Generative AI is changing how startups scale

For years, young companies have tended to follow a familiar script: they added headcount as they raised capital. Employment growth was not success in and of itself, but it served as a key signal of it to investors looking for evidence of a startup’s momentum.

Now, fresh research by UNC Kenan-Flagler Business School professors shows that generative AI is weakening that long-standing link between startup growth and employment.

Research by Finance Professors Abhinav Gupta, Elena Simintzi and Franklin Qian at UNC Kenan-Flagler and PhD student Yifan Sun of Penn shows that generative AI is changing how startups scale, allowing them to become more productive, raise money and reach key milestones with fewer workers.

“When we thought about startup success so far, we thought about startups scaling up,” Simintzi says. “In other words, a startup that could add employees was considered successful. With generative AI, startups are both leaner and successful.”

The study, which analyses nearly 95,000 U.S. startups founded between 2018 and 2021, uses the release of ChatGPT in late 2022 as a natural experiment. Their findings point to a sharp decoupling of growth from headcount.

Fewer people, more output

Startups most exposed to generative AI — meaning a larger share of their work involves tasks that AI tools can perform — reduced employment by about 8% after ChatGPT’s launch. The decline emerged within two quarters and has since then persisted.

What is more, those same companies became markedly more productive. Output rose even as headcount fell, as tasks previously done by displaced workers could be more efficiently done with the help of generative AI. This productivity gain translated to stronger startup fundraising. It seems headcount growth is no longer a prerequisite for momentum.

Who gets replaced — and who stays?

The job losses are not evenly distributed.

Junior, execution-heavy roles have borne the brunt of the losses, with senior and managerial positions less affected.

As a result, startups are holding on to more experienced staff even as overall headcount shrinks — a kind of selective downsizing.

“If your job entails a lot of execution, you’re more likely to be replaced by the technology,” Qian says. “In contrast, if you’re a manager, then you’re less likely to be substituted.”

For workers who are displaced, the adjustment is painful.

“There is a real cost for these workers,” Qian says. “About half remain unemployed for up to six months, and many who do find work move into lower-paid roles that are less exposed to AI.”

More startups, not fewer jobs

At first glance, the employment effects look bleak. At the aggregate level, they are far less so.

Generative AI has allowed existing startups to run with fewer people, while also lowering the cost of starting new companies.

So, more startups are being formed — particularly in industries where a larger share of work is exposed to generative AI. Sectors with greater exposure saw about a 7% increase in the total number of active startups after ChatGPT’s launch.

“On the bright side, there is no negative employment effect on aggregate,” Gupta says. “This is because we see more startups being created, and these need fewer workers, so there is no overall change in employment.”

In other words, jobs lost at incumbent startups are broadly offset by jobs created at new ones. The labor market churns, but it does not shrink.

Venture capital adapts

Investors have started to rethink how they deploy capital.

Some venture capital firms are spreading their bets more widely, backing more companies but writing smaller cheques. Among those with portfolios concentrated in industries most affected by generative AI, initial funding rounds are about 12% smaller on average, while the number of investments rises about 8%.

“We do see VCs investing in more companies, but making smaller bets,” Gupta says.

A new definition of success

The paper’s conclusions aren’t dystopian, but they are disruptive.

“Yes, there is a cost for younger people already employed at these GenAI exposed startups,” Simintzi says. “But, at the same time, we see a lot of opportunity for startup creation.”

The research findings suggest that growth is becoming less tightly linked to payroll size and more towards measures of output per employee.

As Simintzi put it: “The message of the paper is actually a positive one. Because we can create not more jobs, but more firms.”