Gartner: AI Infrastructure Is Now the Line Item Driving $6.37 Trillion in IT Spend

Gartner now projects worldwide IT spending will hit $6.37 trillion in 2026, a 14.2% jump over 2025 and an upward revision from the 13.5% growth it forecast earlier this year. The revision matters more than the topline number: analysts don’t usually raise a trillion-dollar forecast mid-year without a specific driver, and Gartner names it directly. Data center systems spending is up 62.5% to $822 billion — by far the fastest-growing of the six segments Gartner tracks, ahead of software (+15.5%), infrastructure-as-a-service (+29.3%), devices (+9.8%), IT services (+5.3%), and communications (+4.4%). Distinguished VP Analyst John-David Lovelock put the scale in stark terms: “building the compute capacity required for AI is the largest infrastructure project ever attempted by humanity.”

That reframes the number: this isn’t a general tech-spending boom, it’s AI-infrastructure buildout pulling everything else along with it. Gartner’s own AI-specific breakout, published a week earlier, shows the same dynamic at a finer grain — worldwide spending on AI models and platforms is set to grow 63.4% in 2026 to $64 billion, with foundation model spending alone more than doubling (104.2%) and domain-specific language models growing 210%. The data-center line in the trillion-dollar forecast and the AI-platform line in the smaller one are describing the same capital cycle from two altitudes.

The capacity crunch behind both numbers is real enough that vendors are looking past terrestrial limits for a fix. Bain’s analysis of orbital data centers notes SpaceX filed with the FCC in January for an “orbital data center system,” and estimates a 100-megawatt orbital constellation currently costs roughly 50% more per kilowatt than a comparable ground facility — a premium Bain expects to close only once launch costs fall from today’s ~$600/kg toward $50-$100/kg.

For companies budgeting AI initiatives against this backdrop, Gartner’s headline flags rising hardware and memory costs as a real headwind stretching device replacement cycles elsewhere in the budget — the infrastructure buildout isn’t free money, it’s a reallocation.