SpaceX is ramping up production and consuming computing power at an alarming rate! The GPU shortage alarm is ringing again for 2027: Socit Gnrale is looking at CoreWeave (CRWV.US), Nebius (NBIS.US), and Oracle Corporation (ORCL.US) as the biggest potential winners.

BNP Paribas noted that comments from SpaceX (SPCX.US) management regarding the expansion of artificial intelligence computing power and the monetization of cloud services have widespread implications for the entire AI infrastructure ecosystem, including new cloud service providers CoreWeave (CRWV.US) and Nebius (NBIS.US), as well as major clients of NVIDIA Corporation: Microsoft Corporation (MSFT.US) and Oracle Corporation (ORCL.US). Analyst Stefan Slawinski stated, "With SpaceX releasing its first quarterly earnings report as a public company, Musk outlined an ambitious infrastructure roadmap aimed at expanding its AI computing footprint from around 2 GW by the end of 2026 to 'close to 10 GW rather than 5 GW' by the end of 2027. Coupled with SpaceXs repeated emphasis on its relationship with NVIDIA Corporation, we are beginning to question whether GPU supply will once again become a key constraint for AI infrastructure deployment in the next 12 to 18 months, bringing both risks and opportunities." The GPU computing power bottleneck is exacerbating, highlighting the value of "strategic priority." SpaceX's aggressive expansion, along with Microsoft Corporation, Alphabet, Meta, Amazon.com, Inc., Oracle Corporation, CoreWeave, and Nebius accounting for a total of eight giants, is expected to generate over 30 GW of incremental AI demand by 2027. However, the market's revenue expectations for NVIDIA Corporation correspond to only 19 GW of delivery capacity, creating a demand gap of approximately 17%. This means that the GPU supply-demand imbalance could persist well into 2027. In this context, major clients with "priority allocation rights" to NVIDIA Corporation hardware or with the capability to develop their own chips will gain significant competitive barriers and excess market benefits. Slawinski pointed out, "Within our coverage, we continue to view CoreWeave, Nebius, and Oracle Corporation as major beneficiaries due to their strategic partnerships with NVIDIA Corporation. We believe that once the market tightens again, this will position them favorably for substantial GPU allocation, but there is also a risk of being squeezed out if NVIDIA Corporation prioritizes its increasingly close relationship with SpaceX." Analysts believe that SpaceX's statements further corroborate that the shortage of AI computing power is unlikely to ease, even by 2027. SpaceX management indicated that memory output is growing by approximately 20% annually, while the underlying AI demand is increasing at over 200%, which means supply growth continues to lag significantly behind demand. Even if the direction is broadly accurate, it indicates that the currently favorable pricing environment is unlikely to normalize in the short term. Perhaps more importantly, Musk speculated that systems based on the next-generation Vera Rubin architecture could eventually realize a commercial monetization potential of $30 billion to $50 billion per GW, far exceeding the previous industry expectation of $15 billion per GW. Slawinski stated that SpaceX has revealed a capital recovery period of less than one year for its new computing power deployments. This greatly reinforces the business model of new cloud providers profiting from leasing computing power, indicating that the commercial returns of computing power infrastructure remain extremely strong. Pricing pressure accompanies market share squeeze. If NVIDIA Corporation decides to prioritize its strategic cooperation with SpaceX, a company of immense scale and rapid growth, even new cloud providers (such as CoreWeave and Nebius) or traditional cloud giants (like Oracle Corporation) that are also closely linked with NVIDIA Corporation face the real risk of their chip allocations being monopolized by SpaceX. For small and medium-sized cloud providers lacking direct strategic partnerships with NVIDIA Corporation or lacking self-developed chip capabilities, the risks of chip supply disruptions and hindered computing power deployments become even more severe. Supply-side growth (e.g., annual memory output increasing by about 20%) is severely lagging behind demand-side growth (underlying AI demand increasing by over 200%). This significant supply-demand imbalance means that the current high-profit, strong pricing power seller's market environment for computing power will not normalize or decline in price in the short term. Slawinski bluntly stated that there is a clear "logical disconnect" between the market's current expectations for NVIDIA Corporation's performance and the ambitious computing power expansion plans of major giants. GPU supply will once again become the core constraint for AI deployment in the next 12 to 18 months. SpaceX's computing power roadmap further demonstrates to the market that the AI computing power shortage will be challenging to alleviate even by 2027, and the manufacturers that can firmly lock in GPU supply sources will be the biggest winners in this competition.