SPXNDXDJIBTCETHOILGLD10YGOOGAAPLNVDATSLAMSFTMETASOLXRPLINKLTCDOTBNBSPXNDXDJIBTCETHOILGLD10YGOOGAAPLNVDATSLAMSFTMETASOLXRPLINKLTCDOTBNB
Home AI

Why Tech’s 127,180 Job Cuts in 2026 Are Funding a $690 Billion Compute Bill

With four months still to run, 2026 has already been a worse year for technology employment than 2025 was. Layoffs.fyi counts 127,180 job losses across 281…

Overhead view of a wooden desk with printed quarterly report pages, coffee, glasses, an empty blank office badge on a lanyard, and a tablet showing the Oracle, Apple, Meta, Microsoft and Amazon logos

With four months still to run, 2026 has already been a worse year for technology employment than 2025 was. Layoffs.fyi counts 127,180 job losses across 281 companies so far this year, against 122,606 across 278 companies for the whole of last year. The industry passed the annual mark in the third week of August, which is the sort of milestone nobody schedules a press release around.

The reflex reading is that tech is in trouble. It is not. Revenue is growing, the hyperscalers are printing record quarters, and the same companies doing the cutting are simultaneously committing to the largest capital-spending program in the sector’s history. Both things are true at once, and the relationship between them is the story.

The Money Did Not Vanish, It Changed Line Items

Microsoft, Alphabet, Amazon, Meta and Oracle have collectively committed to somewhere between $660 billion and $690 billion of capital expenditure in 2026, close to double the 2025 figure. Amazon alone is tracking toward roughly $200 billion. Alphabet sits in the $175 billion to $185 billion range, Meta between $115 billion and $135 billion, Microsoft above $120 billion.

Payroll is an operating expense. Data centers are a capital expense. Moving a dollar from the first bucket to the second does not make the dollar disappear, it changes how the market values it and how the accountants treat it. Operating expense hits earnings the quarter it is spent. Capital expense is depreciated across years, and in the meantime it shows up as an asset and as evidence that a company is winning the most consequential infrastructure race since broadband.

Investors have been rewarding the second category and punishing the first. That is not a neutral incentive structure, and executives respond to incentive structures.

Oracle Is the Cleanest Version of the Trade

Oracle makes the arithmetic legible. Its fiscal 2026 capital expenditure came in near $55.7 billion, up from $21.2 billion the previous year, a more-than-doubling driven entirely by the buildout behind its cloud ambitions. Its annual report confirmed 21,000 layoffs, around 13% of a 162,000-person workforce, and further cuts followed in August.

One line went up by roughly $34 billion. Another shed 13% of the people. Those are not unrelated decisions taken by unrelated committees.

A company that lays off 13% of its staff while tripling its capital budget has not lost confidence in its future. It has decided which version of the future it wants to own, and people are not in it.

We saw the same pattern in Microsoft and Meta’s capex guidance, where the spending commitments landed in the same quarter as headcount discipline, and again in the hyperscaler capex numbers running through Nvidia’s order book. The demand Nvidia books is, at one remove, the payroll these companies decided not to carry.

The Depreciation Wave Explains the Timing

Here is the part that is easy to miss. The cuts are not primarily about paying this year’s server bill. They are about the bill that arrives in 2028.

Capital spending on this scale converts into a depreciation charge that lands on the income statement for years afterward, long after the excitement of the announcement has faded. A company that spends $200 billion on compute is committing to absorb tens of billions of annual depreciation for the rest of the decade. Margins that look fine while the spending is capitalized will look considerably worse once it starts running through earnings.

Cutting payroll now creates the margin headroom to absorb that later. It is a pre-payment, executed while the labor market is soft enough to permit it and while investors are applauding anything with the word infrastructure attached. The companies moving fastest are the ones with the largest depreciation exposure coming, which is exactly what you would expect if this reading is right.

The Cuts Are Aimed at Specific Bets, Not Spread Evenly

Apple’s August reductions are instructive because they were surgical. The company cut more than 200 roles split roughly between the Vision Pro organization and the Siri and software teams, largely shutting down a Vision Pro gaming group and thinning the unit producing immersive video. The Siri cuts hit staff working on the previous generation of the assistant, because the replacement needs a different skill set entirely.

Apple framed it as realigning teams to evolve the business, and said new roles would be created alongside the losses. Both halves of that statement are probably accurate, and neither changes what happened to the people in the old Siri org. This is what a strategic pivot looks like from the inside: not a company shrinking, but a company deciding that the expertise it spent a decade accumulating is the wrong expertise for the next decade.

TikTok, LinkedIn and Netflix all announced reductions in the same month, according to tracking of the August wave. The first quarter set the tone, with 81,747 tech workers losing jobs, the heaviest quarterly figure in at least two years.

What Would Have to Change

For context, this is still not 2023, when 265,660 people lost tech jobs across 1,194 companies. The current wave is narrower and concentrated in larger firms, which is consistent with a capital-reallocation story rather than a solvency one. Coverage of the milestone has rightly framed it as a record with months to spare rather than a collapse.

The uncomfortable question is what stops it. If the AI infrastructure buildout delivers the productivity gains its sponsors promise, headcount does not come back, because the whole premise was that it would not need to. If the buildout disappoints, the depreciation arrives anyway and the pressure on operating costs gets worse, not better. When we looked at the 88,000 AI-attributed job cuts earlier this year, entry-level roles were absorbing the first hit, and nothing since has reversed that.

Watch the fourth-quarter capex guidance. If the big five hold their 2027 spending plans while the layoff tracker keeps climbing, the conversion of payroll into compute is a structural feature of how these businesses now operate, not a cycle anyone is waiting out.