Tech Is Firing Everyone to Pay for AI

This is a story about a transaction. A cold, deliberate, historically unprecedented transaction between the technology industry and its workforce. In the first five months of 2026, more than 142,000 tech workers lost their jobs. That is a 33% increase over the same period last year. Workforce analytics firm TrueUp projects the full-year total could reach 370,000 — a figure that would rival the post-pandemic record. And unlike the 2023 layoff wave — which was a correction for pandemic over-hiring — this one is happening while the companies doing the cutting are posting the strongest financial results in their history.

Google, Amazon, Microsoft, and Meta have collectively committed to $700 billion in capital expenditure in 2026 — nearly double what they spent in 2025. Amazon alone has pledged $200 billion. Meta is spending over $100 billion on AI data centres while simultaneously cutting 8,000 workers. The pattern is consistent and unmistakable: human payroll is being liquidated to fund AI infrastructure. The workers are not being let go because the business is struggling. They are being let go because the business no longer needs them in the same proportions it once did — and because the compute that replaces them costs more than they do.

The Companies Cutting the Most

The layoffs are not evenly distributed. They are concentrated in roles that AI can demonstrably replace — software engineers, data analysts, content moderators, customer support, HR operations, and mid-level product managers. The cuts are heaviest at the companies with the largest AI ambitions, because those are the companies both most able to deploy AI replacements and most motivated to redirect payroll toward compute.

The cumulative toll since 2020 is staggering: more than 1.45 million tech workers have been affected by layoffs since the start of that year. The industry added these people during a period of explosive digital growth driven by pandemic behaviour shifts — then corrected the over-hire, then corrected again for macro pressures, and is now correcting a third time for AI. Each wave has been framed differently. This one is being framed as transformation. Critics are framing it as something else entirely.

"AI Psychosis" — When Executives Lose the Plot

Box founder Aaron Levie put a name to what many in the industry have been observing: "AI psychosis." His diagnosis, shared publicly this week, was pointed: the people deciding that AI can replace your job are the ones least likely to understand what your job actually involves. They see a language model that can write a coherent email and conclude that the person who was writing emails is now redundant. They see a code generation tool that produces syntactically correct functions and conclude that the engineer who was reviewing, testing, debugging, and architecting systems is now optional.

The Stanford Human-Centered AI Institute published data this week that adds empirical weight to Levie's concern. Software developer employment for workers under 26 has fallen nearly 20% since 2024. The entry-level coding roles — the ones where engineers learn the craft, build the judgment, and develop the intuition that makes a senior engineer worth ten of them — are disappearing fastest. AI is not just replacing current workers. It is compressing the training pipeline through which the next generation of engineers was supposed to emerge.

The DuckDuckGo Signal — Users Are Voting With Their Clicks

Not all of the backlash against the AI-everything pivot is coming from laid-off workers. Some of it is coming from the consumers those workers were building products for. DuckDuckGo installs are climbing — and the reason, according to user feedback and commentary, is simple: people want Google to stop forcing AI into search and just give them links.

Google's AI Overviews — the AI-generated summaries that now appear at the top of most search results — have been controversial since launch. Publishers hate them because they reduce click-through to original sources. Users who liked the traditional search experience find them cluttered and occasionally wrong. The growing DuckDuckGo install base is a measurable signal that a meaningful segment of users will switch search providers to escape the AI-first experience, even if that means sacrificing Google's depth and personalisation.

Google's partial response — adding more website links inside AI Overviews, announced this month — reflects the tension between two incompatible commercial goals: maximising the AI experience that keeps users engaged, and maintaining the referral traffic relationships with publishers that make the web's content economy function. As covered by Ars Technica, the change is being read by publishers as a partial concession rather than a genuine reversal.

Microsoft Copilot — Reverted, Rethought

In a move that TechRadar described as a sign that Microsoft is "seemingly undecided on what to do with Copilot and AI on Windows 11," the company reverted the Copilot AI assistant to its original design this week — rolling back interface changes that had been introduced over the past several months. The reversion is significant because it suggests Microsoft's internal research is showing that the AI-first interface changes are not landing with the mainstream Windows user base in the way the company had hoped.

Fortune's headline from earlier this month — "Microsoft lost its way in the AI race. Can Copilot get it back on course?" — captures the broader problem. Microsoft invested billions in OpenAI, integrated Copilot across its entire product suite, and made AI the centrepiece of its Windows 11 strategy. The Copilot reversion suggests that neither the product nor the communication strategy is working as intended. For a company that has staked its next decade on AI being as transformative as the internet was for its previous decade, that is a real concern.

The Paradox in Numbers

The defining feature of this moment is not the layoff number alone — it is the simultaneity. Record revenues. Record AI investment. Record layoffs. All happening at the same companies, in the same quarters, described in the same earnings calls where executives celebrate the business's strength and mourn the human cost in the same breath.

Glassdoor's Employee Confidence Index for the tech sector fell 6.8 percentage points year-over-year to 47.2% in March — the largest year-on-year confidence drop of any industry. Workers know what is happening. They can read the same earnings reports their bosses present to investors. The model is not hidden: their salaries are being reallocated to GPU clusters, and the GPU clusters are being used to build tools that will do what they did, faster and cheaper.

Apple Delays Smart Glasses — One Bright Spot for Caution

In a piece of news that cuts against the relentless acceleration narrative, Apple has delayed its iPhone-connected smart glasses to late 2027, according to sources cited by Techmeme today. Apple is targeting the mid-tier $200–$500 eyewear market — the same segment where Ray-Ban Meta glasses have found traction — and the company is aiming to disrupt it the way it disrupted the watch market with Apple Watch.

The delay is notable because it suggests Apple is still willing to take extra time to get a product right rather than ship to a competitive timeline. In an industry where every other company is racing to be first with AI hardware, Apple's willingness to push to late 2027 is either disciplined patience or a sign that the product is not yet ready. Given Apple Watch's trajectory — initially mocked, now the world's best-selling watch — betting against Apple's patience on new hardware categories is a historically poor wager.

What Happens Next

TrueUp's projection of 370,000 layoffs for the full year of 2026 would put this cycle second only to 2023's post-pandemic correction. But the structural difference matters enormously. The 2023 wave was a correction of excess. This wave is a structural reallocation. When corrections end, hiring resumes in the corrected categories. When structural reallocations complete, the eliminated categories do not come back. The question for 2027 and beyond is whether the AI productivity gains materialise fast enough — and broadly enough — to create new categories of work that absorb the workers being displaced from old ones.

The honest answer, as of June 1, 2026, is that nobody knows. The optimists point to the 275,000 open AI jobs, the new categories of work being created around AI development and governance, and the historical pattern of technology creating more jobs than it destroys over the long run. The pessimists point to the speed of this transition — which is compressing timelines that previous waves spread across decades into years — and to Stanford's data showing the career ladder's bottom rungs disappearing first, before young workers have had a chance to climb them.

What is certain is that the transaction is real. The technology industry is trading its human workforce for compute. The bet is that the compute will generate enough value to fund a new economy of work on the other side. That bet may be right. But the workers being cut today will not be comforted by a historical trend. They are living on the wrong side of the graph, right now, while the data centres go up.