Gartner raised its 2026 AI infrastructure spending forecast by 4.8 points and its 2027 forecast by 8.1 points, revisions that reflect opportunistic pricing rather than new capacity. DRAM and HBM manufacturers are running at full capacity. They cannot expand output much in the near term. That constraint now governs the entire AI accelerator market, and it puts a governor on the constant upward revisions in AI spending forecasts seen for the past several years.

Gartner's latest numbers show a modest uptick for AI datacenter infrastructure, software and services in 2026 and 2027. The largest changes concern the servers and networks that underpin AI systems. AI software forecasts moved only a few tenths of a point since the January forecast. AI services revenue forecasts for this year and next were actually trimmed by a few tenths of a point.

What is driving the higher infrastructure numbers?

Price, not volume. With HBM makers unable to add meaningful capacity, suppliers pass price increases through to customers. Those increases inch the spending forecasts up. But HBM capacity remains the gating factor, so revenue forecasts will no longer leap every couple of months as they did after the GenAI boom began in late 2022. The forecast revisions, in the words of the analysis, reflect "expected opportunistic pricing more than capacity expansion."

The money at stake remains enormous. Before the GenAI boom, IT spending growth of 5 percent to 8 percent per year drew constant complaint. AI spending has now delivered enormous incremental sums into the IT sector on top of that baseline.

How much of the IT budget does AI now consume?

By the source's math, using Gartner overall IT spending figures and the AI spending forecasts, AI went from 13.7 percent of total IT spending in 2024 to 31.7 percent in 2025. Part of that jump came from AI growth. Part came from shrinkage: the rest of the IT spending pie contracted by 12.6 percent.

Anyone who claims AI is not eating into other parts of the IT budget is wrong, the analysis states. That trend is expected to continue indefinitely, even as companies spend to modernize and consolidate traditional servers to free room and budget for AI projects.

That consolidation boom shows up in recent financial results from Dell and Hewlett Packard Enterprise. It will not last, according to the analysis. At some point companies will return to incremental installations of traditional servers, and spending on traditional back office and analytics systems will resume its slide.

Will AI spending pass non-AI spending in 2027?

Gartner has not yet issued an AI spending forecast for 2027 in full. The source's own projection, marked as a guess, suggests aggregate AI spending on IT hardware, software and services will exceed non-AI spending by a small margin in 2027 — the first time in history. The gap is expected to widen thereafter.

That crossover assumes AI projects demonstrably pay off. The analysis frames the risk directly: increased efficiencies, productivity, reduced costs and increased revenues from AI "is not a foregone conclusion." A single recession could stop the AI spending spree, sustain it, or accelerate it at the expense of legacy IT systems.

The only certainty is the memory constraint. Until DRAM and HBM makers add capacity, price increases — not shipment growth — will carry AI infrastructure revenue upward, and forecast revisions will stay measured rather than explosive.

By Grace Kim