Where the Free Megawatt Hides: Anatomy of Stranded Capacity
Most operators are sitting on capacity they paid to build and can't sell. It isn't lost — it's stranded behind the safety margins that were reasonable when the facility was commissioned. Here's where it hides, and what it takes to get it back.
The gap between built and sellable
A data center's nameplate capacity is a design-day number: worst-case ambient, every rack at contracted draw, one cooling unit failed, one feed lost. Real halls almost never operate there. Contracted load runs below allocation, IT draw runs below contract, and the margins stack — provisioning margin on top of contract margin on top of equipment derating.
Stack those layers and the arithmetic is consistent across the industry: on a typical hall, 8–12% of built capacity is powered, cooled, insured — and unsold. On a 100 MW campus that is a hyperscale customer's worth of power, already connected to the grid.
The four hiding places
Contract vs. draw
Tenants contract for peak and run at average. A customer holding 2 MW typically draws 1.2–1.4 MW at their own peak. Without measured coincident-peak data, the difference stays reserved — and unsellable.
Static cooling envelopes
Cooling limits are set from commissioning-day CFD, often a decade stale. Halls re-racked and re-densified since then are governed by thermal maps of a room that no longer exists. Behavioral models routinely find 15–30% more thermal headroom than the static envelope admits.
Failover reservations
N+1 reserves are sized for a failure at design load, not actual load. When the hall runs at 72% of design, the reserve is oversized by the same margin — capacity held hostage to a worst case that is itself overstated.
Stranded pockets
Power and cooling strand separately. A row can have electrical headroom but sit in a warm aisle; the next row has cold air and no breaker positions. Neither shows up as sellable until placement is optimized across both constraints at once.
Why audits don't find it
A manual capacity audit is a snapshot: one team, one spreadsheet, one month, one operating point. It can safely release only what it can safely defend — usually the most obvious 2–4%. The rest requires proving safety across thousands of failure scenarios and load patterns, which is not a job for a spreadsheet. It's a job for physics simulation at machine scale: model the facility as it actually behaves, stress every candidate configuration against N+1 worst cases, and release only what survives.
THE TEST THAT MATTERS
Capacity is only free if it survives the worst day — a cooling unit down, a feed lost, ambient at design extreme, every tenant at coincident peak. Anything released without that proof isn't reclaimed capacity. It's borrowed risk.
What it's worth
At market colocation rates near $120/kW-month, each reclaimed megawatt is roughly $1.4M of annual recurring revenue — at close to 100% margin, because the power, cooling, and building are already paid for. Against the alternative of new construction at $10–13M per MW and multi-year grid queues, reclaimed capacity is the cheapest megawatt an operator will ever sell.
The free megawatt is real. It's sitting in the gap between how your facility was designed and how it actually runs. Finding it takes physics; selling it takes proof.
HEADROOM RESEARCH
Want the stranded-capacity read on your own halls?