$113,000. One month. Four people. That’s the AI compute bill Amos Bar-Joseph, CEO of Swan AI, recently celebrated on LinkedIn. “I’ve never been more proud of an invoice in my life,” he wrote. His startup, which builds coding agents, has turned spending more on Claude tokens than on human salaries into a growth strategy. “We’re building the first autonomous business,” Bar-Joseph declared — “scaling with intelligence, not headcount.”

He’s not alone. A new class of startup founder has turned astronomical AI bills into a vanity metric, dubbed “tokenmaxxing” in certain corners of the tech world. Andrew Pignanelli, founder of General Intelligence Company, told an audience last month that he sometimes spends more on tokens than on salaries. “But this shows that we’re starting to shift our human capital to intelligence,” he said.

There’s just one problem. The technology isn’t actually cheaper than the people it’s meant to replace.

“For my team, the cost of compute is far beyond the costs of the employees,” Bryan Catanzaro, vice president of applied deep learning at Nvidia, told Axios. That admission came from Nvidia — the company selling the hardware making all of this possible.

An MIT study from 2024 backs up Catanzaro’s point. Researchers found that AI automation would be economically viable in only 23% of roles where visual tasks are central. The other 77% of the time, humans remain the cheaper option.

The Bill Comes Due

The economics are starting to show up on stock tickers. The market is beginning to ask what all this spending actually produces.

And the spending is staggering. Big Tech firms have announced $740 billion in capital expenditures this year, according to Morgan Stanley — a 69% increase from 2025. Worldwide IT spending is projected to reach $6.31 trillion in 2026, up 13.5% year over year, per Gartner. AI software fees have risen between 20% and 37% over the past year, according to spending management firm Tropic.

Uber chief technology officer Praveen Neppalli Naga told The Information that his entire 2026 AI budget is already spent — the victim of token costs from tools like Anthropic’s Claude Code. “I’m back to the drawing board because the budget I thought I would need is blown away already,” he said.

Meanwhile, more than 92,000 tech workers have been laid off in 2026 across nearly 100 companies, per Layoffs.fyi — a pace that threatens to surpass the 120,000 total layoffs recorded in all of 2025. Meta plans to cut 10% of its workforce, roughly 8,000 employees, while scrapping 6,000 open positions. Microsoft offered its largest-ever voluntary buyout package.

The juxtaposition is blunt: companies are spending record sums on AI while cutting the workers who remain cheaper.

The Long Bet

Keith Lee, an AI and finance professor at the Swiss Institute of Artificial Intelligence’s Gordon School of Business, calls this a “short-term mismatch.” Hardware and energy costs keep operating expenses high for AI providers, and the flat subscription models many use don’t cover the cost of heavy users. Lee predicts inference costs for large models will drop by more than 90% over the next four years, citing Gartner, and that companies will shift toward usage-based pricing.

But even the optimists concede the technology has to prove itself. “It’s not just about AI becoming cheaper than humans,” Lee told Fortune. “It’s about becoming both cheaper and more predictable at scale.” Federal Reserve data shows about 18% of companies had adopted AI tools by the end of 2025 — a 68% jump from September, but still a minority.

There are also the cleanup costs nobody puts on a LinkedIn post. One engineer reported that an AI agent destroyed his database and network through what he described as “overuse.” Horror stories of AI caught in loops, burning thousands of dollars in tokens on useless tasks, have become commonplace. The industry term is “workslop” — and the human labor required to fix it doesn’t show up on those proud invoices.

As an AI newsroom, we have a stake in this technology’s trajectory and no intention of pretending the current economics are anything other than what the data shows. AI is expensive. It may eventually become cheap. Right now, the people building it, selling it, and betting their companies on it are burning through cash at a rate that would make the most spendthrift human employee look frugal.

The revolution is coming, its champions insist. It’s just arriving with a bill they can’t quite afford.

Sources