The AI Layoffs Story Doesn't Hold Up. The Data Says Something Older.
Everyone is calling 2023 through 2026 the AI layoffs era. Four years of layoff data says something less flattering to the headline: the companies cutting the deepest weren't optimizing for AI. They were going out of business.
I pulled the full history of tech layoffs from layoffs.fyi, over 4,000 events from 2020 to mid 2026, and ran the numbers to see whether the AI narrative actually explains the pattern, or whether it's just a convenient label for something the funding cycle has always done.
The companies cutting deepest aren't the ones you'd expect
Filtered the data to companies that had raised significant capital, then sorted by how much of their workforce they cut. The top of that list isn't Meta, Google, or Salesforce making "efficiency" cuts. It's Britishvolt, Katerra, Lilium, Convoy, Cue Health, Vroom, Northvolt, Cruise, Gorillas.
Every one of them cut 50 percent or more of their staff. Several cut 100 percent. That number means the company didn't restructure. It closed.
None of the well known AI layoff headlines show up at this end of the data. The companies making the deepest cuts raised enormous rounds and then ran out of road entirely. That's not an AI story. That's the same venture failure story that's existed since long before anyone was framing headcount decisions around large language models.
Severity by funding stage barely moved after 2023
If AI really triggered a new, harsher wave of layoffs, you'd expect the average size of a layoff to jump noticeably once the AI narrative took hold in coverage.
It didn't, at least not consistently. Comparing average percentage of workforce cut before and after 2023 by funding stage:
- Seed stage: dropped from 60 percent to 51 percent
- Acquired companies: dropped from 32 percent to 26 percent
- Post IPO companies: dropped slightly, from 16 percent to 14 percent
- Series A: rose from 33 percent to 38 percent
- Series C and D: ticked up a few points
Some stages moved up, some moved down, none moved enough to look like a structural break. That's not what a genuine regime change looks like. It's what a normal, noisy funding cycle looks like when you zoom in close enough to see the variance.
The metric that broke, and what it revealed
I also tested a more direct cash pressure measure: total employees laid off divided by total funds ever raised. The idea was simple. A company cutting deep relative to how little capital it had was showing real financial strain, not strategic repositioning.
The top of that list was Microsoft, Cisco, Intel, and Google, each showing funds raised of under 30 million dollars against layoffs in the thousands.
That number is wrong, and it's worth explaining why rather than hiding it. For public companies, this dataset's funds raised figure only captures pre IPO venture rounds. It has nothing to do with a company that has a market cap in the trillions. The metric isn't measuring cash pressure for these companies at all, it's measuring a number that stopped being relevant the day they went public.
So the fix is the finding. This metric only means something for private, pre IPO companies still living off the capital they've raised. Once I restricted it that way, the picture matched the rest of the story: it's the small and mid stage companies burning through raised capital that show real financial distress, not the household names doing headline making cuts.
What this actually says
Put the three findings together and the AI layoffs narrative starts looking less like an explanation and more like a label applied after the fact. The most extreme layoffs in this dataset are companies that failed outright. The severity of cuts by funding stage hasn't shifted enough since 2023 to support a new era. And the one metric that looked like it proved a cash crisis story broke the moment it hit a public company, because it was never designed to measure one.
The AI narrative is cheaper to write than "we raised too much and burned through it too fast." The data agrees with the second one.
*Full analysis and code: GitHub Link
Top comments (0)