The AI Industry in 2028: Where the Money Is Going

I build AI products for clients and run engineering at Avenuor, so I read these forecasts with one question in mind: what changes for the people building and buying AI systems? Here is what the numbers say and how I interpret them.

Total spending: from $1.5 trillion to over $3 trillion

Gartner's forecasts have risen through 2026. In September 2025 it said worldwide AI spending would total $1.5 trillion in 2025. Its January 2026 forecast put 2026 at $2.5 trillion. By September 2026 it had raised that to $2.7 trillion, a 49.5% increase over 2025, with the release's forecast table projecting about $3.6 trillion for 2027.

Two things are worth noting. The forecast was revised upward twice in one year, which tells you how fast commitments are moving. And Gartner reclassified categories in September 2026, separating cross-functional agents and assistants from AI software, so year-to-year segment comparisons need care.

Where the money goes by segment

Gartner's September 2026 forecast for 2026 breaks down roughly like this (source):

Shares are my calculation from Gartner's table. Gartner notes that hyperscalers and service providers buying AI-optimised servers remain the largest single area of spending.

IDC tells a consistent story from the infrastructure side. It raised its 2026 forecast for AI infrastructure spending to $497 billion, about 56% year-over-year growth, and projects $1.08 trillion in 2029 and $1.21 trillion in 2030. IDC and Gartner define infrastructure differently, so their totals are not directly comparable, but both show the same direction.

Why infrastructure dominates

Every model call runs on hardware someone had to buy, power and cool. Training frontier models takes enormous clusters, and serving millions of users takes more. Until those data centres exist, nothing above them can scale, which is why the largest cheques in 2026 go to servers, chips and cloud capacity rather than to applications.

For most businesses this layer is invisible: you rent it through a model API or a cloud provider. What matters is the knock-on effect. Heavy infrastructure investment is what pushes the cost per unit of AI capability down over time, and that is what makes new use cases affordable for smaller companies by 2028.

Where growth is fastest

Infrastructure is the biggest bucket, but not the fastest-growing one everywhere. Gartner's 2026 segment forecasts show where momentum is:

• AI platforms and models: forecast to grow 63% in 2026.
• AI-optimised cloud infrastructure (IaaS): forecast to grow 96% in 2026.
• Securing AI: a market Gartner expects to reach $4.8 billion in 2027. Small in absolute terms, but it exists because agents now take real actions.

Investment: record money, concentrated

Stanford's 2026 AI Index reports that total corporate AI investment reached a record $581.69 billion in 2025. Private investment grew 127.5%, generative AI more than doubled and captured nearly half of all private AI funding, and the number of newly funded AI companies rose 71%. The United States committed 23 times more private AI investment than China, though the report notes China's state-directed spending is likely understated.

The same chapter reports that organisational adoption reached up to 88% of surveyed organisations in 2025, up from 55% in 2023.

Adoption is wide, value is narrow

High adoption figures come from surveys of larger organisations. Across the whole US economy, the Census Bureau found only 19.8% of businesses used AI in any business function as of May 2026, with small firms well below large ones.

Value is narrower still. McKinsey's 2025 State of AI survey found that while 88% of organisations use AI somewhere, about 39% report any enterprise-level EBIT impact, and most of those say it is under 5%. That gap between spending and measured return is the most important number in this post, and I cover how to close it in how to measure AI automation ROI.

The risks to watch before 2028

Infrastructure overbuild

In October 2026, Gartner warned that over $3.5 trillion of AI infrastructure spending could be exposed to impairment by the end of 2030 if capital spending continues at its current pace. If that happens, it is mainly a problem for the companies building data centres. For AI buyers, excess capacity tends to mean cheaper compute.

Falling unit costs, rising total costs

Gartner predicts that by 2030, inference on a one-trillion-parameter model will cost providers over 90% less than in 2025. It also predicts inference costs per agentic workflow will increase more than fivefold through 2028, because agents make many more model calls per task. Both can be true: each call gets cheaper while each workflow uses far more calls. Cost controls belong in the architecture from day one.

Cancelled agent projects

Gartner also expects over 40% of agentic AI projects to be cancelled by the end of 2027. That is the flip side of record spending: a lot of it will go to projects that never reach measured value.

What this means for businesses building with AI

• Expect capability to get cheaper per unit. Do not over-engineer around today's prices. Build so you can swap models, and route simple tasks to smaller models. More in small language models and on-device AI.
• Budget for services and integration, not just licences. AI services is the second-largest segment for a reason: connecting models to real data and workflows is most of the work. Agents are also changing how the software itself is bought; see AI agents vs SaaS.
• Put a number on every project before you start. Pick one metric, measure the baseline, and agree on what success looks like. The spending boom will not rescue a project with no defined outcome.
• Treat security as a line item. Agents that can send emails or change records need permission checks, approvals and logs. The emerging securing-AI market exists because many teams skipped this.
• Plan for skills, not just spend. The money only turns into working products if someone can build with it. I cover which roles grow and which shrink in AI jobs in 2028.

Work with me

If you are deciding where your own AI budget should go, I can help you scope projects with a measurable outcome and build them to production. See AI integration and AI agent development, or book a scoping call.