Gartner says $2.59 trillion. Now what?
August 15, 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 organisations actually spend, rather than what the industry builds - is $407 billion, up 34.8% from $302 billion. Those are two very different numbers and only one of them belongs in your board pack. Here is which, and what each is good for.
Read the two numbers separately
| Figure | 2026 | What it actually 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 organisations spend. The only peer-comparable line. |
| 2027 worldwide | $3.49T | Direction of travel, not a plan input. |
The distinction matters because the headline number gets quoted in board decks as though it were a demand signal. Most of it is supply: capital being deployed by hyperscalers and chip makers on the expectation that demand arrives. When you benchmark, benchmark against $407 billion.
Where the money concentrates by industry
Financial services leads at roughly $68 billion with 79% adoption. Healthcare follows at roughly $45 billion. The pairing of spend and adoption is the part worth reading closely - $68 billion at 79% adoption is not a few large experiments, it is broad deployment.
If you sit in financial services, the implication is uncomfortable and specific: your competitors are not piloting. A gap there is a production gap, and production gaps compound because the instrumentation, governance, and cost discipline that make the second workflow cheap are only built during the first. See AI FinOps for CFOs.
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. Those look contradictory and are not.
The total grows because infrastructure commitments are locked in years ahead and because adoption is still spreading. Individual programmes get deferred because scrutiny has arrived at the same time - fewer than a third of decision-makers can tie AI value to financial growth. Money is being redistributed inside a growing envelope, toward programmes that can show an outcome. A rising market is not protection; it is cover for a reallocation you may not notice until your line moves. See a quarter of planned AI spend is slipping to 2027.
What the infrastructure share tells a buyer
Over 45% of $2.59 trillion in capital equipment has to earn a return, and it earns it through consumption. That is worth holding in mind whenever you model AI spend downward on the assumption that falling unit prices lower bills.
Per-token prices have fallen consistently and bills have gone up anyway, because cheaper tokens make longer contexts, more retries, and more agent loops economically reasonable. An infrastructure build of this size is a structural reason to expect that pattern to continue rather than reverse. Budget in consumption and derive the dollar figure from it, not the other way round. See falling token prices, rising bills.
What a market forecast can and cannot do for you
Useful for: board framing (is our growth rate unusual for our sector, or ordinary?), timing arguments (a deferral in 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 AI vendor negotiation.
Not useful for: setting your own number. A market forecast contains no unit of your business. "Peers spend X% of IT budget on AI" tells you nothing about whether your support agent is worth $0.31 per resolved ticket. Budgeting off a peer percentage is how a programme ends up with a defensible-looking number and no defensible unit underneath it - which, per the deferral data, is exactly the profile that gets cut.
The three lines to bring to the board
- Your growth rate against $407 billion at 34.8%. Materially above needs an explanation. Materially below in financial services needs a different one.
- Your unit economics for one workflow, with a counterfactual. This is the line the market forecast cannot supply and the only one that survives a deferral review. See proving AI ROI.
- Your consumption trajectory in tokens. With infrastructure of this scale seeking utilisation, a dollar-denominated forecast will keep being wrong in the same direction. See AI spend forecasting.
The forecast tells you the room is getting bigger. It does not tell you whether you are spending well in it. Only the second question gets asked in a budget review.
Related
- A quarter of planned AI spend is slipping to 2027
- AI spend board reporting
- State of FinOps 2026 for CFOs
- Inference vs training spend
- AI capitalisation: capex vs opex
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, rising to $3.49 trillion in 2027. Enterprise AI spend specifically - what companies buy 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 $407B figures?
The $2.59 trillion is total worldwide AI spending, most of which is infrastructure: servers, data centres, and silicon, forecast to exceed $1.43 trillion on its own. The $407 billion is enterprise AI spend - what buying organisations spend on AI services, software, and consumption. Only the second figure 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. Financial services combines high spend with high adoption, which means competitors are not experimenting - a laggard there is behind in production, not behind in pilots.
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 can be true: the total grows while individual programmes are reallocated toward those with demonstrated value.
How should a CFO use a market forecast like this?
For direction and context, not for a budget number. It is useful for board framing, for showing whether your growth rate is unusual, and for anchoring a vendor conversation. It cannot tell you what your own AI spend should be, because it contains no unit of your business.
What does the infrastructure share imply for buyers?
AI infrastructure runs from $975.6 billion to $1.43 trillion, over 45% of the total. That capital has to earn a return, which means sustained pressure to raise consumption. Plan for pricing that rewards volume, and price your budget in consumption rather than assuming falling unit rates lower the bill.