Goldman Sachs tracks AI-driven job losses and warns of falling demand

Goldman Sachs has started tracking monthly how many jobs disappear because of AI and how many return thanks to tools that augment people rather than replace them. The first figures look worrying: automation is already producing a steady net loss in employment, and behind it lies a more uncomfortable question: what happens to demand when a growing number of people have less money?

What the numbers show

According to Goldman Sachs, about 25,000 jobs disappear every month because AI directly replaces a human worker. Some of the losses are offset by "augmentation" models, where an employee isn't laid off but gets tools that speed up their work. But the scale of the offset is still noticeably smaller: only about 9,000 positions come back. As a result, the economy sees a net reduction of roughly 16,000 jobs a month.

Over a year, that works out to about 192,000 net losses. These are no longer one-off reshuffles inside individual companies but a recurring stream of cuts that can be tracked as a separate macroeconomic indicator. Importantly, this is not a forecast for the distant future but monthly dynamics that banks and researchers are starting to see right now. In other words, changes in the labor market are becoming systemic rather than episodic.

• 25,000 jobs are lost to direct automation
• 9,000 positions return through work "augmentation"
• The net reduction is 16,000 jobs a month
• Annualized, that is about 192,000 losses
• Chief financial officers expect up to 502,000 AI-related layoffs in 2026

Surveys of chief financial officers, cited by the NBER and the Federal Reserve, add an even harsher layer to this data. They expect 502,000 jobs to disappear because of AI in 2026 alone. For comparison, 55,000 such layoffs were recorded in all of 2025. If these estimates hold, this is not a smooth transition but a nearly ninefold acceleration.

The problem isn't the robots

Discussion of AI-driven cuts usually boils down to efficiency: businesses cut costs, processes get faster, margins grow. But this logic has a weak spot. A salary is not only a company expense but also someone's future payment elsewhere in the economy. A person who has been replaced spends less on housing, food, transport, subscriptions, education and electronics. For a single firm that is a saving; for the whole system it is a contraction of purchasing power.

If you remove workers, who will buy everything you sell?

This question sounds especially sharp against another trend: money and income are concentrating at the top. According to data from the first part of the series, 21.5 trillion dollars have concentrated in seven companies, and income from capital grew 12 times faster than wages. This means the gains from automation are not going where they turn back into mass consumption, but where they more often settle in assets and financial instruments.

Where the balance is shifting

The very idea of "augmentation" shows that AI doesn't have to be a layoff machine. In many roles, a model can strip out routine, speed up analysis, prepare drafts, and help with customer support or information search. But for that, a company must deliberately build processes around the employee rather than around replacing them. So far, the numbers suggest the replacement scenario is developing faster than the augmentation scenario, especially in standard office functions.

For business, this creates a double risk. In the short term, cuts do improve financial reports. But in the medium term they may hit the sales market, especially in sectors tied to the mass consumer. If AI raises productivity but at the same time lowers total household income, the economy ends up with a paradox: producing gets easier, selling gets harder. And this gap can widen even while the companies themselves temporarily gain in profit.

What this means

The AI story goes beyond HR and corporate efficiency. If automation keeps removing jobs faster than it creates new roles and augments existing ones, the main constraint on business will be not supply but demand. Put differently, companies will have ever more ways to produce and ever fewer customers who can afford the output. This is a question not only of technology but of the resilience of the entire consumer economy.