Late-April Q1 2026 earnings calls from Microsoft, Alphabet, Amazon, and Meta confirmed a combined $725 billion of 2026 capital expenditure, up 77 per cent from 2025's $410 billion record. That works out to roughly $2 billion of new infrastructure spend every day this year, around three-quarters of it AI-related. Microsoft's AI business cleared a $37 billion annual run rate in fiscal Q3, up 123 per cent year-on-year. Google Cloud's customer backlog jumped past $460 billion. Amazon spent $44.2 billion in a single quarter and reaffirmed a full-year plan approaching $200 billion. Meta raised both ends of its 2026 guidance by $10 billion to $125-145 billion.
None of these numbers moved when Maine voted, when Seattle froze, when Compass walked away from Prince William. Capital allocation cycles quarterly inside hyperscaler boards; consent cycles through ballots, hearings, and litigation calendars on five-to-eight-year clocks. Microsoft's $37 billion AI run rate is the structural change inside the prints rather than the headline aggregate, because it is the threshold at which AI revenue can be argued to fund AI capex on a forward basis without diluting existing shareholders. Meta is in a different position: capex is rising faster than monetisation can plausibly catch up, which is why a 16,000-job restructuring lands alongside the $10 billion guidance increase.
The consequence shows up at the supplier end. GE Vernova's Q1 backlog rose from 83 GW to 100 GW in a single quarter, with $2.4 billion of data-centre electrification orders in the quarter alone, exceeding the whole of 2025. Customers pulled aggressively into 2030, and roughly 10 GW of available production capacity remains across 2029 and 2030 combined. The behind-the-meter gas escape, the standard route round a slow grid queue, has now become its own queue paced by factory throughput rather than utility planning. Capital can outrun consent, because consent cannot lobby a 2030 turbine slot. It cannot outrun a turbine that has not been built yet.
