Gartner's 2026 AI spending forecast
Published 15 August 2026
Gartner forecasts worldwide AI spending of $2.59 trillion in 2026, up 47% year over year, reaching $3.49 trillion in 2027. Enterprise AI spend - what buying organizations actually spend rather than what the industry builds - is $407 billion, up 34.8% from $302 billion. Those are two different numbers and only one of them is a benchmark you can use.
Which number is which
| Figure | 2026 | What it measures |
|---|---|---|
| Worldwide AI spending | $2.59T (+47%) | The whole industry, dominated by supply-side build-out. |
| AI infrastructure | $975.6B → $1.43T | Servers, data centres, silicon. Over 45% of the total. |
| Enterprise AI spend | $407B (+34.8%) | What buying organizations spend. The only peer-comparable line. |
| 2027 worldwide | $3.49T | Direction of travel, not a planning input. |
The headline gets quoted in slide decks as a demand signal. Most of it is supply - capital deployed by hyperscalers and chip makers on the expectation that demand arrives. Benchmark against $407 billion.
What the infrastructure share implies for your bill
Over 45% of $2.59 trillion is capital equipment that has to earn a return, and it earns it through utilization. Hold that in mind whenever a plan assumes falling unit prices will lower spend.
Per-token prices have fallen consistently and bills have risen anyway, because cheaper tokens make longer contexts, more retries, and more agent loops economically reasonable. An infrastructure build of this scale is a structural reason to expect pricing that rewards volume, and to expect the pattern to continue rather than reverse. This is the practical case for consumption-denominated budgets: forecast tokens per completed task and derive the dollar figure, not the other way round. See LLM cost trends 2025-2026 and why LLM bills spike.
Where the money concentrates by industry
Financial services leads at roughly $68 billion with 79% adoption. Healthcare follows at roughly $45 billion. The spend-plus-adoption pairing is the part worth reading closely - $68 billion at 79% adoption is broad production deployment, not a handful of large experiments.
If you sit in financial services, a gap is a production gap rather than a pilot gap, and production gaps compound: the attribution, governance, and cost discipline that make the second workload cheap only get built during the first. See LLM cost benchmarks by industry.
How this squares with spend being deferred
Forrester expects enterprises to defer roughly a quarter of planned AI spend into 2027. Gartner forecasts 47% growth. They look contradictory and are not.
The total grows because infrastructure commitments lock in years ahead and adoption is still spreading. Individual workloads get deferred because scrutiny arrived at the same time, with fewer than a third of decision-makers able to tie AI value to financial growth. Money is being redistributed inside a growing envelope. A rising market is cover for a reallocation, not protection from it. See a quarter of AI spend is slipping to 2027.
What a market forecast is good for
Useful for: sector framing (is our growth rate ordinary or unusual?), timing arguments (deferring inside a 35%-growth market is a competitive decision, not a neutral one), and vendor conversations - a supplier quoting scarcity in a market growing this fast is quoting a strategy, not a constraint. See LLM purchasing guide.
Not useful for: setting your own number. A market forecast contains no unit of your business. "Peers spend X% of IT budget on AI" says nothing about whether your support agent is worth $0.31 per resolved ticket - and budgeting off a peer percentage produces a defensible-looking figure with no defensible unit under it, which is precisely the profile that gets cut in a deferral review.
The three lines that matter more than the forecast
- Your own growth rate against $407 billion at 34.8%.
- A unit economic for one workflow, at full cost, with a counterfactual. See AI value management.
- Your consumption trajectory in tokens, not dollars. See token budget implementation.
The forecast says the room is getting bigger. It does not say whether you are spending well in it, and only the second question gets asked in a budget review.
Related
- A quarter of AI spend is slipping to 2027 - the counterweight to the growth number.
- LLM cost trends 2025-2026 - why falling prices raise bills.
- LLM cost benchmarks by industry - a usable comparison, unlike the market total.
- State of FinOps 2026: 98% now manage AI spend - the practitioner view.
- How to budget for AI spend - the mechanics.
Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →
FAQ
How much will be spent on AI in 2026?
Gartner forecasts worldwide AI spending of $2.59 trillion in 2026, up 47% year over year, reaching $3.49 trillion in 2027. Enterprise AI spend - what buying organizations spend rather than what the industry builds - is forecast at $407 billion, up 34.8% from $302 billion in 2025.
What is the difference between the $2.59T and the $407B?
The $2.59 trillion is the whole industry, dominated by supply-side build-out: AI infrastructure alone runs from $975.6 billion to $1.43 trillion, over 45% of the total. The $407 billion is what buying organizations spend. Only the second is a peer benchmark.
Which industries spend the most on AI?
Financial services leads at roughly $68 billion with 79% adoption, ahead of healthcare at roughly $45 billion. The pairing of high spend with high adoption means broad production deployment rather than a few large experiments.
What does the infrastructure share mean for consumption forecasts?
Over 45% of the total is capital equipment that has to earn a return through utilization. That is a structural reason to expect pricing that rewards volume and to keep expecting bills to rise even as per-token prices fall. Budget in consumption and derive dollars from it.
Is AI spending growth slowing?
Not in aggregate. Gartner's 47% growth sits alongside Forrester's prediction that enterprises defer 25% of planned AI spend into 2027. Both hold: the total grows while individual workloads are reallocated toward those with demonstrated value.
How should a FinOps team use a market forecast?
For context and direction, not as a budget input. It is useful for showing whether your growth rate is unusual for your sector and for anchoring vendor conversations. It cannot tell you what your own AI spend should be, because it contains no unit of your business.