AI’s reality check: The “never-ending utility bill” bites

Key topics:AI IPO hype and trillion-dollar valuations face rising ROI skepticismCompanies hit AI cost overruns as usage caps follow massive token billsDebate grows: productivity gains vs runaway AI costs and unclear returns.By Parmy Olson.In the coming months, some of the world’s biggest AI labs — including OpenAI and Anthropic PBC — will debut on the public markets at valuations close to $1 trillion each. Alphabet Inc. is tapping equity and debt markets to fund its AI spending, while older companies like Hewlett Packard Enterprise Co., Dell Technologies Inc. and Nokia Oyj are benefiting from the gold rush too.Everyone is piling in on the bet that artificial intelligence will make companies more profitable and productive. But the truth is messier. The debate about AI’s return on investment stubbornly persists, in part because companies still lack broad, measurable improvements they can cite, with spending justified company-by-company. Nvidia Corp. Chief Executive Officer Jensen Huang this week promised “insane” returns for investors in AI (calling those who questioned the technology’s potential returns “crazy”), but the next few months of initial public offerings have every possibility of proving a peak rather than the start of another boom..Read more:.CoreWeave’s IPO: A turning point for AI investment transparency.One set of cautionary signals comes from companies that are finding AI useful — but far too costly. Uber Technologies Inc. recently set usage caps on AI coding tools for its staff after the company blew through its entire annual AI budget in just a few months. Walmart Inc. capped staff use of a tool for generating spreadsheets and presentations, shifting away from giving employees an unlimited number of tokens, a data measurement for usage. One unnamed company is said to have accidentally spent $500 million on tokens in a single month, after introducing Anthropic’s Claude to its workforce but failing to provide usage limits. Plenty of grumbling about cost is coming from the financial sector too, as my colleague Lionel Laurent points out.Last year, the hot new trend among corporate users was tokenmaxxing, or using AI as much as possible to boost productivity and climb internal leaderboards. But more firms are now learning the hard way that the more tokens they spend, the more AI costs.The consultancy Bain & Company summed the problem up in a recent report titled Your AI Budget Is Growing. Your Returns Aren't. It found that while some companies reported cost reductions of 10% to 20% after using AI, many were plowing more money into the technology before those savings had fully materialized, essentially making a leap of faith. The bull case is that we’re still in an experimentation phase in which many companies aren’t measuring all the right things. AI’s biggest value could come not just from cutting costs but from producing more useful work. When I asked Anastasis Germanidis, the co-founder of Runway AI Inc., about how much his tools for generating video would shrink Hollywood budgets, he countered by suggesting that wasn’t the objective. “We’re going to have way more visual storytelling out there in the world,” Germanidis said. “More with the same amount of budget.”Tim Rocktäschel, co-founder of London- and San Francisco-based Recursive Superintelligence, tells me that many companies currently lack the skills to integrate AI properly. OpenAI and Anthropic have recognized that gap and are hiring more forward-deployed engineers, or software designers who also work with customers to help use AI in the most effective way, he said. For Rocktäschel’s own startup, AI agents have helped develop models that would have typically required hundreds of researchers using just a “dozen people.”But many other firms are finding that AI behaves more and more like a never-ending utility bill. Every query, hallucinated draft or bloated coding session has a cost because it is essentially metered software.That's partly the result of the sheer scale at which companies are consuming AI. Providers have always charged corporate customers by usage, but until recently, most employees accessed AI through flat-rate subscriptions or price-capped pilots, so token costs stayed under the radar. Now that firms are deploying agents and coding tools and rolling out AI company-wide, those per-query expenses are finally becoming large enough to bite..Read more:.FT: ‘Of course it’s a bubble’: AI start-up valuations soar in investor frenzy.Now that those costs are out in the open, metered pricing is forcing companies to confront a question that AI labs conveniently postponed as they built up more hype: whether their technology is useful enough to justify the bill. If many more AI users start to go the way of Uber and Walmart, that will be a problem for a market that until now has been riding astonishingly high. .© 2026 Bloomberg L.P..Sign up for your early morning brew of the BizNews Insider to keep you up to speed with the content that matters. The newsletter will land in your inbox every morning on weekdays. Register here.Support South Africa's bastion of independent journalism, offering balanced insights on investments, business, and the political economy, by joining BizNews Premium. Register here.If you prefer WhatsApp for updates, sign up to the BizNews channel here.

AI’s reality check: The “never-ending utility bill” bites

Related Stories

Inside SpaceX's $75bn IPO: AI, Mars & Musk's iron grip

Can the stockmarket swallow Anthropic, SpaceX and OpenAI?

BN Daybreak - Mon 1 June 2026: Oil Climbs; SpaceX IPO frenzy; BEE critiques; Chelsea Flower Show Gold

Anthropic's newest AI tools are making customers decidedly nervous