Before you underwrite NBIS to $800, ask where the power comes from
Every $800 target on Nebius is built the same way. Take booked GPU capacity. Multiply by an assumed utilization and price-per-GPU-hour. Grow it on a hyperscaler curve. Slap a multiple on it.
The model is clean. That's the problem. It's clean because it skips the hard part.
I'm not calling it a short. I run a constraint-signal framework for my own research — find the physical bottleneck a thesis depends on *before* I touch the revenue line. On NBIS the bottleneck isn't demand. It isn't even chips. It's the gap between booked capacity and deliverable capacity, and almost nobody is pricing it.
**What the models skip**
A neocloud revenue model treats booked capacity like it converts to revenue on a smooth ramp. It doesn't. The conversion runs through a chain none of the DCF inputs capture:
secured power → energized substation → interconnection queue position → commissioned hall → racked and burned-in GPUs → billable hours.
Every link is a real-world queue with its own lead time. And the binding constraint across the entire AI buildout right now is the *front* of that chain — power and grid interconnection — not the back.
That kills the linearity the bulls assume. If your $800 case needs capacity online late 2026 to hit the revenue it's discounting, and energization slips two or three quarters — which is the base case for new large loads in most relevant markets, not the bear case — the revenue doesn't show up late at the same NPV. It shows up late *and* into a worse pricing environment, because everyone else's delayed capacity lands at the same time.
**Why this is a leading indicator, not just a risk factor**
This is the part worth keeping regardless of what you think of the stock.
Power and interconnection status is observable *before* the revenue miss hits a print. Substation energization dates. Queue positions. The gap between announced MW and actually-contracted-and-energized MW. These move quarters ahead of the income statement.
So you don't wait for a guide-down to know the ramp is at risk. The bulls are watching booking announcements and treating them as supply signals. They're not. A signed contract tells you someone wants the compute. It tells you nothing about whether the megawatts to deliver it are energized.
One queue you can accelerate with capital. The other you can't. The grid doesn't care how focused your capex is.
**What makes me wrong**
Writing the falsifiers down so this doesn't turn into permabear cope:
* Energized, contracted power keeping pace with announced capacity — the MW gap closing, not widening
* Depreciation-adjusted margins holding as the fleet scales. Neocloud unit economics live or die on the depreciation schedule against GPU useful life, and that's the *other* number the $800 models wave through
* Revenue ramp tracking energization dates, not booking dates
Those hold, the bull case is real and I'll say so.
**The claim**
NBIS isn't worth zero. Demand's real, the focus is real. But an $800 target underwritten off booked capacity and a hyperscaler growth curve is pricing a delivery timeline the physical constraints don't support — and the tell shows up in power and interconnection data before it shows up in revenue.
If you're long, that's what I'd track. Not the next contract headline.
Where's the hole in it? Especially want to hear from anyone closer to the power/datacenter side than me.