← Writing

Note · AI economics · November 2026

Why AI isn't showing up in GDP yet

AI isn't visible in the macro figures yet. A survey of nearly 6,000 executives in the US, UK, Germany and Australia, published by NBER in February, found that about nine in ten firms saw no impact from AI on employment or productivity over the past three years. Torsten Slok, chief economist at Apollo, wrote that “AI is everywhere except in the incoming macroeconomic data.” He sees no sign of AI in productivity, employment or inflation figures so far.

Part of the reason is what firms use AI for. Most AI projects so far aim to cut costs. When a company uses AI to do the same work with fewer people or less spend, its margins go up, but the economy hasn't produced anything more. The value has moved from wages or suppliers to profit. GDP only grows if the people, time and money freed up go into new output, such as other work, new products or investment. If the savings sit in the business or go to shareholders, demand and output stay about where they were.

Doing existing tasks faster has the same limit. A team that clears its work in four days instead of five only adds to output if the fifth day goes into something new. Some of the time saved also goes back into checking and fixing AI output. A Workday survey of 3,200 employees in January found that nearly 40% of the time people saved with AI went on correcting, verifying or rewriting what it produced. The same thing happens at home. A Stanford study of 200,000 US households found that generative AI made chores like job hunting and travel planning much faster, and people spent the time saved socialising or watching TV.

Cost savings can still grow the economy. Much of the productivity growth of the last century came from automation that cut costs, from farm machinery to ATMs, once the people and money freed up went into other work. With AI that hasn't happened at scale yet, and the broad view among economists and researchers is that the main reason is a lag.

Robert Solow said in 1987 that you could see the computer age everywhere except in the productivity statistics, and the gains from computers came through in the late 1990s. Erik Brynjolfsson, Daniel Rock and Chad Syverson call this the productivity J-curve. Firms spend first on process redesign, retraining and data, that spend doesn't show up as output, and measured productivity looks flat until it pays off. Today the dip shows up in two ways.

AI does already show up in GDP in one place, which is investment. GDP counts what businesses spend on new equipment, software and buildings. So when a cloud provider buys servers and chips or builds a data centre, that spending is added to GDP when it happens, before the data centre has produced anything. Investment in information processing equipment and software is about 4% of US GDP, but Harvard economist Jason Furman calculated that it accounted for about 92% of US GDP growth in the first half of 2025. The real boost is smaller than that, because many of the chips and servers are imported and imports are subtracted from GDP, as the Congressional Research Service points out. Either way, this building work adds to growth now, while the productivity gains from using AI are still to come through.

In payments and finance operations, the projects that will show up in the numbers are the ones that change how much a team can handle.

A chatbot, or the AI built into Excel and other office tools, helps one person get through their work faster. Many finance people say it has changed how they work, and they put the time saved into analysis and decisions that matter more. That value is real, but it's hard to measure and it builds slowly, while the team's size, its process and the number of cases it closes each week usually stay the same. So it takes a long time to show up as extra output.

An AI agent works differently because it completes tasks, or whole chunks of a task, autonomously. Take a queue of chargebacks, reconciliation breaks or payment exceptions. The agent pulls the data, checks it against the rules and resolves the routine cases, and passes the rest to a person. The agent isn't fully autonomous, since people still review its work and approve what needs approving, but each case moves forward without someone driving every step. Hence the team's capacity goes up by an amount the business can measure and plan with. The same team can handle more volume as the business grows, and people spend their time on the cases that need judgement. That extra output is what eventually shows up in GDP. We wrote about how that works in What changes when an AI agent works the queue.

Sources: Yotzov, Bloom, Davis et al., Firm Data on AI, NBER Working Paper 34836 (2026); Fortune, 17 Feb 2026 (Slok); ERP Today on Workday, Beyond Productivity (2026); Blank, Schubert and Zhang, The Household Impact of Generative AI, SIEPR (2026); Brynjolfsson, Rock and Syverson, The Productivity J-Curve (AEJ: Macroeconomics, 2021); Fortune, 7 Oct 2025 (Furman); Congressional Research Service, AI Investment and the U.S. Economy.