Business Projections and Actual AI Use Nationwide

As of July 2026, 21.7% of U.S. businesses reported using artificial intelligence (AI), up from 17.7% at the start of the year. Roughly one in five American businesses has now adopted AI in some part of its work, and the number continues to rise.

Adoption is concentrating in larger firms, a select group of industries, and specific cities. At the same time, there is a gap between how much businesses predicted they would use AI and how much they actually did; they fell short of their own expectations.

21.7% of U.S. Businesses Now Use AI, but the Pace Is Moderating

On its own, the national trend reads as a clean growth story: between January and July, AI use rose 4 percentage points, from 17.7% to 21.7%, while the share of businesses saying “no” fell from 71.1% to 67.5%. Adoption is broadening across the U.S. business base.

U.S. businesses’ initial estimations didn’t deliver. While in January their growth expectations for the next six months were 4.7 percentage points, in July, that number shrank to only 2.8 percentage points for the following 6 months. Also back in January, 22.4% of businesses forecast that they would be using AI by July, but the actual figure landed at 21.7%: a 0.7-percentage-point shortfall.

Businesses Keep Overestimating Their Own Forecasts

Companies in 28 of 48 states used less AI in July than they had projected back in January, and so did 14 of the 25 largest metros. Most company-size groups undershot as well (that 0.7-percentage-point shortfall). The pattern suggests that six months ago, businesses collectively overestimated how quickly they could put AI to work.

The widest gaps are among mid-to-large and multi-market organizations. Firms of 100–249 employees fell 8.7 percentage points short of their own forecast, the largest gap of any cohort, while multi-sector and typically large companies missed by 7.2 percentage points. At the state level, New Mexico (−17.5 percentage points) and Nebraska (−12.4) recorded the largest overestimates. This counterweighs the adoption headline: enthusiasm met the reality of on-field adoption.

Tampa’s AI Usage Surges; Phoenix Shows the Biggest Drop

When looking at the country’s 25 largest metros by population, 15 beat the national average of 21.7% AI usage. Only five saw a decline in adoption, but the swings within that group are quite dramatic. Tampa increased 14.9 percentage points to 32.0%, the highest rate of any major metro. Detroit expressed the lowest AI usage back in January, but by July, climbed 11.4 percentage points to 24.5% — the second greatest change after Tampa.

The reversals cluster in the Sun Belt and mid-Atlantic regions. Once the leading metro back in January with 30.1%, Phoenix had the biggest decrease in AI usage, down 8.6 percentage points to 21.5%. Baltimore (−4.8), Houston (−4.2), Portland (−1.7), and San Diego (−0.5) are also cooling. As for the country’s two biggest metros, New York (17.9%) and Los Angeles (19.7%), both sit below the national average.

AI Use by Business Size and Sector

The Largest Firms Use AI at Almost Double the Rate of the Smallest

AI use climbs in tune with headcount, from 20.8% among the smallest firms (1–4 employees) to 38.8% among those with 250 or more. Every step up the size ladder adds adoption, and the largest companies are now nearly twice as likely to use AI as the smallest.

When it comes to meeting their projections, many cohorts, especially large ones, fell short. Even so, there is one that stands out. Mid-large firms of 100–249 employees look like the most hesitant group in the economy: their usage grew to only 30.5% compared to their projection of 39.2% (−8.7 percentage points). This is the size group showing the most significant pullback of forward-looking expectations, reducing their growth expectations from 11 percentage points back in January to 6.3 percentage points in July.

Inside the Sectors: Information Leads at 43.7%

Among the office-based sectors most relevant to how and where people work, the Information sector leads decisively at 43.7%, with Professional, Scientific & Technical Services (37.6%), and Finance & Insurance (36.0%) close behind. The Information category covers software publishers, telecommunications and data-hosting providers, and media, broadcasting, and internet-publishing firms — businesses whose core work is producing and processing information, which is exactly what today’s AI tools are built for. Real Estate sits further back at 29.0%, and Health Care at 24.7%.

Finance & Insurance is the most confident sector in the survey: it registered the only jump in forward-looking optimism among sectors, increasing the growth expectations from 2.5 percentage points in January to 5.7 percentage points in July. At the other end of the spectrum, multi-sector companies cooled down the most, from 12.6 percentage points back in January to only 4.6 in July. Not to mention they undershot their January forecast by 7.2 percentage points, the widest margin of any sector.

Size Meets Sector: Where Adoption Concentrates

Layering the two divides on top of each other reveals where AI use is densest. Among the largest Information-sector firms (250+ employees), 77.0% use AI — the highest rate anywhere in the data. Real Estate shows one of the biggest AI adoption differences between company sizes from any industry, ranging from 28.7% at the smallest firms to 52.4% at 100–249 employees.

A deeper look by industry challenges the assumption that bigger companies are always ahead in AI adoption. In Finance & Insurance, adoption among firms of 100–249 employees climbed 10.9 percentage points between January and July (to 62.7%), while at firms of 250+, it dropped 10.2 percentage points (to 42.2%). Two cohorts that began the year nearly equal ended up 20 points apart six months later.

How Businesses Use AI, and Why Some Don’t

The Function Gap: AI Clusters in the Front Office

Among businesses participating in the survey, Sales & Marketing is the field where AI is most widely used (14.3%), ahead of Strategy & Business Development (12.4%) and Research & Development (11.2%). Asked about the next six months, businesses expect that same ordering to hold: Sales & Marketing tops the list at 62.8%, with Strategy & Business Development (57.4%) and Research & Development (48.3%) close behind. Even functions where AI is barely present today, like human resources and quality control, expect to expand their use.

These “now” and “next six months” figures come from two separate questions answered by different groups. The current-use figures cover all the firms, including those that do not use AI in any function, while the six-month projections only include those that are planning on using AI in the next six months. This group separation explains the wide gap between current and projected use.

Additionally, because the two percentages rest on different bases, we can point to the direction of intent, not the precise path it will take: to the customer-facing and knowledge-work functions that define office life first.

What Employees Do with AI: 85% Write, 50% Search

Among firms whose staff used generative AI, 85.4% applied it to writing or editing documents, emails, and communications. That is by far the most common use. Searching for information (49.9%) and summarizing or analyzing documents (44.6%) followed. Overall, 20.8% of businesses said their employees had used generative AI to assist with work-related tasks.

43.7% Say AI Supplements Workers; Just 2% Cut Jobs

The supplement also answers the question: how is AI impacting headcount and operations at these businesses? The answer is more augmentation than replacement. 43.7% of firms say AI supplemented or enhanced a task done by an employee; only 10.1% say it performed a task a person used to do, and 10.6% used it to introduce an entirely new task. Where AI did replace human tasks, 70.9% of firms said it replaced only a small number.

On jobs, the fear-versus-reality gap is wide. 95.7% of AI-using businesses report that it had no impact on total employment whatsoever; in fact, 2.0% of businesses saw a decrease, while 2.3% got an increase. Strikingly, the heaviest adopters were the most likely to be hiring: for 3.0% of firms with 250+ employees and 4.5% Information sector firms, AI usage led to an increase in total employment. Also, the adoption of AI required little operational changes or investments, with 64.3% of AI-using firms making none at all. The most common changes were developing new workflows (15.4%) and training existing staff (15.0%), with only 1.3% of firms hiring staff specifically trained in AI.

The Barriers: Why 6 in 10 Future Non-Adopters Say “Not Applicable” 

The future non-adopters are the businesses that are planning on not using AI in the next six months. For them, the main reason for that decision is neither fear nor cost but relevance. 61.6% of future non-adopters say the technology simply does not apply to their business. Behind that sit a knowledge gap (22% cite limited understanding of what AI can do), privacy and security concerns (20.7%), and the sense that the technology is not mature enough yet (13%).

The barriers also shift with size. Smaller firms are far more likely to say AI is “not applicable” (63.3% at 1–4 employees) than the largest firms (42.2% at 250+). Instead, larger firms cite privacy concerns and the fear that the technology is not yet mature enough.

The Case for Staying Flexible

Read together, the numbers point to unevenness. AI adoption is concentrating in larger firms, in a select group of knowledge industries, and in specific talent-dense cities, while much of the map moves slowly and some of it moves backward. Even the businesses leaning in are recalibrating how quickly the payoff will come, and pointing that ambition first at the customer-facing, knowledge-work functions that define office life.

For all its revolutionizing potential, AI is still a relatively new technology that continues to advance unevenly and at an unpredictable pace. In this context, flexibility becomes an asset.

When a company cannot yet know whether AI will reshape its workflows in six months or six years, committing to fixed overhead is a hard bet to place. Industry research points the same way: JLL notes that “flexible space gives organizations the real-time optionality to adapt as AI reshapes team structures and headcount needs,” while a Cushman & Wakefield report found that 55% of global occupiers were already using flexible office solutions in 2025, with 17% planning to increase their use.

According to Luke Robinson, Regional President, North America for WeWork, “the companies navigating AI best are the ones that keep their options open. They want to be able to add a team, open a market, or reshape a floorplate without signing up on a decade-long lease. Flexible space is how they buy that optionality while the picture is still coming into focus.”

In a data set defined by how differently the economy is adopting AI — by size, by sector, by geography, and by function — the ability to stay flexible while the picture clarifies may be the most broadly useful advantage of all.

---

This report is a product of WeWork Business Intelligence, the company’s data and research engine for understanding the complexities of the modern workplace. By combining global real estate trends with real-time market data, Business Intelligence gives leaders the evidence they need to navigate the changing world of work and deploy flexible, high-performing workspace strategies for the future.

---

Methodology

• Figures are drawn from the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS), comparing the publication of 16 July 2026 with 15 January 2026, alongside the BTOS AI Supplement (fielded 17 November 2025 – 8 February 2026; published 23 April 2026).
• The main survey asks two AI questions: whether a business used AI in any function in the last two weeks, and whether it expects to over the next six months. The supplement adds detail on the functions AI is used in, the tasks employees apply it to, its effect on employment, and the reasons non-adopters give.
• Analysis focuses on office-relevant sectors. Sectors and states with suppressed data (including Alaska and Rhode Island in the main state comparison) are excluded where noted, as are size-within-sector cells the Census suppressed for confidentiality.
• “Forecast miss” compares July usage against what the same group projected for this window six months earlier. Some supplement questions allowed multiple responses, so shares can sum to more than 100%.