Artificial intelligence stocks have experienced significant volatility this year, swinging between optimism and skepticism. Investors initially backed AI infrastructure builders, then suddenly worried about rapid capacity expansion.
This uncertainty severely affected neocloud providers such as CoreWeave (NASDAQ:CRWV), Nebius Group (NASDAQ:NBIS), and IREN (NASDAQ:IREN), whose business models rely on leasing advanced GPU clusters to AI developers. Conventional wisdom held that if hyperscalers oversupplied compute, specialized providers would lose their edge.
However, a new bottom-up model from Morgan Stanley indicates the market misjudged the risk. Rather than oversupply, the coming years may witness one of the largest infrastructure buildouts in tech history.
The AI Buildout Is Only Accelerating
Morgan Stanley estimates that total compute capacity among the five major hyperscalers will rise from 30.5 gigawatts in 2025 to 116.6 gigawatts by 2028—a nearly fourfold increase. Each added gigawatt requires design, financing, and construction, implying $4 trillion to $8 trillion in capital expenditure through 2028.
The bank’s breakdown of projected capacity growth illustrates the scale:
| Company | 2025 Compute Capacity | 2028 Compute Capacity | % Increase |
|---|---|---|---|
| Amazon (NASDAQ:AMZN) | 13.8 GW | 35.8 GW | 159% |
| 5.0 GW | 31.6 GW | 532% | |
| Microsoft (NASDAQ:MSFT) | 7.5 GW | 20.3 GW | 171% |
| Meta Platforms (NASDAQ:META) | 3.5 GW | 21.2 GW | 506% |
| SpaceX (NASDAQ:SPCX) | 0.7 GW | 7.8 GW | 1,014% |
| Hyperscaler Total | 30.5 GW | 116.6 GW | 282% |
Source: Morgan Stanley bottom-up hyperscaler compute model.
These figures undermine the excess-compute bear narrative that pressured AI infrastructure stocks earlier this year. Demand cannot quadruple in three years if the industry is awash in idle capacity.
Why Nebius Still Has Room To Grow
The demand outlook clarifies why Nebius occupies a vital niche. In Q1, revenue surged 684% year over year, and management reaffirmed a target of $7 billion–$9 billion annualized revenue run rate by year-end. Long term, the firm aims for $51 billion in annual revenue by 2030 from owned infrastructure alone.
Nebius recently unveiled an asset-light partnership model: third-party data center owners fund new facilities while Nebius provides its AI cloud platform, customer base, and operational expertise. This lets Nebius expand geographically without heavy balance-sheet strain, earning recurring licensing and revenue-share fees. The approach accelerates scaling and adds a growth lever beyond self-funded infrastructure.
Key Takeaway
The bear case against neoclouds rested on two premises: cooling AI demand and financing struggles. Morgan Stanley’s model counters the first by projecting hyperscaler capacity jumping to 116.6 GW by 2028. Nebius’s partnership model counters the second by shedding much of the capital burden.
Execution risks remain, and Nebius must hit its targets. Yet if Morgan Stanley is right, the market’s 35% selloff of Nebius shares appears a reaction to a non-existent problem—potentially an opening for long-term investors.
