🏗️ Data centers — is the physical capacity to house AI still growing?

Latest observations

Trend lines appear after 6 comparable observations of the same definition — the monthly series therefore need about 6 months. "Not yet measured" means missing, not zero. Machine-readable: /us/api/datacenter_us.json.

Observation tables

Question this table answers: what are the values and dates in each row of "Cooling degree days — CONUS population-weighted, base 65°F"? (source or retrieval date on each row)

Question this table answers: what are the values and dates in each row of "Private data-center construction spending — millions of USD, not seasonally adjusted"? (source or retrieval date on each row)

Question this table answers: what are the values and dates in each row of "Unfilled orders, Electrical Equipment Manufacturing — millions of USD, seasonally adjusted"? (source or retrieval date on each row)

Method, scope, and why this chapter is not judged

• Not judged: no frozen criterion exists for calling these levels good or bad, so the status is "observing". Any future criterion will be defined in a versioned document and frozen before a badge appears.
• Regional aggregates only: this chapter never publishes facility-level data, site locations, or company names. All series are national or regional totals from official statistics.
• The lead-time series is a proxy: no public monthly statistic of transformer lead times exists (checked 2026-08-29). Unfilled orders of Electrical Equipment Manufacturing (which contains transformer manufacturing, NAICS 335311) is the closest public proxy — it measures the order queue in dollars, not delivery weeks.
• Cooling degree days use the CPC definition as-is (population-weighted, base 65°F, degree-day method per the ASHRAE Handbook — Fundamentals). We do not compute or re-base degree days ourselves.
• Boundaries with other chapters: grid operations (demand, generation, prices) belong to the electricity chapter; rulemaking belongs to the regulation chapter; drought and water indicators belong to the water chapter (planned). This chapter covers only the facility side: construction, components, and cooling conditions.
• Data-center construction spending is a United States-only series — no equivalent monthly public statistic was found for other countries (checked 2026-08-29).
• What this chapter can say: how much private money went into data-center construction each month (Census VIP, millions of USD, not seasonally adjusted); whether the order backlog of U.S. electrical-equipment manufacturers (Census M3 unfilled orders; the industry includes transformer makers) is growing or shrinking month by month (a lead-time proxy); how much cooling the CONUS climate demanded each day (NOAA CPC population-weighted cooling degree days, base 65°F).
• What this chapter cannot say: how many data centers exist, are under construction, or where (dollar totals only; no facility, site or company data by design); transformer delivery times in weeks (the backlog is measured in dollars of orders, not in delivery time); whether any level is good or bad, or how the U.S. compares with the EU on these series (different concepts and bases; only the Japan edition shares the 65°F degree-day base).

Sources, licenses, keys, and rate limits

• U.S. Census Bureau, Value of Construction Put in Place Survey — data-center category, monthly, millions of USD. Public domain; retrieved via the Census API (free key required; usage limits per the Census API terms). Compiled by AI Lifeline.
• U.S. Census Bureau, M3 Survey — unfilled orders, Electrical Equipment Manufacturing, monthly, seasonally adjusted. Public domain; same Census API and key.
• NOAA Climate Prediction Center, Degree Days Statistics — daily, population-weighted, base 65°F. Public domain; no key; retrieved once per day (no formal rate limit published — we keep to one fetch per day by design).