For Project Controls escalation you can put in a document

No official data-center PPI exists. Escalation data for DC construction is annual, retrospective and delivered as a PDF. So we built the index the estimating, scheduling, change-management and risk functions need: a daily, decomposed, weight-published DC Build index with an append-only audit trail — and the last mile into your workflow: a basis-of-estimate bridge, a contingency table graded on every vintage, a program view, a market panel at county resolution, and a long-lead board built from primary filings.

DC Build · YoY

+8.4%

construction input costs · as of 2026-10-02 · 2018-01=100

Long-run basis · shortfall

62%

of 12-month vintage-true windows under-provisioned · mean 6.2pp · 101 anchors

Publishes on record

since 2026-07-08 · append-only, never restated

Build components

published weights · 2 carry a live proxy tail

Cite

The toolkit — from estimate through procurement

Basis of estimate

Escalation calculator

Escalate your own base estimate between any two months of the index, with the $/MW bridge that shows which packages moved it.

Open →

Program

Portfolio view

Your projects — base, delivery, carry basis — aggregated into escalation to date, at delivery, and the realized band. Stays in your browser.

Open →

Contingency

Escalation grades

Every contingency basis graded on every vintage since 2018, strict and extended. The basis with the best error has been the worst contingency.

Open →

Market conditions

DC market panel

Construction wages and headcount at county resolution for 19 real data-center markets, against the national rate.

Open →

Long-lead

Long-lead board

Switchgear, transformers, generators, HVAC, pumps — PPI beside 9 vendors' stated order books.

Open →

Index

DC Cost Index

Build, Ops and Hardware indexes with published weights, component modes and state parity multipliers.

Open →

Output side

Compute prices

What a token and a GPU-hour cost — the price of what the facility produces, beside the price of building it.

Open →

Supply

AI capacity

Operational, under-construction and planned critical-IT MW by company, with the filing behind each figure.

Open →

Defensibility — the three receipts

1 · A citable number

Every headline carries a reference string

Series, as-of date, rebase, value and a link that reproduces the view — the Copy button under each KPI. CSV and JSON downloads sit beside every table; /data lists every artifact with its JSON Schema.

2 · A graded contingency

Expected vs realized on every vintage

The scatter puts each anchor month's carried basis against what the index actually did over 12–48 months. The published shortfall rates are recomputed from the same dots. Bases that look best on error have under-provisioned most often — the table says so.

3 · A history that cannot be restated

What the index read on any past day

Point in Time reads an append-only ledger of every publish — the row, the timestamp, the git commit. The vintage store beneath it keeps every release of every input. A counterparty cannot argue the history moved.

Vocabulary — what the numbers here are, and are not

Escalation	The ratio of the DC Build index between two months, applied to your base estimate. National. Not a quote.
Basis of estimate	Your base cost and base month. The calculator's bridge shows which of the twelve packages moved the ratio, in your dollars.
Contingency basis	A realized annualized rate of the index over a stated window — long-run, trailing 3-year, current momentum — carried forward from the last complete month. Graded, not forecast.
Realized band	The p10–p90 of every like-length window in the sample. A range of what has happened; not a probability distribution.
Long-lead	Equipment packages whose vendor order books we track from primary filings, beside the PPI leg we publish for each. Backlog basis is badged; bases are never summed.
$/MW	Yours to supply: the index is a ratio, so any unit you give it comes back in that unit. The market panel's MW columns are denominated and status-split.
Energization	Not modelled. The capacity tracker records stated dates per site with the filing behind each; the long-lead board is the earliest signal we have.

Methodology, weights and every negative result we have published — including the lead-lag study that found no forward model is warranted — are on /datacenter and /dc-scoreboard. The index is not a contract-referenced series today; its weights can change with notice on the methodology page.