Big Tech is spending so much money building AI infrastructure that it is, quite literally, carrying the entire U.S. economy on its back. The capital expenditure ("capex") going into AI systems — data centers, GPU clusters, networking, power — is on track to add 2.5% to U.S. GDP growth in 2026 and more than 3% in 2027. In Q1 2026 alone, AI-related capex was responsible for an astonishing 75% of all U.S. economic growth. The combined 2026 capex commitment from the top-five hyperscalers — Amazon, Microsoft, Alphabet, Meta, and Oracle — is now tracking above $440 billion, and that figure understates the real number because it ignores private AI labs, sovereign projects, and enterprise deployments. Stopping the boom is no longer a policy debate; it is a recession trigger. In the geopolitical context of USA versus China, the policy answer is settled.

The 2.5% GDP Number Is Probably Too Low

The official tally — 2.5% of 2026 GDP growth from AI capex — is a hyperscaler-only estimate built on top-down BEA inputs and the disclosed 2026 capex guidance from the five companies that dominate the AI infrastructure spend. It does not capture the rapidly expanding cohort of private AI labs (OpenAI, Anthropic, xAI, Mistral) buying their own clusters; it does not capture sovereign AI projects (UAE, Saudi Arabia, Japan, India) standing up national capacity; and it does not capture the enterprise deployment wave — banks, retailers, pharma, defense — pulling forward five-year IT plans into 18-month AI rollouts. Stacking those layers, real AI capex is closer to 3.5-4% of GDP this year and could exceed 5% by 2028.

The "Token Factory" Phase vs the ROI Phase

Right now the money is going into building token factories — physical AI data centers stuffed with NVIDIA Blackwell, AMD MI400, and increasingly custom silicon (Google TPU v6, AWS Trainium 3, Microsoft Maia 200). The output of a token factory is not a service yet; it is the capacity to produce inference and training tokens. This is the canal-and-railroad phase of an industrial revolution: enormous fixed-cost spend with minimal revenue contribution at the moment of deployment.

The ROI phase begins when those factories run real workloads at scale: AI agents writing production code, automating workflows, condensing legal review, generating drug-discovery candidates, replacing customer-support headcount. The economic literature is consistent — every previous general-purpose technology platform (electricity, the internet, mobile) produced returns on capital that exceeded the build-out cost by an order of magnitude, but only after the buildout was largely complete. We are not there yet.

The capex-to-revenue gap

Microsoft disclosed roughly $13 billion in annualized AI revenue in its most recent quarter, against ~$89 billion in AI capex — a 6.8× gap. That ratio is structurally unsustainable forever, but historically it is what every infrastructure cycle looks like in years one to four. The bull case requires the gap to close from the revenue side; the bear case requires it to close from the capex side.

Why the Government Will Not Stop the Boom

Some policymakers are nervous about AI capex concentration; the FTC has flagged vertical integration; the EU continues to litigate digital markets law. None of it matters in the near term. Pulling back AI spend would mechanically print a recession headline within two quarters. No administration is going to volunteer for that outcome, particularly when the political framing is dominated by USA-vs-China strategic competition.

Beijing has tripled state-directed AI investment since 2024 and is racing to deploy domestic GPU alternatives at scale. Washington's read is that this is a national-security competition with a productivity dividend on the other side. That framing virtually guarantees policy support — through tax incentives, accelerated permitting for power and data centers, and continued export-control posture against advanced semiconductor sales to China. The NVIDIA position in this geopolitical equation is the single most important corporate variable.

How AI Capex Trickles Down to the IA13 Cohort

Hyperscaler capex does not vanish. Every billion that Amazon, Microsoft, Google, Meta, and Oracle spend on AI infrastructure flows directly into a narrow basket of providers — the IA13. The cohort spans GPU and accelerator silicon, networking, data-center physical layer, and power generation. Order books are committed 12-18 months out for most of these names.

The key feature of the IA13 thesis is concentration: there are only a handful of companies on Earth that can supply hyperscale-grade infrastructure at the volumes required. The AMD-Meta multi-year compute deal is the latest data point on how locked-in this demand is. Our AMD-vs-NVIDIA framework tracks the indicators that distinguish bull and bear scenarios for the GPU duopoly.

What Investors Should Watch Next

Hyperscaler capex guidance every quarter

The single highest-signal data point is the trailing-twelve-months AI capex line in the 10-Q filings of the top-5 hyperscalers. If guidance for 2027 keeps moving up — which has been the pattern for six consecutive quarters — the IA13 thesis stays intact. The first quarter where two or more hyperscalers cut guidance simultaneously would be the cleanest possible top-of-cycle signal.

Power, not GPUs, is now the binding constraint

Grid operators in PJM, ERCOT, and MISO are at limits. Constellation Energy and NextEra have multi-year nuclear PPAs pre-booked with hyperscalers. Watch any policy moves on nuclear restart, transmission permitting, or behind-the-meter generation — those unlock the next 2 percentage points of GDP contribution. GE Vernova's order book is a leading indicator; its multiple has expanded with the thesis.

The capex-to-revenue convergence

Microsoft's 6.8× capex/revenue ratio will compress one of two ways. Bull: AI revenue accelerates (agents at scale, enterprise penetration, pricing power). Bear: capex slows because the marginal data center cannot be filled with paying workloads. The first path produces a multi-year compounding window for the IA13 cohort. The second produces a sharp drawdown.

Sovereign AI buildouts

UAE, Saudi Arabia, India, Japan, and a growing list of mid-size nations are committing tens of billions to sovereign AI capacity. These are largely additive to U.S. hyperscaler spend, not competitive with it, and the supplier base overlaps almost completely with the IA13. Sovereign demand could add another 50-100 basis points to global AI capex growth through 2028.

For the macro view on AI's contribution to GDP, the BEA quarterly releases are authoritative. For corporate capex, SEC EDGAR 10-Q filings carry line-item detail for each hyperscaler. And Bloomberg's S&P 500 dashboard tracks the sector concentration that the IA13 thesis depends on.