Can the AI-led job losses really persist?

We’ve all been cut off mid-call when the prepaid balance dies. Now imagine your whole business runs that way — and the bill is a mystery until it lands. That, increasingly, is life with AI: a worker that is brilliant and tireless, but charges by the word and falls silent the moment the money runs out. In 2026, the companies that rushed to hire this worker are discovering both sides of that bargain the hard way.

And they had every reason to rush. As per FinOps Foundation( Linux Foundation industry body), resolving one routine customer query costs a company around 6-8 dollars if a human does it. An AI agent does the same job for somewhere between 1 and 3 dollars! Multiply that across millions of queries, and the appeal is obvious.

But here is the twist now catching companies unawares: the price of AI has been tumbling but the bills have been climbing anyway.

The price keeps dropping. So why is the bill bigger?

The raw cost of AI has collapsed, down roughly 98% in two years (Stanford AI Index Report 2025). AI is billed in “tokens”; think of them as tiny fragments of words. If nothing else had changed, every AI bill on earth would be shrinking.

The catch is in how AI itself has changed. A few years ago, asking it a question was like stopping a stranger for directions — you got a quick, one-line answer.

Today’s AI behaves more like an over-eager new recruit who insists on showing all their work skills: it plans, looks things up, checks itself, second-guesses, and tries again — chewing through thousands of words for a task that once took a handful.

Consulting firm EY put a figure on it: a customer-service task that cost about $0.04 to run in 2023 costs around $1.20 today — roughly 30x more. Each word got far cheaper, but the machine simply uses a great many more of them.

The wider numbers tell the same story. Even as the price per unit fell drastically, enterprise AI bills have risen by an estimated 320%, according to multiple industry analyses.

In a 2026 survey by the FinOps Foundation, a corporate-finance body, 73% of organisations said their AI costs came in higher than planned. And there is little relief in sight: research firm Gartner reckons that by 2030, prices could fall another 90% and bills would still go up — simply because everyone is using so much more.

The bit nobody saw coming: It stops when the balance hits zero

A salaried worker turns up whether they handle two queries or two hundred; their pay does not change. AI is the opposite — you pay for every single use, and the meter never sleeps. That brings two huge shocks.

The first is bill shock. It was reported that ride-hailing giant Uber burned through its entire 2026 AI coding budget by April, and that even Microsoft pulled its own engineers’ access to an AI coding tool once the charges piled up. When companies pay per use, a busy week can cost a small fortune — and they often don’t notice until the invoice lands.

The second is harsher: when the budget runs dry, the work simply stops. Hit your limit, or suffer an outage, and every process wired to that AI freezes — like a factory line when the power trips.

This is not hypothetical. A run of 2026 breakdowns — including Anthropic’s Claude in March and Microsoft’s Copilot in May and June — left companies watching their chatbots go mute and their automated tasks stall mid-step.

The analytics firm Ookla logged 3.72 million AI-outage reports in the US between January 2025 and April 2026, and found the newer, cleverer “agentic” systems broke down 210% more often than simple ones. What’s more striking is that 68% of the businesses it surveyed now run at least one process that cannot function without AI.

When a company swaps its support desk for a bot, the people who used to be the backup are gone — so when the AI falls over, there is often no Plan B at all. It is partly why some firms, the fintech Klarna among them, have quietly begun hiring humans back to work alongside the machines.

What it means for you?

None of this means AI doesn’t work, or isn’t worth it. Plenty of companies are saving real money. The point is simpler: the math is still being worked out, and the argument that “AI is cheap” hides a moving target.

So if you are sizing up a company that has gone all-in on AI, a few plain questions help. Is its AI spending a one-off saving, or a bill that grows every month? Are today’s low prices even built to last — or is the provider running at a loss to grab customers, the way a new delivery app pampers you with discounts that quietly vanish later?

And what happens to the business if its AI supplier raises prices, hits a cap, or goes offline for a day? Also, it’s worth remembering that the MIT research group NANDA found in 2025 that only about 5% of company AI projects delivered a clear, measurable payoff. In other words, using AI and profiting from it are not the same thing.

So the question buried under the headlines is a personal one: if AI is this expensive and this temperamental, does the threat to those jobs actually ease?

On one side, the cost reality is a real brake. A company that expected to swap a salaried team for a cheap bot, and instead ran up a seven-figure AI bill, watched the system stall when the budget ran dry, and couldn't prove the results beat the humans, has every reason to slow down. When the maths stops working, the firing tends to pause.

On the other side, a lot of the job cuts pinned on AI were never really about AI. They were about margins and over-hiring in the boom years.

Before you let news headlines drive what you do with a career or a portfolio, it's worth asking the same question these companies are now being forced to ask: not "can AI do this?", but "what does it actually cost to make AI do this — and is it worth it?"

The bottom line

AI was sold as the perfect worker: one that never sleeps, never tires and costs a fraction of a human. Much of that is true. But it pays to remember what kind of cost it is. A salary is fixed; a token is metered. A colleague can cover for you, but an AI goes quiet the moment the money runs out. The real question for 2026 isn’t “can AI do the job?” It clearly can. It is “what does it cost to keep doing the job, day after day after day?”