AI Costs More Than the People It Replaced

AI Costs More Than The People It Replaced

ByJemma Green,

Contributor.

Forbes contributors publish independent expert analyses and insights.

Dr Jemma Green is cofounder and chairman of www.powerledger.io

Jul 02, 2026, 12:21pm EDT

--:-- / --:--

This voice experience is generated by AI. Learn more.

This voice experience is generated by AI. Learn more.

Summary

The tech industry faces a paradoxical crisis: companies are shedding human jobs to invest in AI tools that are currently more costly than the workers they replace. Major players like Uber and Microsoft report exorbitant AI spending, with budgets exhausted rapidly and little correlation to tangible value. This "tokenmaxxing" culture, where AI usage is incentivized over actual productivity, fuels massive waste. Despite widespread layoffs justified by AI reallocation, studies indicate AI is economically viable in only a fraction of roles. The unsustainable model of subsidized AI pricing is unwinding, forcing a market correction. The industry must now shift from indiscriminate spending to architecting efficient, AI-native solutions that prove their worth, or risk a significant bubble burst.

Giant robot throwing man in a trash can. Artifical intelligence replacing jobs concept. Vector illustration.

getty

Something odd is happening in the tech world right now: the technology that was supposed to make human labour obsolete is, at this moment, more expensive than the humans it was meant to replace. Companies are laying off workers to fund the very AI tools that cost more than the workers they just let go. The circular logic of it would be darkly comic if tens of thousands of livelihoods weren't caught in the middle.

Uber’s CTO, recently disclosed that the company burned through its entire 2026 AI coding budget in four months. By March, 84 percent of Uber's engineers had adopted Claude Code, and roughly 70 percent of committed code now originates with AI. The usage was enormous. The corresponding value was murkier. Uber's COO and President, Andrew Macdonald, conceded publicly that token usage didn't seem to correlate directly with useful features shipped to users.

Uber is not an outlier. Microsoft, which has invested approximately $13 billion in OpenAI and writes up to 30 percent of its own code with generative AI, instructed engineers in a major division to stop using an AI coding assistant because the bills became untenable. One unnamed company, per Axios, ran up a $500 million Claude bill in a single month after management forgot to set a usage cap. These are structural miscalculations about what intelligence costs when you purchase it by the syllable.

Bryan Catanzaro, Nvidia's vice president of applied deep learning, put it bluntly: the cost of compute for his team now far exceeds what the company spends on the employees using it. The company that manufactures the hardware powering the AI revolution acknowledges that the technology is more expensive than the people it was supposed to augment.

And yet Catanzaro's boss, Jensen Huang tells the industry that a $500,000 engineer should be consuming at least $250,000 worth of AI tokens annually, and that Nvidia is working toward a $2 billion annual token budget for its engineering force. Tokens, he has suggested, should be a recruiting perk. The message from the top of the supply chain is unmistakable: spend more, faster.

MORE FOR YOU

Companies have obliged. Big Tech has announced $740 billion in capital expenditure this year, a 69 percent increase from 2025. Gartner projects AI agent software spending alone will reach $207 billion in 2026, up 139 percent from the prior year.

Here is where the arithmetic turns perverse. Alongside this spending, more than 115,000 tech workers have been laid off in 2026 across more than 150 companies. Meta eliminated 8,000 positions. SentinelOne cut 8 percent of its workforce to redirect resources toward AI. Wix cut a fifth of its people. Block halved its headcount. Atlassian shed 1,600 jobs.

Automation worker concept with 3d rendering robot working in office

getty

The stated rationale is consistent: operational efficiency, reallocation toward AI. But the MIT study found that AI automation is economically viable in only about 23 percent of roles. For the remaining 77 percent, humans remain cheaper. Goldman Sachs' chief economist has stated plainly that he does not view AI investment as strongly growth-positive. Sequoia Capital partner David Cahn has put a number on the resulting gap: AI companies need roughly $600 billion in annual revenue to justify current infrastructure spending. The gap, as of mid-2026, is widening, not closing.

So the present moment looks like this: companies cutting human labour to fund artificial intelligence that currently costs more than the labour it replaces, in pursuit of productivity gains that most studies cannot yet verify, at a pace that is exhausting annual budgets in weeks.

The cultural dimension may be the most telling part. Amazon built an internal leaderboard called KiroRank to track AI usage among engineering teams. It was quietly taken down after employees began gaming it — burning tokens on meaningless dumb tasks solely to climb the rankings. Meta built a similar tracker called Claudeonomics. Amazon encouraged staff to "tokenmaxx," treating consumption itself as a performance indicator. When you reward people for how much they spend rather than what they produce, spending becomes the output.

Boards demanded their CEOs adopt AI. Then came indiscriminate deployment — what the industry calls tokenmaxxing. In the third phase, leadership teams are waking up to the bill and asking an overdue question: does every task actually require the most expensive model available? Roughly 95 percent of enterprise AI usage still runs on the costliest frontier models, even for work that doesn’t demand that sophistication.

When a resource becomes cheap enough to waste, people waste it without a second thought. When it becomes expensive enough to matter, they develop a sudden, fervent interest in efficiency. Artificial intelligence appears to be hurtling toward that same reckoning — except the waste is measured in billions, and it arrives on a heavy monthly invoice.

The private equity side of the market is arriving at the same conclusion from the opposite direction. At recent Crypto Valley Conference private equity panel, Giuseppe De Filippo (Head of Private Capital Markets at Julius Baer) stated that SaaS transactions are stalling because horizontal pricing no longer works and valuations have not caught up to that reality.

AI can now generate a usable interface in hours, which means the design layer companies spent years polishing is worth less than it was a year ago. What AI cannot generate is twenty years of domain logic baked into a niche ERP system for a mining operation or a water utility. The moat has moved from how software looks to what it knows.

For a long time, the working assumption was that costs were falling. Per-token prices have indeed dropped, and Gartner forecasts that running the largest models could be nearly ninety per cent cheaper by 2030. The catch is that consumption has scaled faster than prices have fallen. A study by Faros AI found that "code churn," lines of code deleted versus lines added, increased by more than 800% under high AI adoption. More tokens in, more work thrown away.

The prices companies are paying for AI usage now are not real prices. OpenAI, Anthropic, Google and Meta are all pricing inference below the cost of serving it, burning venture capital to buy market share. OpenAI spends nearly two dollars for every dollar it earns on inference. Sam Altman admitted publicly that the company loses money on its $200 per month subscriptions. The subsidy model started unwinding this year.

The spending story and the returns story have been running on separate tracks. For years, subsidised inference prices, venture-backed losses and the promise of eventual productivity kept both tracks moving in the same direction. In June 2026, the market noticed they were diverging. Chipmakers lost roughly $1.3 trillion in market value in a single session, the steepest one-day drop for the PHLX semiconductor index since the pandemic crash of March 2020. Nvidia, Micron, and AMD led the losses. South Korea's benchmark index fell 10% in a day and briefly halted trading. SpaceX slid below its IPO price within days of listing. Accenture is down 52% in six months. The selloff was not a verdict on the technology. It was a verdict on the timeline.

TOPSHOT - A currency dealer reacts as she monitors exchange rates in a foreign exchange dealing room at the Hana Bank headquarters in Seoul on February 2, 2026. South Korea's benchmark index Kospi tumbled more than five percent on February 2, in line with a sell-off across Asian markets amid fresh worries about an AI-fuelled tech rally that has sparked fears of a bubble in the sector. (Photo by Jung Yeon-je / AFP via Getty Images)

AFP via Getty Images

In April 2026, Anthropic moved enterprise customers from flat rate plans to usage based billing tied to actual compute. GitHub followed weeks later with the same shift for Copilot, after years of quietly absorbing up to eight times the subscription value for heavy users. Analysts project that when pricing normalises to reflect real infrastructure costs, enterprise AI bills rise another 30 to 50 percent above current levels.

The profitability path requires either prices going up or compute and energy costs coming down faster than consumption grows. Neither is happening. OpenAI's own projections show $14 billion in losses this year, with $44 billion in cumulative losses before any profit appears in 2029. Ray Dalio has described the current moment as the early stages of a bubble. The parallel to the late 1990s is instructive. The internet was real technology. And it still produced a crash.

If the person selling the compute calls the spending "the most fair criticism right now, there is a ton of waste,” what does the person paying the bill call it? If the cost of tokens has already exceeded the cost of the employees they were meant to replace, when does the comparison start running in the other direction? And if the answer is that costs will eventually fall far enough to close the gap, the follow-up is: who absorbs the losses in the years between now and eventually?

History already sketched the answer: the internet was real, it still crashed, and what followed wasn’t less internet – it was internet that finally paid for itself. AI is heading for the same sorting, and the divide is already visible. Lisa Emme, co-founder of Inversion AI, says, "The mistake is treating AI as a feature you add. AI-native companies rebuild around the model – and once you do, you stop paying frontier prices for work a specialised model does better and cheaper. That’s not cost-cutting; it's architecture. The winners this decade won't run the biggest model. They'll have built systems where the right model runs the right task, moving toward workflows that need a human in the loop less and less."

That is the duller future the boom hasn’t priced in – not how much intelligence you can buy but how much you can put to work. The industry has answered every question about what AI can do. It has not answered the only question that now matters: whether it pays for itself before the money runs out.

Editorial StandardsReprints & Permissions

LOADING VIDEO PLAYER...

FORBES’ FEATURED Video