It is the same scene on repeat in 2026: a big tech company posts record cloud numbers, announces another data center deal worth tens of billions, and then quietly trims thousands of people from the payroll. This week it was Uber confirming around 3,300 job cuts, PayPal trimming hundreds in India, and Oracle sitting on reports of another wave that could run to 10,000 roles globally. Layoffs.fyi puts total tech losses this year near 130,000, with more than 6,300 roles gone in the last ten days alone.

The easy headline is that AI ate the jobs. The real story is more uncomfortable for anyone running a business: this is what happens when a company decides to spend everything on compute and then goes looking for the money to pay for it. Oracle is the cleanest example in the industry right now.

Uber's Math: 3,300 Roles, 10 Percent, Fewer Managers

Uber went first this month. CEO Dara Khosrowshahi told staff in an internal memo that the company is cutting roughly 3,300 jobs globally, about 10 percent of the workforce, in its largest reduction since the Covid-19 pandemic. The goal is to remove management layers, simplify teams, and rethink where the company operates. Reports from India suggest 200 to 300 roles there are caught in the sweep.

Notice what the memo did not say. Bloomberg reported earlier that Uber actually blew through its annual AI budget in about four months and had to cap internal usage. So the company is simultaneously spending more on AI, running out of budget for it, and cutting people to reinvest. That is not a struggling business. That is a business reallocating.

Oracle Is the Story Nobody Is Reading Carefully Enough

Oracle is the one to watch, because the numbers there are genuinely wild. Headcount fell by about 21,000, or 13 percent, in the fiscal year that ended in May, dropping from roughly 162,000 to 141,000. Business Insider reported managers are drawing up lists for another round, with internal estimates of 7,000 to 10,000 roles globally and 2,000 to 3,000 in India, where Oracle employs around 30,000 people. Employees have been watching Slack counts and waiting for a 6 am email since September 1 came and went without one.

Now the part that should make every operator pause. Capital expenditure hit 55.7 billion in fiscal 2026, up from 21.2 billion the year before, as Oracle builds data centers for customers including OpenAI. Free cash flow went negative to the tune of 23.7 billion. The company raised 43 billion in debt and 5 billion in equity, expects roughly 40 billion more this year, and interest expense alone climbed to 4.6 billion. Restructuring charges hit 1.8 billion, a 391 percent jump, against a 2.1 billion cap in its own 10-K, leaving about 300 million of headroom.

And here is the kicker: revenue grew 17 percent to 67.4 billion, cloud infrastructure revenue grew 77 percent, and remaining performance obligations closed the year at 638 billion. This is a growing company cutting people because it has committed so much capital to AI infrastructure that payroll is now the line item that has to give. TD Cowen estimated in January that cutting 20,000 to 30,000 roles could free up 8 to 10 billion in incremental free cash flow. Seven months later, that reads less like a forecast and more like a plan.

PayPal, Microsoft, Apple: The Rest of the Wave

Oracle is not alone. PayPal initially faced reports of around 600 job losses in India before clarifying that about 220 employees, roughly 4 percent of its India teams, were affected as of the end of August. Microsoft has reportedly placed around 400 to 500 employees in India on performance improvement plans. Apple cut roughly 200 roles tied to Siri, Vision Pro, and related software teams last month. Block's Jack Dorsey set the tone earlier in the year by cutting the company nearly in half, from over 10,000 people to just under 6,000.

The pattern is consistent: the cuts land hardest in the middle. Management layers, legacy sales, support, and the work between a product decision and a shipped feature. Gartner's Helen Poitevin warned that companies cutting headcount to demonstrate quick AI returns are making a mistake, because workforce reductions create budget room, not return. The organizations that win, she argued, invest in the skills and operating models that let humans guide and scale autonomous systems.

Why This Is an Allocation Story, Not an AI Story

This is the framing that matters. The lazy version of this news cycle says AI is replacing workers, full stop. The more accurate version is that AI is forcing companies to make brutal capital allocation choices, and labor is the flexible line item. When Oracle commits hundreds of billions to data centers, borrows at scale, and watches interest expense climb by a billion dollars in a single year, something has to absorb that cost. Severance is cheaper than a GPU cluster. That is not a technology verdict. That is an accounting one.

It is also why the India numbers sting so specifically. India's five largest IT companies reduced their combined workforce by a net 6,981 employees in fiscal 2026 after adding 12,718 the year before, and the campus-hiring model of building bench strength is largely gone. Hiring is now tied to specific projects and actual demand. The first quarter of fiscal 2027 showed a small reversal, led by TCS adding 9,000 net roles, but the structural shift is not reversing. Global capability centers are hiring for AI, data, cybersecurity, and digital engineering, not for generalist fresher pipelines.

What This Means for Anyone Running a Stack

If you run a small shop, do not read this as a warning to avoid AI. Read it as a warning about how you fund it. The companies getting into trouble committed to massive fixed infrastructure costs before the revenue was locked in, then discovered the only lever left was people. The ones that come through this cycle cleanly treat AI as a variable cost tied to actual output, the way a sensible operator treats any other tool.

The other lesson is quieter. When a company runs a 6 am email layoff with no HR call and DocuSign paperwork the same morning, it is spending trust it cannot buy back. The engineers watching Slack counts drift downward today will remember this when the next recruiter calls.

What Comes Next

Oracle's fiscal calendar made September 1 the date everyone watched, and when the email did not arrive, the dread moved to September 15 and then to late September. The company has not confirmed anything publicly and may be trying to fit the cuts inside its existing restructuring cap. Uber's cuts will roll out through the rest of the year. Every earnings call for the next two quarters will feature the same choreography: AI investment up, headcount down, efficiency praised, and nobody saying the quiet part about payroll funding the data centers.

The takeaway for the rest of us is simple. Watch the capital allocation, not the press releases. When a company's capex outruns its free cash flow by 20-plus billion dollars, the headcount question is not if, it is when. And the people who survive this cycle are not the ones who can write code fastest. They are the ones who understand where the money actually goes, because that is the skill every one of these restructurings is quietly selecting for.

— Allan Ali, Sylt.ing