side (demand for underlying AI increases by more than 200% per year). This serious imbalance between supply and demand means that the current sellers' market environment with high margins and strong pricing power in the computing power market will never be dull or reduced in the short term.

Slowinski said bluntly that there is a clear “logical disconnect” between the market's current performance expectations for Nvidia and the giants' ambitious computing power expansion plans, and GPU supply will once again become a core constraint on AI deployment within the next 12 to 18 months.

SpaceX's computing power roadmap once again proved to the market that AI computing power shortage will not be fundamentally mitigated until 2027, and that manufacturers that can firmly lock in the GPU supply source will be the biggest winners in this competition.