The Race for AI Capability: Who Will Win?

The Nebius Dilemma: Explosive demand but power limitations

Nebius Group (NBIS) faces a “luxury” problem: it sells all its capacity before bringing it online. In Q3, every available megawatt was instantly absorbed by eager customers wanting to secure NVIDIA's Blackwell chips.

But here is the uncomfortable reality: the lack of energy is the real bottleneck, not the lack of demand.

Nebius's strategy is ambitious:

• Contracted capacity: 2.5 GW for 2026 (vs 1 GW months ago)
• CapEx 2025: $5 billion ( tripled from $2 billion )
• Projected ARR revenue: $7-9 billion by the end of 2026

But the competitors do not sleep:

Microsoft is accelerating its existing cloud advantage:

• AI capacity will increase by +80% during 2025
• Azure will grow ~37% in Q2 (demand continues to outstrip supply)
• Even so, expect to be capacity constrained until the end of the fiscal year
• Total revenue: $79.5-80.6 billion (growth 14-16%)

CoreWeave is also in the race, although with setbacks:

• Delays in delivery of infrastructure from data center partners
• Reduced 2025 projection: $5.05-5.15 billion ( vs $5.15-5.35 billion previously )
• Adjusted operating income: $690-720 million ( from $800-830 million)

The market outlook:

NBIS has skyrocketed 144% in 6 months ( vs 6.9% of the software/internet sector ). But the book valuation is only 4.66X compared to 39.95X of sector peers—indicating that there is still room.

The crucial thing: at this moment, whoever can connect power to data centers faster wins. It's not about who has better technology, it's about who closes the energy gap first. And that requires massive capital, strategic partnerships, and impeccable execution.

The bottleneck is not digital. It is physical.