Nearly 140,000 US tech jobs have been cut since the start of 2026. In a growing share of the announcements, AI appears somewhere in the explanation.
TechCrunch has been maintaining a running list of the major cuts where employers name-checked AI. Reading it in one sitting is more informative than any individual announcement, because the interesting variable is not whether AI is mentioned. It is how precisely it is mentioned — and for most of these companies, the answer is not very.
The list
| Company | Cut | Date | How AI was framed |
|---|---|---|---|
| Amazon | 16,000 | Jan 2026 | Vague — "we will need fewer people" as AI adoption rises |
| Dell | ~11,000 (10%) | Jan/Mar 2026 | Vague — savings redirected to AI-optimised servers |
| Salesforce | <1,000 | Feb 2026 | Explicit — fewer support cases; Agentforce handles the work |
| Block | ~4,000 (40%) | Feb 2026 | Vague — "intelligence tools" enabling a new way of working |
| Atlassian | 1,600 (10%) | Mar 2026 | Vague — and says AI "doesn't change... the number of roles required" |
| Snap | ~1,000 (16%) | Apr 2026 | Vague — reduce repetitive work, increase velocity |
| Coinbase | ~700 (14%) | May 2026 | Vague — engineers ship in days what took weeks |
| PayPal | 4,500+ (20%) | May 2026 | Vague — "aggressively adopt AI" |
| Cloudflare | 1,100 (20%) | May 2026 | Explicit — the majority were "measurers"; middle management made obsolete |
| GM | 500–600 | May 2026 | Vague — AI played "a role", not the sole cause |
| Meta | 8,000 (10%) | May 2026 | Mixed — moved 7,000 into AI roles simultaneously |
| Intuit | ~3,000 (17%) | May 2026 | Vague — "reallocating resources toward AI" |
| Cisco | ~4,000 (5%) | May 2026 | Vague — realigning around silicon, optics, security, AI |
| 1,500–3,000+ | through May 2026 | Vague — rolling performance reviews, no official figures | |
| GitLab | ~350 (14%) | Jun 2026 | Vague — restructuring for "100x growth requirements" |
| Oracle | 21,000 (13%) | Jun 2026 | Explicit — AI adoption "resulted in reductions to our workforce" |
| Microsoft | 4,800 (2.1%) | Jul 2026 | Contradictory — roles "not being replaced by AI", but AI "changing how work gets done" |
| Monday.com | ~600 (20%) | Jul 2026 | Vague — co-founder denied it was to replace people with AI |
| IBM | 3,000–9,000 | rolling | Mixed — ~200 HR roles to AI agents, while tripling entry-level AI hiring |
Four companies actually said something
Strip out the language that could describe any restructuring in any year and very little remains.
Oracle is the clearest: AI adoption and deployment "have resulted in reductions to our workforce." That is a causal claim, in a filing-adjacent register, about 21,000 people.
Cloudflare is the most specific about who. The majority of the 1,100 were "measurers" — middle management whose function was tracking and reporting on work — described as made obsolete. That is a claim about a category of job rather than a headcount target, and it is falsifiable.
Salesforce named the product. Fewer support cases require humans because Agentforce handles them. Under 1,000 roles, and a mechanism you could audit.
IBM is precise in a smaller way: roughly 200 HR roles replaced by AI agents — while simultaneously tripling entry-level AI hiring.
Everything else is compatible with a company that would have cut costs regardless and reached for the available vocabulary.
The tell is in the contradictions
Two entries undermine the framing from inside.
Atlassian cut 1,600 people while stating that AI "doesn't change the mix of skills we need or number of roles required." Both things were said. Only one of them explains a layoff.
Microsoft managed both positions in one announcement: the roles were "not being replaced by AI", but "AI is changing how work gets done." That is a sentence engineered to be quoted either way.
Meta cut 8,000 and moved 7,000 people into AI roles at the same time. That is not AI eliminating work. It is a company reallocating its workforce toward a bet — and saying out loud that "success isn't a given in AI".
The market read it too
This is the most useful data point in the whole picture, and it is not a matter of opinion.
Financial Times analysis found that companies citing AI as layoff justification underperformed the Nasdaq by nearly 10% over the 30 trading days following the announcement.
That is the opposite of what the framing is meant to achieve. "We are cutting costs because our AI works" is supposed to read as a company getting more efficient. Investors priced it as something else — either as an admission that demand is soft and AI is the cover story, or as scepticism that the claimed productivity exists.
The market, in aggregate, does not appear to believe these companies.
Why founders should care about the vocabulary
If you are building something, three things follow.
The talent market is not what the headlines imply. 140,000 cuts alongside aggressive hiring at Anthropic and OpenAI, and IBM tripling entry-level AI hiring, is not a labour surplus. It is a violent reallocation. The people you can now hire are not necessarily the people the AI-labs are competing for, and that cuts both ways.
"AI made us efficient" is a claim you will be asked to substantiate. The FT data suggests sophisticated audiences already discount it. If you use that line in a fundraise, expect to be asked which function, by how much, and measured how — which is exactly the question Oracle, Cloudflare, Salesforce and IBM can answer and the other fifteen cannot.
Middle layers are where the pressure lands first. Cloudflare's "measurers" formulation is worth taking seriously precisely because it is specific. The roles most exposed are the ones whose output is coordination and reporting rather than the work itself. That is a structural observation about org design, and it applies to a 40-person company as much as an 11,000-person one.
What this list does not tell you
It is worth stating the limits.
Being on this list means the employer mentioned AI, not that AI caused anything. Attributing a layoff to AI is a communications decision, made by people who know it sounds better than "we over-hired" or "revenue missed."
Companies with genuine AI-driven efficiency gains who chose not to say so are absent entirely. So the list measures the willingness to attribute, and the FT's market data measures whether that attribution was believed.
On current evidence, the attribution is common and the belief is not.
