The four largest US technology companies will collectively spend up to $725 billion on AI infrastructure in 2026, a 77% jump from last year, while over 92,000 tech workers have already lost their jobs in the first four months alone, making this the most explicit trade-off between compute and headcount in the industry's history.
Mark Zuckerberg said it plainly in an internal memo. Meta has two major cost centres: compute infrastructure and people. The more you spend on one, the less you have for the other. It is unusual for a CEO to articulate a strategic reallocation with that much transparency. Normally the message gets softened into language about organisational efficiency, strategic focus, and right-sizing. Zuckerberg skipped all of that. He committed Meta to between $115 billion and $135 billion in AI infrastructure this year, nearly double the prior year's capital expenditure, and announced 8,000 job cuts alongside 6,000 unfilled roles that will stay vacant. The jobs are funding the data centres. He said it explicitly. The workers heard it.
The aggregate numbers are harder to contextualise against anything in tech history because nothing quite like this has happened before. Google, Amazon, Microsoft, and Meta reported combined capital expenditure of more than $130 billion in Q1 2026 alone, according to their earnings disclosures. Microsoft has set its full-year 2026 capex at $190 billion, significantly above analyst consensus, with CFO Amy Hood noting that $25 billion of that reflects rising memory chip and component costs. Amazon has committed $200 billion in AI investment for the year. Oracle, carrying the weight of its own data centre expansion programme, has cut up to 30,000 workers in the largest single workforce reduction of this cycle. Amazon cut approximately 30,000 total, including 16,000 in January. Meta's cuts total 8,000 with more signalled for the second half of the year. Microsoft offered voluntary buyouts to 8,750 US employees in the first such programme in the company's 51-year history. By April, more than 92,000 tech workers had received notices, a number that already makes 2026 the worst year for tech employment on record.
What makes this structural rather than cyclical is the Snap disclosure. The company cut 1,000 jobs, 16% of its workforce, and explained the rationale with a single data point: more than 65% of its new code is now AI-generated. Smaller, more focused teams can produce the same output. That is not a claim about future capability or projected productivity gains. It is a present-tense operational statement about what is happening inside a production engineering organisation right now. Autodesk cut 1,000 jobs. Epic Games cut 1,000. These are not consumer-facing AI companies that have been publicly building toward this transition for years. They are software businesses that develop tools and games, and they are reducing headcount because AI generation has materially compressed the labour requirement for their core functions.
The financial logic underlying all of this is straightforward to describe and genuinely uncertain in its outcomes. Google's Q1 2026 cloud revenue grew 63% year over year to $20 billion, driven substantially by AI workloads. That number is the strongest evidence that the infrastructure investment is generating returns. Jefferies analyst Brent Thill looked at that figure and told the Financial Times the bear thesis on AI was garbage. The counterargument, visible in Meta's stock dropping 7% in extended trading despite strong headline earnings, is that investors are watching the magnitude of forward capital commitments and asking when the revenue growth catches up. Microsoft, Amazon, and Alphabet are also raising debt to fund infrastructure that, in significant portions, is not yet generating corresponding revenue. McKinsey projects that AI infrastructure will require $6.7 trillion globally by 2030. That is a demand forecast, not a guarantee. The companies spending the most are making the largest bets that the forecast proves directionally correct.
Mustafa Suleyman, Microsoft's AI division head, said earlier this year that AI could replace most white-collar jobs within 12 to 18 months. His company is simultaneously offering buyouts to 8,750 employees in the first such programme in its history, targeting workers over 50 with long tenure, specifically because their per-head cost is higher and their roles are among those most likely to be compressed by the AI tools their employer is building. There is a particular quality to that combination: the people with the most institutional knowledge and the highest accumulated salaries are the first to be replaced by the systems their company is investing most heavily to build. The voluntary framing softens the optics without changing what it is.
Zuckerberg's acknowledgment that he does not have a crystal ball plan for the next three years is the most honest statement about the current moment that any of these CEOs has made publicly. The companies spending the most on AI infrastructure are operating on conviction rather than certainty. They believe the productivity gains will compound, the revenue will follow, and the leaner headcount structure will make them more profitable organisations than the ones they were in 2022. Some of them will be right. The workers who have already lost jobs are not waiting for the outcome of that bet. They are updating their CVs in a market where the companies most likely to hire them are deploying the same AI tools that replaced their previous roles. That is the condition the industry has arrived at in 2026: record infrastructure investment, record workforce reductions, and a level of executive honesty about the connection between the two that was not present in any prior cycle.
Also read: ElevenLabs is moving from voice generation into voice labor and the call center industry has not fully priced that in • AI dictation apps are proliferating fast but the startups building them are running out of time before the platforms absorb the category • Agentic search changes what a benchmark score actually means and founders are not reading the fine print
