Forrester puts AI at a fifth of 2027 tech spend, and services pay for it

Forrester published its 2027 predictions on 21 October with a number that frames next year’s technology budgets: AI will account for 20% of all business technology spending. The analyst firm forecasts global technology spending of $6.1tn (£4.6tn) in 2027.

“AI will command an unprecedented share of technology budgets in 2027, intensifying pressure on leaders to prove its value while maintaining control over costs, infrastructure, and risk,” said Sharyn Leaver, chief research officer at Forrester, in the firm’s announcement.

Brian Hopkins, Forrester’s vice-president for its emerging tech portfolio, set out the detail behind the number in a post published the same day. Most of the AI money buys servers and equipment, he writes, but it reaches companies as AI tokens and credits. Only 53% of technology leaders expect next year’s budget to beat inflation, according to Forrester, so nearly half will lose real purchasing power while their businesses ask for more AI and pay more for it.

Forrester’s predictions are written about enterprises, but the arithmetic reaches further down. Much of a mid-market firm’s technology spending goes to outside suppliers: a managed service provider for day-to-day IT, consultancies for projects and per-seat subscriptions for software. If the AI line grows faster than the budget, Forrester’s forecast says which of those lines pays for it.

Where the extra money comes from

“Third-party services will take the hit,” Hopkins writes. Technology leaders already plan their deepest cuts in services, according to Forrester, which adds that AI is steadily lowering the cost of delivering consulting and managed services. Hopkins sets out Forrester’s recommendation: “Open your services contracts and your seat-based software renewals six months early.” The money recovered, he writes, should go into planning for AI consumption costs and context engineering, the work of feeding AI tools a company’s own information.

That puts two groups of suppliers under pressure at once. Managed service providers and consultancies face clients who expect AI to make their work cheaper, and who now have an analyst forecast saying it already does. Software vendors that charge per seat face renewals negotiated alongside usage-based AI bills that sit outside the subscription. The AI line in a 2027 budget is increasingly a metered consumption cost, while the lines it is drawn from are fixed-term contracts with notice periods.

Half the tokens, 5% of the bill

According to The Register’s report on Forrester’s AI predictions, the firm expects open-weight models, which companies can download and run themselves, to handle half of AI tokens in 2027 but take only 5% of enterprise spend. Hopkins’s post describes the same call as open-weight models doing “a large share of enterprise AI work on a sliver of the spend”.

Forrester names Gemma, Llama and Qwen as models capable enough for the work that generates the most tokens: document processing, customer support conversations, internal knowledge search and the routine steps inside agentic workflows, The Register reports. Frontier models, according to the same report, are kept for the smaller number of tasks where errors are expensive.

For a mid-market firm, that split changes what an AI bill is made of. Most of the volume sits in routine work such as invoices, support tickets and document searches, and Forrester’s prediction is that this work moves to models priced at a fraction of the frontier rate. The rest of the spend goes on the few tasks where a wrong answer costs money.

The harness around the model

Hopkins’s third published prediction concerns the software around the model. Forrester expects the harness around a model, rather than the model itself, to drive three-quarters of AI platform decisions in 2027. The harness is the software layer that directs work, assembles context, controls which tools an agent can use and checks its results, and Hopkins writes that “accuracy and running cost both get settled in that layer”. Platform vendors have already begun offering the choice of model as a drop-down setting, according to Hopkins.

If the model is interchangeable and the cost is settled in the runtime and in the company’s own data, the valuable work is integration: connecting AI tools to a firm’s systems and keeping the context they draw on accurate. Managed service providers and consultancies already sell that work. Forrester describes a squeeze on the old services line and recommends new spending on the context engineering those same suppliers could deliver.

Forrester also predicts that at least three well-known enterprises will reverse a flagship AI deployment or push it back a year in 2027, because the business units held accountable for AI have no way to enforce policy inside the agent software they use. Over half of generative AI decision-makers already name ‘agentic sprawl’ as a challenge, according to the firm.

The 2027 budget Forrester describes is mostly a reallocation. AI takes a fifth, inflation takes the growth for nearly half of technology leaders and services contracts make up the difference. The suppliers on those contracts are being asked to fund the AI line at the moment AI makes their own work cheaper to deliver, and the context engineering Forrester wants the savings spent on is integration work that some of them already sell.