Technology Firms Report More AI Friction Than the Sectors They Sell To
On five measures KPMG publishes side by side, the sector that builds AI reports worse conditions than the cross-sector average. Its share reporting a return at scale is lower too, on a question that is not worded identically.
Adoption Is Universal, Return Is Not
KPMG put the technology sector at the front of every input measure in its Global Tech Report 2026. Eighty-six per cent of respondents report a defined enterprise-wide AI strategy, ten percentage points above the cross-sector figure. All report active AI initiatives, and more than 90 per cent report active deployment across AI, cloud, data and cybersecurity. Seventy-four per cent invest more than US$50 million a year in digital technologies and 51 per cent invest more than US$100 million.
The money is returning in aggregate. Eighty-three per cent report realising more than US$50 million in digital value over the past 12 months and 62 per cent more than US$100 million, with AI accounting for roughly 21 to 40 per cent of total digital value.
The exception is consistency. Only around 10 per cent of technology organisations report achieving AI at scale with consistent ROI across multiple use cases. The technology cut is based on more than 155 leaders at companies with revenue of US$1 billion or more, split 37 per cent EMEA, 38 per cent Americas and 25 per cent Asia-Pacific.
What Retrenchment Looks Like in Practice
The survey does not name companies, but the past year has produced examples of what a technology firm does when an internal AI deployment costs more than it returns, and they are not pilot projects.
Microsoft cancelled most of its direct Claude Code licences after six months of widespread internal adoption, and Uber spent its entire 2026 AI coding budget in four months despite having encouraged employees to use the tools, both reported by Fortune in May 2026. Meta built an internal leaderboard to track token consumption. The mechanism is the same in each case: unit prices fall while agentic workloads consume far more tokens per task, so total spend rises even as the per-token cost drops. Gartner expects inference costs to decline 90 per cent by 2030 and enterprise AI expenses to climb anyway.
"For my team, the cost of compute is far beyond the costs of the employees," says Bryan Catanzaro, vice president of applied deep learning at Nvidia.
The pattern is not confined to the largest firms. In the 2026 State of AI Cost Governance report from Mavvrik and Benchmarkit, published in July, 62 per cent of organisations said unexpected AI costs had materially altered a business decision over the past year. Among those, a quarter delayed or cancelled an AI initiative outright, a third imposed emergency spending freezes and 40 per cent escalated to the board. Only 11 per cent said they could forecast AI costs to within 10 per cent.
The Same Survey Prices the Friction
Back in the KPMG data, the cost story is one of several. On every measure of execution friction that the report publishes with a cross-sector comparison, technology companies report worse conditions.
Seventy per cent say access to skills and talent is a constraint on turning strategy into delivery, against 53 per cent across sectors. Fifty-four per cent report employees feeling left behind, against 38 per cent. Forty-six per cent report disconnected AI projects, against 32 per cent. Leaders expect a steeper rise in cyber-risk exposure over two years, at 41 per cent against 25 per cent. Sixty-one per cent say technology plans go out of date quickly, and more than half say business strategies are frequently rendered irrelevant by unexpected technology innovations, against 35 per cent elsewhere.
Those sit beside two measures the sector leads on for good reasons: 83 per cent report the IT function leading AI implementation, against 73 per cent, and 83 per cent report employees trusting AI outputs enough to inform decisions, against 62 per cent. The report's own reading is that rapid scaling is outpacing change management and coordination in parts of the sector, and that across most technologies the largest single group of organisations sits in a maturity state it describes as funded but hitting blocks with scaling.
Where the Comparison Runs Out
The 10 against 24 comparison in the takeaways needs the caveat the report does not supply.
The two figures come from the same programme but not from identical questions or samples. The technology cut asks about organisations achieving AI at scale, with consistent ROI across multiple use cases; the parent report, covering 2,500 executives in 27 countries and eight sectors, asks about achieving ROI across multiple use cases, with no reference to scale, and puts that at 24 per cent, down seven percentage points from the previous round. Adding a scale condition should be expected to lower the number on its own. The revenue floors differ as well, US$1 billion against US$100 million, so part of the gap may reflect company size as much as sector, and technology and telecommunications make up 11 per cent of the parent sample, which means technology firms are counted inside the figure they are being measured against.
The two numbers are not a like-for-like comparison and the headline does not rest on them. What survives is the pattern underneath: five friction measures reported on the same basis as their cross-sector counterparts, all running against the technology sector, and a parent figure that is falling.
What Asia's Largest Deployments Do and Do Not Show
Asia-Pacific accounts for 25 per cent of the technology sample, roughly 39 respondents, and KPMG publishes no regional breakdown of the technology findings. Neither does the parent report, which draws 29 per cent of its respondents from Asia-Pacific. There is no published figure for how many Asian technology companies sit inside the 10 per cent.
What the region does have is disclosure from three of its largest technology employers. Infosys, TCS and Wipro have each deployed Microsoft 365 Copilot to more than 100,000 staff, according to a Microsoft case study published in June 2026. Infosys reports 91 per cent monthly active users. TCS reports 86 per cent of licensed associates using AI daily, with productivity improvements of 20 to 25 per cent in research and content production and work-cycle time down 25 to 35 per cent. Wipro reports 95 per cent monthly active usage, 7.5 million prompts a month and 250,000 full-time-equivalent days saved per quarter.
"The real opportunity with AI lies in how deeply it is embedded into everyday work," says Salil Parekh, chief executive and managing director of Infosys. K. Krithivasan, his counterpart at TCS, describes the programme as "an integral part of building AI-first culture and shaping Human + AI operating model".
Those are the largest enterprise AI rollouts anyone in the region has put numbers to, and they are published by the vendor. The case study reports adoption, usage and saved effort. It reports no cost, no net return and no difficulty, which is the half of the ledger the KPMG respondents are describing. Read alongside the Mavvrik finding that only 11 per cent of organisations can forecast AI spend within 10 per cent, high usage is not by itself evidence of a return at scale.
Deepika Giri, associate vice president at IDC, has written separately that organisations in Asia-Pacific without high-quality, AI-ready data can see productivity losses of around 20 per cent when they move past pilots, and that most do-it-yourself agentic projects stall on governance, interoperability and measurement of return.
What Would Settle It
One observation is worth making, and only one, because the report stops short of a verdict itself. KPMG has not said the technology sector is failing at AI. It has documented that the sector leads on strategy, deployment, investment and trust, that it reports more friction than any other sector on five measures, and that its share reporting consistent return at scale is the low figure in the set. Those three facts belong to the same picture.
Respondents expect AI and automation to move from 59 per cent on track or fully scaled today to 96 per cent in 12 months. Whether that is a plan or an expectation is not answerable from this report; the last time KPMG asked the cross-sector question, the number went down.
Three things will show which reading holds. Whether the ROI-at-scale figure recovers in the next edition, due to report in the first quarter of 2027. Whether KPMG publishes a regional cut, which would show if the 10 per cent is evenly spread or concentrated. And whether the 54 per cent reporting employees who feel left behind falls first, since that is the measure tied most directly to the coordination problem the report names.
Sources
KPMG Global tech report 2026: Technology and the parent KPMG Global tech report 2026, KPMG International, for all survey figures and quoted commentary; Fortune, 22 May 2026, for the Microsoft, Uber, Meta and Nvidia detail; the 2026 State of AI Cost Governance report from Mavvrik and Benchmarkit, via CFO Dive; Microsoft's June 2026 case study on Infosys, TCS and Wipro; and IDC research on industrialising AI in Asia-Pacific.
