AI infrastructure market size in 2026: from $1T in capex to $42B you can rent

How big is the AI infrastructure market in 2026? It depends on which layer you measure. Count every dollar spent on data centers and the answer passes a trillion. Count only the AI cloud capacity a team can rent today and it is $42 billion. Both numbers are real, and the gap between them says a lot about where the market is heading.

This post lays out the main forecasts layer by layer, with the source for each one.

The numbers at a glance

Layer	2026 figure	Forecast	Source
Total worldwide AI spending	$2.7T	Infrastructure is over 45% of it	Gartner
Global data center capex	$1T+	$1.7T by 2030	Dell'Oro Group
Capex from Amazon, Google, Microsoft and Meta	$725B	Up 77% from $410B in 2025	Company guidance
AI infrastructure spend	$497B	$1.08T in 2029, $1.21T in 2030	IDC
AI-optimized cloud infrastructure (IaaS)	$42B	Up 96% year on year	Gartner
Neocloud revenue	$25B+ (2025)	Approaching $400B by 2031	Synergy Research Group

Layer 1: total AI spending, $2.7 trillion

Gartner puts total worldwide AI spending at $2.7 trillion in 2026. That covers everything: software, services, devices and infrastructure. Infrastructure makes up over 45% of it, which makes the hardware and the buildings it sits in the single biggest piece of the AI economy.

Layer 2: data center capex, past $1 trillion

Dell'Oro Group puts global data center capex at over $1 trillion in 2026, heading to $1.7 trillion by 2030.

A large share comes from four companies. Per their own guidance, Amazon, Google, Microsoft and Meta plan $725 billion of capex in 2026, up 77% from $410 billion in 2025. A jump that size in one year is the clearest sign of how hard the largest buyers are leaning in.

Layer 3: AI infrastructure, $497 billion

IDC sizes AI infrastructure spend at $497 billion in 2026, up 56% on the year. Its forecast runs to $1.08 trillion in 2029 and $1.21 trillion in 2030.

The forecasts themselves keep moving up. Inside a single year, IDC raised its 2029 figure from $758 billion to over $1 trillion. When forecasters revise that far that fast, the usual reason is that actual spending keeps outrunning the earlier models.

Layer 4: the AI cloud you can rent, $42 billion

Most of the spending above is capex: companies building and owning capacity. The layer that matters to most AI teams is smaller. It is the capacity you can rent by the hour or by the contract.

Gartner sizes AI-optimized cloud infrastructure at $42 billion in 2026, up 96% year on year. That is the market most AI startups and research labs actually buy from. It is small next to the trillion-dollar capex figure, and it is nearly doubling every year.

Neoclouds: $25 billion and climbing

Part of that rentable market runs through neoclouds, the GPU-first clouds built around AI workloads. Synergy Research Group puts neocloud revenue at over $25 billion in 2025, approaching $400 billion by 2031. CoreWeave, the best-known of them, went public in March 2025. For more on what separates neoclouds from hyperscalers, read what neoclouds are.

What the numbers mean for GPU buyers

Three things follow from the gap between a trillion dollars of capex and $42 billion of rentable cloud.

Supply is going up fast, and it is fragmented. SemiAnalysis tracks 209 GPU clouds. Every one of them is competing to fill capacity, and each has its own console, contract and API.

Prices move. A market growing this fast doesn't hold still. AWS cut its H100 on-demand price 44% in one announcement in June 2025, and one-year H100 contract prices then rose about 40% between October 2025 and March 2026. See how GPU rental prices moved for the full timeline.

Capacity is tight anyway. SemiAnalysis reported on-demand capacity sold out across all GPU types in 2026, and Thunder Compute reports most new NVIDIA allocation is reserved into 2027. Spending is rising, but so is demand.

Put together, that is a big, fast-growing market with no single place to see it. Buyers compare prices by hand across a handful of providers they already know, and miss most of the supply.

Where Opengrid fits

Up front: Opengrid is pre-launch. The market view ships first and the routing API is in development.

Opengrid is building the market layer for that rentable slice: a live view of pricing, availability, capacity and regions across GPU clouds, then one API that routes each job to whichever provider wins on price, availability, region and reliability. The bigger the market and the faster it moves, the more a single view of it is worth. For how that layer has played out one level up, read OpenRouter for GPUs.