Data-Center Construction as a Macro Variable

The datacenter buildout stopped being a sector story in 2026. Two independent sources — the Federal Reserve's July 2026 Monetary Policy Report (a first-party central-bank technical report) and ConstructConnect's construction-starts data — now describe data-center construction as large enough to move national statistics. The Fed reports business fixed investment accelerating to 11% annualized growth in Q1 2026 versus 5.5% in 2025, and attributes it directly: "Most of the strength in investment appears to be connected to building the infrastructure necessary to AI services." ConstructConnect measures the same phenomenon from the ground: data centers are now more than 20% of all US nonresidential building starts.

The macro significance cuts two ways, and the second way is the one that matters for positioning. On the upside, the buildout is currently a primary driver of US business investment and, by extension, headline GDP growth. On the downside, that same weight means a capex air-pocket would no longer show up only in hyperscaler earnings — it would show up in national accounts. The Fed's own framing raises the stakes of the Hyperscaler Capex Cycle inflection question from a stock-selection problem to a macro one.

Two secondary effects are worth holding. First, the buildout subtracts from measured GDP through the trade channel: AI-related high-tech imports "soared" in Q1 2026 and net exports took roughly 0.4 percentage points off growth. Second, the productivity payoff has not arrived yet — the Fed puts labor productivity at 2.1% annualized since late 2019 and attributes only a "modest" contribution to AI adoption specifically. In other words, as of mid-2026 the AI economy's measurable macro contribution is overwhelmingly the construction of capacity, not the use of it. That is the demand-side question ("who pays the rent") stated in national-accounts terms.

Key Claims

• Business fixed investment grew ~11% annualized in Q1 2026 vs 5.5% in 2025, with the Fed attributing most of the strength to AI-services infrastructure. Evidence: moderate (technical-report — Federal Reserve, first-party) (Fed MPR July 2026)
• Data centers are >20% of all US nonresidential building starts. Evidence: moderate (analysis — ConstructConnect, industry-primary data) (ConstructConnect July 2026)
• YTD 2026 data-center construction spending through May: $58.1B — more than 4x the same-period 2025 record. Evidence: moderate (analysis — ConstructConnect) (ConstructConnect July 2026)
• Still accelerating, not plateauing: 6-month moving average ($12.0B/mo) runs above the 12-month moving average ($9.7B/mo). Evidence: moderate (analysis — ConstructConnect) (ConstructConnect July 2026)
• AI-related high-tech imports "soared" in Q1 2026; net exports subtracted ~0.4pp from GDP growth. Evidence: moderate (technical-report — Federal Reserve) (Fed MPR July 2026)
• The productivity payoff has not yet landed: labor productivity ~2.1% annualized since late 2019, with only a "modest" contribution attributed to AI adoption. Evidence: moderate (technical-report — Federal Reserve) (Fed MPR July 2026)
• Two independent sources agree directionally — a central bank measuring investment flows and a construction-data firm measuring building starts describe the same step-change. Evidence: moderate (two independent sources, one first-party) (both sources)
• [FORECAST] Power-infrastructure construction is projected to grow +30.8% in 2026 vs 2025 — a projection, not a measured figure. Evidence: weak (single analysis forecast) (ConstructConnect July 2026)

Benchmarks & Data

• Business fixed investment: +11% annualized (Q1 2026) vs +5.5% (2025)
• YTD 2026 DC construction spending through May: $58.1B (>4x 2025 same-period record)
• May 2026 monthly spend: $7.9B (8th-highest month on record)
• Projects YTD 2026: 91
• 12-month moving average: $9.7B/mo · 6-month moving average: $12.0B/mo
• Preconstruction pipeline: ~100 projects, >$101B planned (excludes the separate ~$100B "Project Kestrel")
• Data centers as share of US nonresidential building starts: >20%
• Net-export drag on GDP from AI-related imports: ~0.4pp (Q1 2026)
• Labor productivity: ~2.1% annualized since late 2019 (AI contribution "modest")
• Geographic concentration: VA, TX, LA, IL, NC ≈ 60% of new DC starts spending (trailing 12 months); New England + Western US combined ≈ 6%
• [FORECAST] Power-infrastructure construction growth 2026: +30.8%

Open Questions

• If hyperscaler capex inflects down, how large is the GDP hit and with what lag? The Fed has established the exposure but not sized the downside.
• What share of the $58.1B YTD construction spend is hyperscaler-owned vs neocloud vs colo vs vertically-integrated (see Vertically Integrated Compute)?
• Does the ~0.4pp import drag grow as more of the capex dollar shifts to accelerators (imported) from structures (domestic)?
• Why is the geographic concentration so extreme (5 states ≈ 60%), and is it a power-availability artifact or a permitting/tax artifact?
• When does the productivity channel start showing up, and what would the first evidence look like in the data the Fed publishes?
• Is "Project Kestrel" (~$100B, excluded from the ConstructConnect pipeline figure) a single named project, and whose? Not identified in the source.

Related Concepts

• Hyperscaler Capex Cycle — this concept is that cycle measured in national accounts rather than company guidance
• Power: The Binding Constraint — the +30.8% power-construction forecast is the bottleneck showing up in construction data
• Contracted Backlog & RPO — the demand-side counterweight: who has actually committed to pay for this capacity

Backlinks

Pages that reference this concept:

• Hyperscaler Buyers
• Alphabet

Changelog

• 2026-07-22 — Created from 2 sources (fed-mpr-july-2026-datacenter-construction, constructconnect-datacenter-report-july-2026). Establishes the macro-exposure thread: datacenter construction now moves US national statistics; productivity payoff not yet visible.