AFQY | A Few Quiet Yarns
← All newsAI: more stories in this category

More Agents, Bigger Bills, the Same Profit Question: Enterprise AI Heads Into 2027

13 November 2026· AFQY News

More Agents, Bigger Bills, the Same Profit Question: Enterprise AI Heads Into 2027
Image: Carl Lender, CC BY 2.0, via Wikimedia Commons

Every spring and autumn the world’s big technology research houses publish their surveys and their forecasts, and most years they disagree about something. This year they mostly agree. AI agents are moving into everyday business processes quickly, the cost of running them is climbing, and the profit most organisations were promised is, for the majority, still on its way.

McKinsey, KPMG, Forrester, IDC and Gartner have all published new numbers since August. Read together, they describe a market that has stopped asking whether to use AI and started asking what it costs and what it returns. New Zealand’s own figures, from Datacom’s 2026 State of AI Index, tell the same story at home.

This is what each found, grouped by the question it answers, for organisations setting their 2027 budgets. We wrote about the first stage of this shift in July, in The applause for AI pilots has stopped. The new numbers show how far it has moved since.

TL;DR

  • McKinsey’s 2026 State of AI survey found 44 percent of respondents scaling AI, up from 38 percent a year earlier, while the share reporting any EBIT impact from AI stayed flat at 37 percent.
  • Forrester expects AI to account for 20 percent of business technology spending in 2027, and expects more than half of CIOs to miss their AI return targets.
  • Gartner forecasts worldwide IT spending of US$7.35 trillion in 2027, up 15 percent, and says memory prices more than doubled in the past year.
  • Cyber security agencies, including New Zealand’s NCSC, now treat AI token limits as an operational resilience risk, and warn of attacks designed to run up an organisation’s AI bill or exhaust its allowance.
  • In New Zealand 91 percent of organisations use AI, but only 15 percent are scaling it across the organisation and only 20 percent have a dedicated AI budget with clear cost tracking, Datacom found.

Scaling is up. Profit is not.

McKinsey’s tenth annual survey, The State of AI in 2026, carries the subtitle On the Road to ROI, and the road turns out to be long. The share of respondents scaling AI rose to 44 percent from 38 percent. Among large companies, the share scaling AI agents rose from 27 percent to 40 percent. Yet the share attributing any EBIT impact to AI stayed flat at 37 percent, and the group McKinsey counts as high performers stayed at about 6 percent. About one in five respondents said the cost of running AI had held back how much they used it.

KPMG’s quarterly pulse of US business leaders, published in September, points the same way. Of 314 leaders at companies with revenue of US$1 billion or more, 62 percent are building, deploying or developing agents, up from 53 percent a quarter earlier. Nearly six in ten report measurable business value, and 37 percent report that value in the form of stronger financial performance.

Forrester’s 2027 predictions put a number on the pressure. It expects AI to account for 20 percent of all business technology spending next year, and forecasts global technology spending of US$6.1 trillion in 2027. Its release does not break the 20 percent down, so it should not be read as the share of a typical organisation’s budget. Forrester also expects more than half of CIOs to miss their AI return targets. According to its research as reported by The Register, only 22 percent of AI decision makers saw a positive EBITDA impact in the past 12 months, while 45 percent need a positive return within a year to count the investment a success.

Forrester expects some of the most visible deployments to go backwards. It predicts that at least three well-known enterprises will reverse a flagship AI deployment or push it back a year, because business units carry accountability for AI without the power to enforce policy in the software their agents run.

Agents everywhere, depending on how you count

IDC’s FutureScape 2027, published at the end of October, gives the most striking adoption figure of the season. In its July 2026 survey, 95.4 percent of enterprises had AI agents in production, in an average of 11 workflows, with a monthly inference budget of about US$118,000. IDC predicts that 60 percent of the world’s 2,000 largest companies will run most of their processes AI-first by 2029, up from 13 percent in 2026. It also predicts that 60 percent of enterprises will fail to turn their data into measurable value by 2028.

Set that beside two other numbers and the definitions start to matter. Forrester predicts that fewer than 10 percent of enterprises will use autonomous agents that work across applications. Datacom found that 13 percent of New Zealand organisations use agentic or autonomous systems. The three are not contradicting each other so much as counting different things: an agent that handles one step inside one workflow, against an agent trusted to act across a business’s systems on its own. When a vendor or a survey quotes an adoption figure, the first question is what it counted.

The bill

Gartner forecast in October that worldwide IT spending will reach US$7.35 trillion in 2027, up 15 percent. Gartner distinguished VP analyst John-David Lovelock told CIO Dive that memory prices had more than doubled in the past year, and that the surge, which Gartner calls memflation, should last through 2027 and probably to 2030.

"As much as CIOs are worried about the cost of AI today, it's not going to be this low again."
John-David Lovelock, Distinguished VP Analyst, Gartner

His reasoning: right now, he said, there is more capacity than demand, so model makers are giving away premium models and subsidising token prices to build usage.

New Zealanders have already felt the memory squeeze at the till. Consumer NZ reported in October that RAM costs about three times its 2025 price, that a 2TB solid state drive which cost about $300 in September 2025 now costs more than $1,200, and that the cheapest MacBook Air rose from $2,199 to $2,599 in mid-2026. The three big memory makers, it said, do not expect the pressure to ease before 2028. Those are shop prices, not enterprise quotes, but they show which way the cost of hardware is heading.

The software side is moving too. IDC expects the share of enterprises whose agentic AI bills rose without any published price change to grow from 20 percent today to 55 percent by the end of 2027. Forrester predicts that charging for AI by consumption will backfire on three major business application vendors, as buyers build their own agents on platforms they already license.

When the meter becomes a risk

In October the cyber security agencies of the Five Eyes countries, including New Zealand’s National Cyber Security Centre, published joint guidance on managing AI token consumption. It treats token limits as an operational resilience risk: if an organisation’s allowance runs out, the services that depend on AI can slow or stop. It describes two kinds of attack: denial of tokens, where an attacker uses up an organisation’s allowance, and denial of wallet, where the aim is to run up its bill. As an example, it suggests alerts as usage reaches 70, 80 and 90 percent of a quota, and allocating tokens by how critical each use is.

Gartner’s strategic predictions, released in September, go further. By 2028, it expects 60 percent of the Global 500 to control AI spending at the point where models are run. By 2029, it expects 60 percent of organisations deploying AI to have a function that ties the total cost of AI to the value it returns. And by 2030, it expects 80 percent of organisations with public-facing AI to have suffered what Gartner calls a cost exhaustion attack, in which attackers deliberately drive up AI usage to inflate costs.

Some organisations have started. KPMG found that 74 percent of the US leaders it surveyed now include a cost review when they approve an AI project, up from 61 percent a quarter earlier, and 43 percent set usage or token budgets.

Closer to home

Datacom’s 2026 State of AI Index, released in September, found that 91 percent of New Zealand organisations use AI, up from 87 percent in 2025 and 66 percent in 2024. Most are still early: 81 percent are in exploratory or implementation stages, and 15 percent are scaling AI across the organisation, up from 12 percent. The share saying AI has transformed their core operations fell from 8 percent to 4 percent.

On returns, a third said the benefits of AI had exceeded its costs, and 30 percent said the two were broadly balanced. On cost, only 20 percent have a dedicated AI budget with clear cost tracking, and one in five have not yet addressed how they manage AI costs at all.

Put the local and global numbers side by side and the shape of 2027 is clear enough. Use is close to universal. Scale is still the exception. The bill is rising, partly from hardware and partly from consumption nobody priced in advance. The organisations best placed when the budget questions arrive will be the ones that can already say what each AI workflow costs and what it returns.