AI Layoffs: Cloudflare Tumbles 24%

Cloudflare lost a quarter of its market cap in 24 hours. The official reason fits in a single line: 1,100 layoffs framed as AI productivity gains. Three weeks earlier, Snap announced a similar cut and its stock climbed 7% in pre-market. Between the two, the narrative flipped.

The same weekend, sociologist Dominique Méda published a Le Monde op-ed with an unambiguous title: « Inside companies, give workers a power equal to those who bring capital ». The critical sociologist frames the problem of who decides to introduce AI in companies. Wall Street just framed the same question, in stock-price form.

When financial markets and critical sociology say the same thing the same week, it is worth pausing.

The market no longer rewards the "replace with AI" pitch

Look at the numbers. Cloudflare reports Q1 2026 on the evening of May 7. Revenue of $639.8 million, up 34% year-over-year, beating analyst expectations. In the same breath, CEO Matthew Prince announces 1,100 job cuts, roughly 20% of staff. The justification is explicit: pivot to an "agentic AI-first operating model", with internal AI usage up 600% in three months.

The next day, the stock plunges 24%. For TechCrunch, it is "the most explicit case of a company directly attributing layoffs to AI replacing human roles". The market ruled that solid quarterly fundamentals are not enough to offset the signal.

Three weeks earlier, on April 15, Snap announced exactly the same thing: 1,000 layoffs, 16% of staff, "65% of new code generated by AI" per Evan Spiegel. The stock rose 7% in pre-market. Three weeks, two structurally identical announcements, two opposite verdicts. The market's reading frame flipped in between. Investors started seeing what these layoffs hide: a loss of service capacity that AI agents do not compensate, and technical debt that will land later.

Méda: who decides on AI inside companies

While Wall Street punishes the "all-in AI" pitch, Dominique Méda publishes a Le Monde op-ed whose title sets the thesis. She directly attacks the promise pushed by Elon Musk and Sam Altman of a world freed from work, where employment would, in her own words, suffer "a massive bloodletting in the years ahead". Méda refuses to stage that bloodletting as liberation. Her answer, in the headline: « Inside companies, give workers a power equal to those who bring capital. »

This thesis lines up with what the market is saying. The sociologist is not arguing about the productivity of any specific AI agent. She raises an upstream question. Who decides to introduce AI inside a company, and against which counter-powers?

The reference frame is well known but never installed in France: German codetermination, formalized in 1976, which grants worker representatives seats on the supervisory boards of large companies. France kept a consultative setup via the CSE works councils. On AI decisions, their voice stays informational, never decisive.

What the Cloudflare sell-off shows is that the market is starting to price that absence of counter-power as a risk. A company laying off 20% of its operators on the promise of an AI agent, without anyone inside able to say "wait, let us look at the actual data", sends a signal of fragile governance. The stock price translates that into a number.

Klarna, IBM, Salesforce: the walk-back has become the rule

To gauge how solid the diagnosis is, look at the companies that tried first. Klarna announced in February 2024 that its AI could replace 700 customer service agents. Sebastian Siemiatkowski made it the marketing argument of the moment. A year later, the same CEO publicly admits: "We went too far". Klarna is rehiring in hybrid mode, because customers walked away and satisfaction collapsed.

IBM did the same on HR. Arvind Krishna confirmed that AskHR and the watsonx suite replaced "a few hundred" HR roles. But on the bottom line, he had to acknowledge to the Wall Street Journal that overall IBM headcount had grown, with HR savings redeployed to roles that AI does not handle. The "we are replacing" storyline holds for the press. The accounting line says "we are reallocating".

Salesforce, same story. Marc Benioff announced 4,000 customer service cuts in September 2025, plain phrasing: "I need less heads". Seven months later, he hires 1,000 new grads and reframes the strategy: "AI won't kill entry-level jobs". The PR pivot is so sharp that the implicit admission shows through: the cut cost more in service quality than it returned in margin.

Three cases, one pattern. The "replace with AI" promise announced in fanfare by the CEO becomes, twelve months later, a quietly managed retraction. What Wall Street finally saw with Cloudflare is that the pattern is now legible before the walk-back. Investors sanction at the announcement.

95% of enterprise AI projects show no ROI

If markets and sociology converge, it is also because the empirical data is in. In August 2025, MIT's NANDA initiative published a study on 300 enterprise AI deployments, with 150 interviews and 350 employee surveys. The verdict: 95% of corporate GenAI projects produced no measurable return. Of the $30-40 billion invested, only 5% of pilots reach revenue acceleration.

The detail that cuts deep: vendor-bought deployments succeed twice as often as in-house builds. The most common decision, "we build our own AI stack internally, under CEO direction", is statistically the worst option. Researchers identify a "learning gap" as the structural cause: systems do not retain field feedback, do not adapt to context. A feedback loop that is missing because the field has no voice in the room.

That is exactly the Méda diagnosis, in data scientist language.

For a manager, the practical question has changed

You are weighing whether to push an AI project inside your team. The default reading frame, the one in McKinsey decks, lists how many roles you replace, how much you save, what ROI you project. It no longer works, neither on the numbers nor on market perception.

The frame that does work is one of decision rights. Who proposes the project. Who signs off. Which operators were consulted on real data and real workflows. What exit clause exists if the agent does not deliver after six months. If the answer to all those questions is "the CEO and the board", the project carries the same risk profile as Cloudflare before May 8.

Cloudflare is a durable precedent, not a disposable news beat. The next company announcing a layoff in the name of AI will be priced on that same grid, unless it can show it set up something other than a unilateral decision. Méda has been theorizing that "something other" for twenty years. Wall Street just joined her through another door.