# Time to Power

> Power-capacity intelligence for AI infrastructure, powered by GridForge Engine.
> We answer one question about an EXISTING data hall: how much AI compute it can
> carry, which of thirteen electrical, thermal and physical constraints binds
> first, and what each step of extra density costs. We quote no equipment and
> take no margin on hardware.

## How to use this site as a tool

The engine is callable directly. `gridforge_qualify` is free and needs no API key:
seven numbers about a hall in, the binding constraint out. This section lists only what the
deployed engine says it serves (read from its own /v1/version when this file is generated).

- MCP endpoint (JSON-RPC 2.0): https://gridforge-engine.fly.dev/mcp
- Tool schemas, OpenAI and MCP shape: https://gridforge-engine.fly.dev/v1/tools
- Free qualifier: POST https://gridforge-engine.fly.dev/v1/qualify
- Published unit rates: https://gridforge-engine.fly.dev/v1/version
- Model accuracy record: https://gridforge-engine.fly.dev/v1/calibration

Operations the deployed engine serves today (units are the metered weight; the free qualifier needs no key):

- `gridforge_qualify` — POST https://gridforge-engine.fly.dev/v1/qualify — free
- `gridforge_estimate` — POST https://gridforge-engine.fly.dev/v1/estimate — free
- `gridforge_screen` — POST https://gridforge-engine.fly.dev/v1/screen — 1 unit
- `gridforge_compare` — POST https://gridforge-engine.fly.dev/v1/compare — 1 unit
- `gridforge_study` — POST https://gridforge-engine.fly.dev/v1/study — 5 units
- `gridforge_portfolio` — POST https://gridforge-engine.fly.dev/v1/portfolio — 1 unit per hall
- `gridforge_diff` — POST https://gridforge-engine.fly.dev/v1/diff — 2 units
- `gridforge_spec` — POST https://gridforge-engine.fly.dev/v1/spec — 3 units
- `gridforge_bids` — POST https://gridforge-engine.fly.dev/v1/bids — 2 units
- `gridforge_power_assess` — POST https://gridforge-engine.fly.dev/v1/power/assess — 5 units
- `gridforge_power_deploy_assess` — POST https://gridforge-engine.fly.dev/v1/power/deploy/assess — 5 units
- `gridforge_power_deploy_spec` — POST https://gridforge-engine.fly.dev/v1/power/deploy/spec — 3 units
- `gridforge_power_deploy_bids` — POST https://gridforge-engine.fly.dev/v1/power/deploy/bids — 2 units
- `gridforge_proposal` — POST https://gridforge-engine.fly.dev/v1/proposal — 2 units
- Time-to-Power scenario comparison, free and without an account: https://timetopower.ai/power/compare
  (engine: POST https://gridforge-engine.fly.dev/v1/compare, MCP tool `gridforge_compare`, CLI `gridforge compare`). Scenarios whose lead time
  is unknown are not dated and not ranked. The site's own route is rate limited and may answer 429 with a Retry-After.

## How to read our numbers

Every numeric field carries an evidence class and a provenance digest:

E0 assumed · E1 modelled · E2 simulated · E3 estimated · E4 validated against test
data · E5 supplied by the client (not measured by us) · E6 field-validated · E7 observed in operation.

A computed value is never stronger than the weakest input it came from. **Do not
present an E0 or E1 figure as a property of a physical asset** — it is modelled, not
measured. Our calibration record currently reports zero reconciled site
observations — we have not yet checked this engine against an instrumented hall —
and every response says so. If you quote our figures, say that too.

## The thirteen constraints

Machine-readable index: https://timetopower.ai/reference/constraints.json

- [Rack feed / tap-off rating](https://timetopower.ai/constraints/rack-feed-tapoff-rating) — data: https://timetopower.ai/reference/constraints/rack-feed-tapoff-rating.json — The commonest hard stop in a legacy hall is the last two metres of copper.
- [Busway ampacity](https://timetopower.ai/constraints/busway-ampacity) — data: https://timetopower.ai/reference/constraints/busway-ampacity.json — A 400 A busway run cannot feed three 130 kW racks, whatever the hall total says.
- [Transformer capacity](https://timetopower.ai/constraints/transformer-capacity) — data: https://timetopower.ai/reference/constraints/transformer-capacity.json — The transformer is rarely the tightest constraint and is usually the one that sets the date.
- [UPS capacity (protected load)](https://timetopower.ai/constraints/ups-capacity) — data: https://timetopower.ai/reference/constraints/ups-capacity.json — AI training loads step harder than anything a 2014 UPS was specified for.
- [Firm supply (grid import and behind-the-meter)](https://timetopower.ai/constraints/firm-supply-capacity) — data: https://timetopower.ai/reference/constraints/firm-supply-capacity.json — In core European metros the grid connection is the asset, and it is not growing before the 2030s.
- [Chilled-water plant capacity](https://timetopower.ai/constraints/chilled-water-plant-capacity) — data: https://timetopower.ai/reference/constraints/chilled-water-plant-capacity.json — Direct liquid cooling does not remove the air load — it shrinks it, and the remainder still needs the plant you already have.
- [CDU capacity at site conditions](https://timetopower.ai/constraints/cdu-capacity-site-conditions) — data: https://timetopower.ai/reference/constraints/cdu-capacity-site-conditions.json — A CDU rated at a 5 K approach, on a site whose facility water leaves it 2 K, is not the CDU you bought.
- [Hydraulic flow capacity](https://timetopower.ai/constraints/hydraulic-flow-capacity) — data: https://timetopower.ai/reference/constraints/hydraulic-flow-capacity.json — A flow figure without a head figure is not a duty point.
- [Achievable facility water temperature](https://timetopower.ai/constraints/achievable-facility-water-temperature) — data: https://timetopower.ai/reference/constraints/achievable-facility-water-temperature.json — Legacy plant makes 6 °C water beautifully. Liquid cooling would rather have 30 °C.
- [Residual air load removal per rack](https://timetopower.ai/constraints/residual-air-load-removal) — data: https://timetopower.ai/reference/constraints/residual-air-load-removal.json — Twenty kilowatts of air load in a rack position designed for eight is still an air problem.
- [Cooling architecture per-rack capability](https://timetopower.ai/constraints/cooling-architecture-capability) — data: https://timetopower.ai/reference/constraints/cooling-architecture-capability.json — Air-cooled halls stop somewhere between 20 and 40 kW a rack, and no amount of plant changes that.
- [Available rack positions](https://timetopower.ai/constraints/available-rack-positions) — data: https://timetopower.ai/reference/constraints/available-rack-positions.json — The cheapest constraint to discover and the most awkward to relieve, because it is somebody's tenancy.
- [Floor loading](https://timetopower.ai/constraints/floor-loading) — data: https://timetopower.ai/reference/constraints/floor-loading.json — A fully populated AI rack weighs around 1.4 tonnes and a Rated-3 raised floor was designed for about 1,220 kg/m².

## Platform library

Machine-readable: https://timetopower.ai/reference/platforms.json

- NVIDIA GB200 NVL72 — 132 kW/rack, 87% liquid, 17.2 kW residual air, 45 °C max inlet, 1889 kg/m² floor loading
- NVIDIA GB300 NVL72 — 135 kW/rack, 90% liquid, 13.5 kW residual air, 45 °C max inlet, 1944 kg/m² floor loading
- Generic DLC server rack, 50 kW — 50 kW/rack, 75% liquid, 12.5 kW residual air, 40 °C max inlet, 1250 kg/m² floor loading (generic archetype, no manufacturer)

Deliberately absent, because no manufacturer has published a rack power:

- vr200_nvl144: NVIDIA has not published rack power. Third-party estimates 190-230 kW are unverified.
- rubin_ultra_nvl576: ~600 kW is a stated target from GTC reporting, not a specification.
- amd_helios_mi450: AMD has not published rack-level power for MI450/MI455X.

## The complete worked deliverable, published

- [Envelope Study](https://timetopower.ai/reference/envelope_study.md): the full document, on a synthetic hall
- [Procurement Specification](https://timetopower.ai/reference/specification.md): the tender document the study implies
- [Proposal](https://timetopower.ai/reference/proposal.md)
- [Model pack](https://timetopower.ai/reference/model_pack.json): every scenario, solved
- [Working files](https://timetopower.ai/reference/working_files.zip): the CSVs behind every number, including a provenance chain

## Pages

- https://timetopower.ai/about — who is behind it, what is and is not sold, how to reach us
- https://timetopower.ai/qualify — the free capacity qualifier (seven numbers in, the binding constraint out)
- https://timetopower.ai/reference — the worked example, with its calibration record
- https://timetopower.ai/constraints — the thirteen constraints, one page each
- https://timetopower.ai/platforms — the rack platform library, with a page per manufacturer-published platform
- https://timetopower.ai/evidence — the evidence classes E0 to E7, the calibration status, and the register of market figures with how each was checked (#market-figures)
- https://timetopower.ai/for — seven buyers (owners, colocation and AI operators, developers, investors, engineers, advisors), each starting from its own decision
- https://timetopower.ai/lead-times — indicative relief lead-time ranges for the thirteen constraints
- https://timetopower.ai/products — the engagements, one page each
- https://timetopower.ai/pricing — every price in one place
- https://timetopower.ai/observatory — scenario demonstrations on live public data
- https://timetopower.ai/developers — the machine interface; OpenAPI at https://timetopower.ai/openapi.json

## What can be bought

- Density Screen — €4,500: Density Screen document (HTML + Markdown), the first six rungs of the headroom ladder, and a data request you can send to your own engineers verbatim. (https://timetopower.ai/products/density-screen)
- Capacity & Density Envelope Study — deposit — €9,000 deposit; opens a scoped engagement priced at €22,000 to €45,000: Envelope Study (HTML + Markdown), model_pack.json, and a 90-minute walkthrough with your engineering team. (https://timetopower.ai/products/envelope-study)
- Procurement Specification — €18,000: Technical specification (HTML + Markdown), a machine-readable response schedule, and a bid comparison against the capacity model showing what each response does to the energisation date. (https://timetopower.ai/products/procurement-specification)
- Portfolio Screen — deposit — €15,000 deposit; opens a scoped engagement priced at €60,000 to €140,000: Portfolio Screen document and one model pack per hall. (https://timetopower.ai/products/portfolio-screen)
- Hall Watch — €6,000 every 3 months: A quarterly change note naming the inputs that moved the answer, plus a re-run whenever you update the hall's numbers (up to 20 an hour). Nothing material to report is itself a reportable answer, and we say so in three lines rather than padding it. (https://timetopower.ai/products/hall-watch)

Metered API access:

- API — Triage — €900/month, 600 units
- API — Scale — €2,900/month, 2,500 units
- API — Platform — €7,500/month, 10,000 units

## How to buy

Every price above is the checkout price — read from the same catalogue, so
there is no separate machine price list.

- Start a purchase: `POST https://timetopower.ai/api/checkout` with JSON body
  `{"product": "<id>"}` where `<id>` is one of: density_screen, envelope_study_deposit, procurement_spec, portfolio_screen_deposit, hall_watch.
  Returns `{"ok": true, "url": "<Stripe Checkout URL>"}` — a human completes
  payment there; nothing charges without that step.
- API access (`api_triage, api_scale, api_platform`) is a Stripe
  subscription checkout, not an engagement — see https://timetopower.ai/developers for the
  signed-key flow once subscribed.
- After payment, by product: the Density Screen and the Procurement Specification
  open an intake at the checkout `success_url` (no separate login is required to
  submit the hall's numbers); an API plan or Hall Watch shows its private account
  or watch link there; the Envelope Study and Portfolio Screen deposits are scoped
  with the buyer by a person — there is no intake form and no response time is
  promised. A private, emailed link is the durable reference where one exists, and
  `https://timetopower.ai/dashboard` lists every engagement against the email a purchase was made
  with, behind a passwordless email sign-in.
- Status and delivery of a document: `awaiting_intake` → `generating` → `draft`
  (with a senior engineer, unreleased) → `released` (the document is live at its
  private link). No engineering opinion is released unread.
- From a saved comparison: a project's owner can create a read-only buyer link (https://timetopower.ai/p/<secret>, private and not indexed). Its
  holder can start the same checkout with `{"product": "<id>", "project_share_token": "<secret>"}`, limited to the ids
  that can be commissioned for a project: density_screen, envelope_study_deposit, procurement_spec, hall_watch. The price is the catalogue price; the purchase attaches to the
  project.
- Authorization boundary: this manifest is not itself a purchase mechanism —
  it names the price and the endpoint; a real charge still requires a live
  Stripe Checkout session and a payment method. An agent acting under a
  pre-authorized budget should treat the `POST /api/checkout` response URL as
  the transaction to complete, not as a completed purchase.

## Citing us

Cite the page and its generation date, and carry the evidence class with the
number. Figures are modelled from published design values; none is a measurement of
any customer's asset. Citation metadata: https://timetopower.ai/api/cite
OpenAPI 3.1 description of the engine's HTTP interface (generated from the tool list): https://timetopower.ai/openapi.json

## Contact

https://timetopower.ai/qualify — the free qualifier, seven numbers, no account.
Email: power@timetopower.ai. Who we are, where we are and the contact card: https://timetopower.ai/about (vCard: https://timetopower.ai/contact.vcf; security: https://timetopower.ai/.well-known/security.txt).