BNP Paribas: SpaceX's Aggressive Expansion May Trigger GPU Shortage Again by 2027, Benefiting CoreWeave and Two Others
Miles Bennett
BNP Paribas estimates eight major players will need over 30 GW of AI compute by 2027, but Nvidia can deliver only about 19 GW — a 17% shortfall that could make GPU supply the binding constraint on AI deployment once again.
How big does SpaceX want to go?
Musk's target: scale AI compute from roughly 2 GW to nearly 10 GW by end-2027 — double the prior market expectation of 5 GW.
This means → SpaceX alone could re-tighten the entire GPU supply chain.
SpaceX has repeatedly stressed its partnership with Nvidia. BNP Paribas now asks: will GPUs become scarce again within 18 months?
How large is the demand gap?
BNP Paribas tallies expansion plans from SpaceX, Microsoft, Alphabet, Meta, Amazon, Oracle, CoreWeave, and Nebius. Combined incremental AI compute demand for 2027 exceeds 30 GW.
Consensus Nvidia revenue forecasts imply only about 19 GW of deliverable capacity — a gap of roughly 17%.
In plain terms = buyers want almost a fifth more GPUs than Nvidia can make, and that imbalance may persist beyond 2027.
Why can't supply keep up?
SpaceX management disclosed that memory output grows about 20% a year, while underlying AI demand is growing at over 200% — an order-of-magnitude mismatch.
This means → even at full-tilt expansion, chipmakers cannot close the gap soon; seller pricing power stays strong.
Musk also projected that next-gen Vera Rubin-based systems could monetize at $30–50 billion per GW, far above the prior industry assumption of $15 billion. SpaceX says its new compute deployments pay back in under one year.
Who is most likely to get GPUs?
BNP Paribas names CoreWeave, Nebius, and Oracle as primary beneficiaries — all three have strategic partnerships with Nvidia and hold priority GPU allocations.
If the market tightens again, that priority access becomes a major competitive moat. Put simply = they have a reserved lane for chips others must queue for.
This reflects a shift in AI infrastructure competition: the key barrier is moving from "can you fund a data center" to "can you secure a steady chip supply."
Where are the risks?
Even CoreWeave and Nebius face crowding-out risk: if Nvidia prioritizes its deepening relationship with SpaceX, existing allocations could be reshuffled.
Smaller cloud providers without a direct Nvidia partnership or in-house chip capability face steeper risk of supply cutoffs and deployment delays.
The BNP Paribas analyst warns of a clear "logic disconnect" between the Street's Nvidia earnings estimates and the ambitious buildout plans across major players — GPU supply will reassert itself as the core constraint on AI deployment within 12–18 months.
Content is for reference only, not financial advice.