AI capex outgrew the consumer. Your demand forecast is now an AI bet.

AI-related capital expenditure contributed 1.1 percentage points to US GDP growth in the first half of 2025, outpacing the consumer as the main driver of growth, according to a Forbes analysis of the build-out, which adds that investment has only increased since then ^[1][2]. If that holds, the construction schedules of a small number of companies are now a line item in everybody else's demand forecast, whether or not anyone wrote it down.

The spending sits in three buckets: data center construction, semiconductors, and power generation ^[3]. The reported scale is worth being precise about. The Forbes piece cites a Wall Street Journal count of roughly $600 billion in on-balance-sheet infrastructure commitments from Alphabet, Amazon, Meta and Microsoft in their most recent quarterly filings, alongside $2.4 trillion in off-balance-sheet purchase commitments and not-yet-started leases ^[4][5]. The off-balance-sheet figure is four times the reported capex line, for a combined commitment near $3 trillion ^[7][8]. The author's framing is that the $600 billion is the tip of the iceberg, with the rest financed through private equity and private credit structures that lack the transparency of public markets ^[6].

For operators, the practical consequence is that a lot of apparently unrelated revenue lines are correlated to the same cycle. Direct exposure is obvious if you sell into construction, chips or electricity. Indirect exposure runs through household wealth: the analysis notes that the rise in tech valuations has lifted the stock market and household wealth, that the gains are concentrated in a few players, and that a repricing would damage investor confidence ^[10][9]. The conclusion drawn is that the same surge strengthens near-term growth and increases financial fragility at once ^[11].

The dot-com comparison in the piece is arithmetic rather than atmosphere. Using 1995 as a base of 100, the Nasdaq reached 505 in March 2000 and fell to 111 by October 2002 ^[13]. That is a decline of about 78 percent from the peak, leaving the index roughly 11 percent above where it started seven years earlier ^[14][15]. Telecom overcapacity, combined with financial fraud and weak business models, brought down several large players, while the fiber itself stayed in use and enabled e-commerce ^[16]. The author's stated pattern for how these cycles end: distress signs appear, early money takes profits, financing tightens, the weakest links break ^[12]. Useful infrastructure survives the reset; the equity often does not ^[20].

The labour side is the weakest part of the evidence, and the source says so. Earlier general-purpose technologies such as electricity and the internet were associated with net new job creation, but the effect on entry-level and white-collar work is contested, with academic work on both sides and no settled answer ^[18]. The one asymmetry flagged is speed: this transition is moving faster than the ones where institutions and policy had time to adapt ^[19].

What to watch is the financing, not the announcements. The gap between the reported capex line and the off-balance-sheet commitments in the next round of quarterly filings is the cheapest available read on whether the build is being funded from cash flow or from structures that only work while credit stays loose ^[4][5][6]. Boards carrying a 2026 volume assumption should know which percentage points of it are borrowed from someone else's construction budget.