Global AI spending is forecast to hit $2.67 trillion in 2026 — a 49.5% jump from $1.79 trillion in 2025 — and more than half of it, $1.48 trillion, is going straight into infrastructure, not software that generates revenue today. That split, reported via a new Gartner-sourced forecast carried by Economic Times, is the clearest number yet showing the AI economy is still being built on borrowed time: capacity first, proof of payback later.
Why it matters: every quarter that infrastructure spend grows faster than monetizable usage, the bet gets bigger that rental rates, utilization, and enterprise adoption will eventually catch up. This is the single number investors, builders, and regulators should be tracking heading into 2026 — and we’ll keep updating it as new forecasts land.
The capex-vs-revenue gap, in numbers
This isn’t a one-off data point — it’s the latest entry in a trend line worth watching as a living story, not a rumor cycle.
- Global AI spend: $2.67 trillion forecast for 2026, up from $1.79 trillion in 2025 (+49.5%).
- AI infrastructure spend: $1.48 trillion in 2026, up from $981.92 billion in 2025 — infrastructure alone is roughly 55% of total AI spend.
- Implication: more than half of every new AI dollar is going to datacenters, chips, and networking — the capital-intensive layer furthest from end-user revenue.
For context, infrastructure spend growing at a similar clip to total spend means the ratio isn’t improving — datacenter and chip buildout isn’t shrinking as a share of the pie even as the market matures. That’s the core of the AI capex vs revenue debate: hyperscalers and neocloud GPU providers are locking in long-duration spend (power contracts, chip orders, datacenter leases) against revenue signals that are still mostly forward-looking guidance, not booked recurring revenue at matching scale.
Why cloud GPU economics are the pressure point
The infrastructure number matters less in isolation than in what it implies for cloud GPU economics. Every dollar of infrastructure spend has to eventually clear two bars: utilization (are the chips running at prices that cover depreciation and power) and rental rate durability (will prices hold as more GPU supply comes online from hyperscalers, neoclouds, and sovereign AI projects simultaneously). Right now, public forecasts are measuring commitment, not clearance. That’s the honest caveat: a $1.48 trillion infrastructure forecast is a spending plan, not a profit statement, and nobody in this data has published matching utilization or margin figures at the same scale.
Litigation tracker: the accountability side of the ledger
While the capex story plays out, the legal exposure side of the AI industry is also growing into its own trackable dataset. As of October 1-2, 2026:
- 149 AI lawsuits are being followed live in U.S. courts by one dedicated tracker.
- A broader lawsuit database counts 246 total AI cases, with 180 active or pending — meaning roughly 73% of tracked cases are still unresolved.
- OpenAI is the most-named defendant across trackers: 53 of 246 cases in one database, 37 in another live tracker — making it the single largest concentration of AI legal risk by company.
- On October 2, 2026, OpenAI reportedly lost a $1 million fee award even after defeating a lawsuit from SpaceXAI — a reminder that procedural wins in AI litigation can still carry real financial costs.
These numbers vary by tracker methodology (what counts as an