The question
At $219.71 a share (October 8, 2026), the market was pricing Nebius on future AI compute revenue. We asked whether its physical capacity, meaning sites, power and GPUs, plus what that capacity costs, could deliver the cash flow the price implied.
What I did
Instead of starting from management guidance, I rebuilt revenue from physical constraints:
- Mapped about 18 data-center facilities and deployments: location, ownership, build status, and connected versus contracted power.
- Estimated when each site comes online and how many GPUs it can hold.
- Applied utilization and realized pricing to get compute revenue.
- Layered in customer contracts and ramp-up timing, with scenarios through 2031.
What we found
The risks sat in the gap between building capacity and earning from it:
- Depreciation vs. useful life. GPUs may lose economic value faster than the accounting schedule assumes.
- Capital intensity. Heavy capex funded with debt and convertibles.
- Timing. Capacity commitments come years before revenue.
- Concentration. Reliance on a small number of large customers.
The recommendation: short
| Case | Price target | vs. $219.71 |
|---|---|---|
| Bear | $85 | −61% |
| Base | $120 | −45% |
| Probability-weighted | $148 | −33% |
| Bull | $310 | +41% |
These figures are from our October 2026 submission, not a current forecast.
Limits
Bottom-up capacity estimates depend on what Nebius discloses, and site timelines move. The valuation schedules were team work; my part was the revenue and operations build.