The bottleneck
isn't chips.
It's power.
Give us the technical information for an existing or proposed AI site, and we tell you how much useful AI compute it can actually support, what blocks it, what fixes it costs, and when the compute can go live. No equipment to sell.
Start with a Density Screen: one hall, five working days, a document that names what binds it. Fixed fee, and it credits in full against the full study.
Seven numbers, no card, no call. It names your binding constraint before you decide whether to buy anything.
Reference photography — not a customer site; not measured evidence.
Nobody in the building can say
what stops the next rack. This does.
The hall you already have — racks, cooling, busway, transformer, grid connection. The facts, assembled once.
Thirteen constraint families solved jointly. One sets the ceiling, named and given its evidence class.
The costed ladder of relief: each rung a limit removed, priced, lead-timed, with the racks it unlocks.
What each rung of relief actually costs, quoted at your site's own conditions, not a vendor's reference case.
The single item that sets your energisation date. Not a typical lead time — yours.
The grid connection is already yours. What actually stops the next rack — busway, transformer, cooling, floor loading — is rarely the constraint people expect, and today nobody in the building has it in writing.
Live, deterministic, and quantified — not illustrated.
This pulls the live EPEX day-ahead curve for the German grid, fetched live rather than illustrated — the same market where a blocked connection or a 160-week transformer lead time turns a density constraint into a cost. Below it, put a number on what an unrelieved constraint is costing your own project.
The grid, streaming in real time.
What is an unrelieved constraint costing you?
Put your numbers in.
Revenue assumption is editable (Derived from public hyperscale colo lease ranges, 2025). On-site figure is a planning timeline, not a guarantee — a Density Screen replaces it with a defensible date for your site.
Most halls don’t fail on cooling.
They fail on a 63 A tap-off.
Thirteen constraints decide how much AI compute an existing hall can carry. One binds first, and it is rarely the one people expect. Seven numbers you already know, and the engine tells you which — free, no contact details.
Fixed fee, named artefact, no hourly billing.
Four engagements, every one priced before it starts. The fee on each card is the fee checkout charges — both are read from the same catalogue.
Density Screen
€4,500One hall, five working days. What binds first, how many racks of your target platform it carries as it stands and after the costed ladder, which item sets the energisation date, and the inputs nobody has measured.
Capacity & Density Envelope Study — deposit
€9,000Five cooling architectures compared under one model, the full headroom ladder with costs and lead times, time to power, sensitivity, economics, and the provenance of every figure.
Procurement Specification
€18,000The tender document the study implies: every duty derived from a constraint in your hall and quoted at your site's own conditions, never at a manufacturer's reference condition. We name no make and no model.
Hall Watch
€6,000 per quarterThe model stays live. Re-solved on a cadence and whenever you change an input, with a change note naming what moved and which input moved it — including when the movement came from our side.
Technology rooted in first principles.
A deterministic constraint engine
Thirteen constraint families — busway, transformer, UPS, grid firm capacity, CDU, hydraulics, floor loading and more — solved jointly over one envelope. The binding constraint is the minimum; a single wrong assumption cannot hide behind a good-looking average.
Evidence class and provenance on every number
Every figure carries an evidence class, E0 (assumed) to E7 (observed in operation), and a provenance chain back to its source. Arithmetic propagates the weakest input — a conclusion can never read as more certain than what it was built from.
The Headroom Ladder
Each rung relieves one binding constraint, priced and lead-timed, with the racks it unlocks. Time to power names the single item that sets your energisation date — not a vendor's typical lead time, yours.
A calibration ledger that says so when it's empty
Every modelled envelope is checked against measured outcomes as they arrive. Today the ledger reads UNCALIBRATED on most constraints, and every deliverable says so. That honesty is the point — and the reason it gets harder to copy with every hall we screen.
Every number carries its evidence class.
Assumed, modelled, simulated and estimated figures are labelled as such — E0 through E3. Nothing here is presented as measured or field-validated (E4–E7) until it actually is. The calibration ledger comparing modelled results against instrumented field data is empty today, and every study and API response says so. That is the honest starting point for a future track record, not a gap we hide.
Vincenzo Grimaldi
Time to Power is built by Vincenzo Grimaldi — a grid networks engineer working on the digitalization of high-voltage assets, with an M.Sc. from RWTH Aachen in cross-domain grid intelligence. The premise is simple: the AI buildout is gated by power, and the engine underneath — GridForge Engine — answers what a hall can actually carry with deterministic, evidence-graded engineering, not a vendor catalog.
Founder principles and prior work: igrimaldi.engineering
No overclaiming.
Let's pressure-test your
time-to-power.
One hall, five working days, a fixed fee. You get the constraint that binds it and the list of inputs nobody has measured.
