Enterprise AI Spending per Employee Is $1,024—the Mean and Median Differ by an Order of Magnitude

Venture capital firm a16z's chart of the week offers an eye-catching number: among companies using AI, the average spend on AI is $1,024 per employee per year, up 96.1% year over year. The chart breaks this down by industry and company size. Retail, wholesale trade, and transportation had the highest increase, +130.7%; companies with 99 to 249 employees had the highest increase, +141.5%. Companies with more than 250 employees, by contrast, saw only +88.4%. The sample is 880 AI-using companies from the Atlanta Fed's Business Uncertainty Survey, in October 2026.

Don't rush to use this number for budgeting. The questions in the comments are more interesting than the chart. The tweet's main text doesn't make clear whether the $1,024 is per month or per year, and someone asks directly, is it monthly. One commenter converts it to an annual basis, about $85/month, higher than a Microsoft license; on a monthly basis, it doesn't match common sense. Another commenter, Joshua Pi'Rwot, dug up the Atlanta Fed's own article: $1,358 per person in 2025 and $2,068 in 2026, employment-weighted and winsorized at the 99th percentile; the un-winsorized 2026 figure is $3,108. Ramp's median is $15.70/month, about $188/year. Putting these numbers together at least shows that the definitions differ. The mean describes the companies spending most aggressively; the median is closer to an ordinary company.

Another controversy is the denominator. One commenter says that one company has a human team, while another runs 11 agents and has zero employees; AI spending isn't in the software budget—it's basically payroll. Averaging per head flattens out this structure. Another one-person company says that if you max out subscriptions, $1,024 is spent in a little over a week. Seats bought for employees and API usage-based billing are also two different ledgers. If no one opens the seats, the mean gets pushed up; if the API budget blows up, developer spending will be far higher than that of ordinary employees. Someone in the comments asks how much is unused seats and how much is API. That question is closer to reality than total spending.

Someone in the comments also posted a video, with text to the effect that if the goal is to save money, you'd be better off not adopting AI in the first place. The reaction is blunt, and it hits the gap between spending and returns.

AI spending is moving from pilot budgets to per-head budgets at no small speed. But the per-head average metric shows neither utilization nor cost per task. A more practical approach comes from the comments: pick two tasks, time them before and after, and only increase the budget if the minutes saved can cover the subscription cost. This suggestion isn't flashy, but it's more useful than the average.