TL;DR: McKinsey State of AI 2026 finds that 89% of respondents’ organizations use AI and 44% are scaling it across the enterprise. Yet only 37% report positive EBIT impact, unchanged from 2025. Large companies are pulling ahead on agents, operating costs constrain one in five organizations, and reported workforce cuts fell below earlier expectations. The practical question is how to turn individual productivity gains into measured business results.

In this article

• What are the key findings of McKinsey’s State of AI 2026?
• Why isn’t AI showing up in EBIT?
• Is the cost of AI becoming a constraint?
• What do AI high performers do differently?
• How reliable is the McKinsey State of AI survey?
• FAQ

McKinsey has now run its State of AI survey for nearly a decade, and the 2026 edition, titled “The state of AI in 2026: On the road to ROI,” is the most candid one yet about the distance between using AI and earning money from it. The authors (Dan Tinkoff, Lieven Van der Veken and Michael Chui, with Tara Balakrishnan) frame it as a road. The data says most companies are still near the on-ramp.

This article is our pillar summary of the full report: every headline number, what changed since the 2025 edition, where the methodology limits the conclusions, and what CIOs, CFOs and CTOs should take from it. If you want the narrower story on coding agents replacing software purchases, read our companion piece, Agentic Coding: McKinsey’s Build-vs-Buy Findings (2026).

What are the key findings of McKinsey’s State of AI 2026?

The 2026 survey has seven findings that matter most for leaders:

1. Adoption has plateaued. 89% of respondents say their organization regularly uses AI in at least one business function, against 88% in 2025.
2. Scaling is rising. 44% say AI is scaling across the enterprise, up 6 percentage points from 38%. 56% use AI in three or more functions, up from 51%.
3. Agents are a large-company story. 40% of respondents at $1B+ organizations report scaling AI agents (27% last year). At smaller organizations it stayed flat at 22%.
4. Coding agents are reshaping IT budgets. About two in ten organizations are scaling software coding agents, and 32% have decided against buying at least one software product because they could build it in-house.
5. Costs now bite. 20% say AI operating costs, including tokens, have constrained their AI use. 60% still expect to increase AI investment next year.
6. Profit impact is flat. 37% report any positive EBIT contribution from AI, essentially unchanged from 2025. AI high performers remain about 6% of respondents.
7. Layoff forecasts overshot. 14% report AI-driven workforce declines over the past year, less than half the 32% who expected them. Even so, 39% now expect declines next year.

The rest of this article goes one level deeper on each.

How many companies use AI in 2026?

Nearly nine in ten. McKinsey reports 89% of respondents’ organizations use AI regularly in at least one function. The line has barely moved since 2025 (88%), which means the adoption question is effectively settled for large parts of the economy. The movement is now in depth: how many functions AI touches and how far each deployment has progressed from experiment to scale.

AI adoption has flattened near 90%; scaling is increasing, especially at large enterprises. Source: McKinsey’s 2026 survey.

The phase data is the more useful signal. Among organizations using AI, the experimenting share fell from 32% to 22%, piloting rose from 30% to 34%, and the share at least scaling rose from 38% to 44%. Companies are moving forward through the funnel, slowly.

A caveat on the long-run line: McKinsey changed its definition of “AI use” several times since 2017 (from “a core part of the business at scale” to “adopted in at least one function” to “regular use in at least one function”), and 2024 had two survey waves. The trend direction is reliable. Exact year-to-year comparisons before 2025 are less so.

Are large companies pulling ahead on AI agents?

Yes, and the gap widened sharply this year. 54% of respondents from organizations with at least $1 billion in revenue report scaling AI across the enterprise, compared with about one-third at smaller organizations. On agentic AI the split is starker: 40% of large-company respondents report scaling AI agents in at least one function, up from 27%. Smaller organizations stayed at 22%.

By tool, large companies lead everywhere. Among $1B+ respondents, 64% are scaling AI chatbots (39% at smaller firms), 31% are scaling software coding agents (17%), and 25% are scaling other AI agents (13%).

One interpretation is that agents reward the things large companies already have: integration teams, data platforms, security review capacity and budget to absorb token costs while use cases mature. Smaller firms can adopt a chatbot in a week. Wiring an agent into an ERP workflow with proper permissions is a different project. For the governance side of agents, see our analysis of McKinsey’s State of AI Trust 2026 report.

Why isn’t AI showing up in EBIT?

Because individual productivity gains are not automatically converting into enterprise financial results. This is the central tension of the 2026 report. 80% of respondents say AI has improved their own productivity and 50% say it helps them make better decisions. At the enterprise level, only 37% report that AI has contributed positively to EBIT, which McKinsey describes as “essentially unchanged from 2025,” despite more organizations scaling.

Self-reported individual productivity gains exceed reported enterprise-level EBIT impact. Source: McKinsey’s 2026 survey.

Senior partner Dan Tinkoff calls this gap “a striking finding.” His explanation, and the report’s, is that time saved by individuals has to be captured through workflow redesign, role changes or growth initiatives before it shows up in profit. If an analyst finishes a report in two hours instead of four and the organization does nothing with the other two hours, the EBIT line does not move.

There are two other plausible explanations the report does not dwell on:

• Measurement. Many organizations have no defined process to quantify AI impact (only about 20% of non-high performers do, per Exhibit 11). Some value may exist but goes uncounted.
• Self-report bias. “AI improved my productivity” is a perception from the respondent about their own work. It is easier to agree with than to verify.

Function-level data shows value does exist in pockets. Respondents most often report cost reductions from AI in supply chain management, service operations and manufacturing, and revenue gains in marketing and sales, product development and software engineering. Software engineering stands out for having the largest share reporting revenue increases above 10% (11%).

Is the cost of AI becoming a constraint?

For a meaningful minority, yes. One in five respondents (20%) say their organization has limited AI use because of AI-related operating costs, including token costs. The share is broadly consistent across company sizes. By industry it ranges from 25% in technology and 24% in financial institutions to 12% in the public sector and consumer goods.

Senior fellow Michael Chui’s comment in the report is the best one-line summary of why: CFOs and chief AI officers have discovered that AI “isn’t ‘too cheap to meter'” for agentic software development and complex reasoning. Per-token prices fell, but token volumes grew faster.

Spending is still rising. 28% of respondents say their organizations spend more than 10% of their total ICT budget on AI, and 60% expect to increase AI investment next year. Pharma, insurance and financial institutions are the most likely to plan increases. We covered the economics of token budgets in detail in our analysis of Accenture’s CIO guide to AI tokenomics.

Did AI cause the layoffs companies predicted?

Not at the scale predicted. In the 2025 survey, 32% of respondents expected AI to reduce their organization’s total headcount over the following year. In the 2026 survey, only 14% report that AI actually contributed to a workforce decline over the past year. Two-thirds (66%) report little or no AI-related change in total employment, and 8% report an increase.

Expectations have not cooled. 39% now expect AI to reduce headcount over the next year, while 43% expect little or no change. Most of the expected reductions fall in the 3–10% range, not mass layoffs. Service operations and supply chain management remain the functions where respondents most expect cuts.

Two readings are possible. Either companies were too optimistic about how fast AI would substitute for work, or the reductions are delayed and will arrive once workflows are actually redesigned. The 2027 survey will be the test. The survey measures reported changes inside respondents’ organizations; it does not count all announced layoffs across the economy.

Employees themselves are relatively calm. Only 13% say AI makes them anxious about their career prospects. The report does find that mid-level managers and individual contributors (47%) are more likely than executives (31%) to report at least one negative effect, such as pressure to take on more work or being overwhelmed by AI output.

What do AI high performers do differently?

They aim higher, redesign more and spend more. McKinsey defines AI high performers as respondents who attribute at least 5% of EBIT to AI and describe its value as “significant.” They make up about 6% of respondents (92 of 1,521 AI users), unchanged from 2025.

High performers differ in ambition, agent scaling and risk management. These associations do not establish causation. Source: McKinsey’s 2026 survey.

The pattern across the report’s exhibits:

• Ambition beyond efficiency. About 80% of both groups pursue efficiency gains with AI. The difference is that 74% of high performers also pursue growth objectives (47% of others) and 65% pursue innovation (49%). High performers are 3.3 times more likely to intend to use AI to fundamentally transform their business within three years.
• Workflow redesign. Nearly three-quarters of high performers report fundamentally redesigning workflows because of AI, up from 55% last year. About one-quarter of other respondents do.
• Leadership and measurement. High performers are twice as likely to say senior leaders show real ownership of AI initiatives and to have defined processes for measuring AI impact.
• Budget. They are more than twice as likely to spend over 15% of their ICT budget on AI, and more than half expect to increase AI investment by 10% or more next year (36% of others).
• Agents and coding. They are twice as likely to be scaling software coding agents and 2.7 times as likely to be scaling other agentic AI. Nearly half have skipped a software purchase in favor of building in-house, versus 31% of others.
• Risk. They are more likely to be working to mitigate AI-driven exploitation of technical vulnerabilities (56% vs 46%) and unauthorized or unintended AI actions (45% vs 33%), the two risks most associated with agents.

One finding runs against intuition: high performers report cost constraints on software coding agents three times as often as others (18% vs 6%). The likely reason is simple. They use coding agents at a scale where token bills become visible. Cost pressure here is a sign of real usage.

A methodology caution applies to this whole section. A sample of 92 is small, and the relationship is correlational. Companies already earning profits from AI may find it easier to spend more on AI, appoint visible sponsors and redesign workflows. The practices are plausible drivers, but the survey cannot prove causation.

What does the State of AI 2026 mean for CIOs, CFOs and CTOs?

The report points to five practical moves:

1. Stop counting adoption. At 89%, usage is no longer a differentiator. Track the share of AI use cases that have reached scale and the share with a measured P&L effect.
2. Put individual gains into workflows. Self-reported productivity gains will not necessarily affect enterprise earnings unless processes, roles or targets change. Pick a small number of end-to-end workflows and measure whether redesigning them around AI improves cost, quality or growth. High performers report workflow redesign more often than other respondents.
3. Treat token costs as a design constraint. Senior partner Lieven Van der Veken describes the fastest-moving companies as “treating operating costs as a design constraint, not an afterthought.” Add cost per task to every agent business case, and decide early which tasks need a frontier model.
4. Revisit the build-vs-buy rulebook. With 32% of organizations already skipping software purchases, procurement and architecture review need a framework for when coding agents make in-house builds viable. Our build-vs-buy analysis covers this in depth.
5. Govern agents before scaling them. High performers manage more risks, especially unauthorized actions and vulnerability exploitation. If you plan to scale agents, these controls should come first.

For smaller companies, the size gap is the uncomfortable message. Agent scaling at sub-$1B organizations did not move in a year. Packaged agent products from SaaS vendors may be the more realistic route than building agent infrastructure in-house.

How reliable is the McKinsey State of AI survey?

Useful for direction, limited for precision. The online survey ran from May 4 to June 8, 2026, with 1,719 participants in 97 countries across industries, company sizes, functions and seniority levels. 36% work for organizations with more than $1 billion in revenue. Responses are weighted by each country’s contribution to global GDP.

Keep four limits in mind:

• It is self-reported. EBIT attribution and productivity gains are respondent estimates, not audited figures.
• Definitions have shifted. The definition of AI use has changed over time, and in 2026 McKinsey asked about cost and revenue impact at the business-function level instead of the use-case level. Function-level comparisons to earlier years are therefore imperfect.
• The high-performer group is small. At n = 92, single percentage-point differences are noise.
• McKinsey sells AI transformation. The survey’s conclusions (redesign workflows, build operating models, invest in change management) align with McKinsey’s consulting offer. That does not make the data wrong, but it is context for how the findings are framed.

FAQ

What is the McKinsey State of AI 2026 report?

It is McKinsey’s annual global survey on enterprise AI use, published in August 2026 under the title “The state of AI in 2026: On the road to ROI.” It surveyed 1,719 participants in 97 countries between May 4 and June 8, 2026.

What percentage of companies use AI in 2026 according to McKinsey?

89% of respondents say their organization regularly uses AI in at least one business function, up from 88% in 2025.

How many companies get financial value from AI?

37% of respondents report that AI has made any positive contribution to EBIT. Only about 6%, the AI high performers, attribute 5% or more of EBIT to AI.

What is an AI high performer in McKinsey’s survey?

A respondent whose organization attributes at least 5% of EBIT to AI use and reports “significant” value from AI. About 6% of respondents qualify (92 of 1,521 AI users).

How many companies are scaling AI agents?

About two in ten overall. Among organizations with $1 billion or more in revenue, 40% report scaling AI agents, up from 27% in 2025. Among smaller organizations, the share is flat at 22%.

Are companies replacing software purchases with AI coding agents?

Yes. 32% of respondents say their organization decided against buying at least one software product or feature because it could be built in-house with agentic coding tools. The share is highest in technology (41%) and healthcare (39%).

Is the cost of AI limiting adoption?

For some organizations. 20% say AI operating costs, including token costs, have constrained their AI use. About one in ten say costs have constrained their use of chatbots, AI agents or coding agents specifically.

Did AI lead to layoffs in 2026?

Less than predicted. 14% of respondents report AI-driven declines in total headcount over the past year, compared with 32% who expected declines in the 2025 survey. 66% report little or no change.

Do companies expect more AI-related job cuts?

Yes. 39% expect AI to reduce their organization’s headcount over the next year, up from 32% a year earlier. 43% expect little or no change, and most expected cuts are in the 3–10% range.

How is the 2026 report different from the State of AI 2025?

The 2025 edition focused on agents, innovation and transformation. The 2026 edition adds new questions on coding agents, build-vs-buy decisions and AI operating costs, and shows that EBIT impact did not improve even as scaling increased.

Final take

McKinsey’s 2026 survey describes a market that has finished adopting AI and has not yet finished absorbing it. Usage is near universal, scaling is creeping up, large companies are moving fast on agents, and employees say they are more productive. The P&L has not followed, and the roughly 6% classified as high performers are more likely to have redesigned workflows. These are survey associations, not proof that any one practice causes higher profits.

For leaders, the useful question for the next budget cycle is no longer whether to use AI. It is which three workflows you will redesign end to end, and how you will measure the result.

Sources: McKinsey & Company, “The state of AI in 2026: On the road to ROI” (August 2026), by Dan Tinkoff, Lieven Van der Veken and Michael Chui with Tara Balakrishnan; McKinsey, “The state of AI in 2025: Agents, innovation, and transformation.” All figures are self-reported survey results weighted by national GDP share. Year-over-year comparisons are affected by changes in McKinsey’s definitions and question design, noted where relevant. High-performer comparisons rest on a sample of 92 respondents and show correlation, not causation.