Uber Blew Its AI Budget in Four Months. Your HR Team Is Next

Uber gave 5,000 engineers access to an AI coding tool in December 2025. By April 2026 - just four months later - the company had burned through its entire annual AI budget. Not 50% of it. Not 75%. Every single dollar. Gone before summer.

The tool in question was Claude Code, built by Anthropic. The same Anthropic whose CEO Dario Amodei spent the first quarter of 2026 warning enterprises that AI would wipe out half of all entry-level white-collar jobs. The same Anthropic valued at almost $950 billion on the strength of an enterprise pitch built around the idea that AI would make corporations more cost efficient by making the humans optional.

SPONSORED

Your employees know the truth. Does your EVP? At Fathom we measure the "Credibility Gap" between your promise and their reality.

Close the Gap

It turns out AI also makes budgets optional. And that is a problem that HR and talent acquisition teams have not yet begun to reckon with.

The Meter Is Running

Here's a scary sentence that should be pinned above every CFO and CHRO's desk in 2026.

Sam Altman told an interviewer in March: "We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter."

Not a licence or a seat, a meter. The same model your energy provider uses. The same model that has produced some of the largest utility bills in corporate history when consumption runs ahead of forecast. And we all know what's been happening to energy bills recently.

This is the detail that got buried under two years of human replacement narrative. While the token salesmen - Altman, Amodei, and the enterprise sales teams behind them - were selling the efficiency revolution to employers, they were also quietly shifting their pricing models from flat subscriptions to usage-based token consumption. Every prompt, document, and AI-assisted decision... metered and charged.

That fancy new AI model that's smarter than the one it just superseded... yeah, it's WAY more expensive, too. Most don't know.

The labour replacement promise was the pitch - it meant incredible cost savings leading to dreamland profits. The meter is now the business model. And the meter does not care whether headcount went up or down.

FIGURE 1 - Uber budget burn

What Bill Shock Actually Looks Like

The Uber case is the most dramatic public data point so far, but it is far from unique.

One healthcare enterprise consumed 1 trillion tokens over six months, generating more than $6 million in unplanned costs before the finance team even understood what was driving the invoices. A developer at another company left an AI coding agent running overnight and woke up to a $6,000 charge for a single session. Another hit $4,200 in API fees over a single weekend.

I recently caught up with a good friend from a top tier professional services firm who told their first bill from Anthropic was almost TEN TIMES what they'd expected it to me. It was millions in unplanned costs but I doubt the bill shock story will go public.

These are not mere rounding errors. These obscene bill shocks are a consequence of a pricing model that most procurement teams did not read carefully enough when they signed the contracts. And even if they had, the AI hype train would have probably seen the contracts signed anyway.

The average organisation's OpenAI API spend reached $384,500 annually as of April 2026 - and AI-native application spend grew 108% in 2025 alone. 78% of IT leaders recently surveyed reported unexpected charges tied to consumption-based AI pricing models. Not a minority experience or fringe cases, but nearly four in five being caught out with bigger than expected bills.

The reason isn't user error and nobody could blame the employees who're being pushed to use the tools more and more - the cost mechanisms are baked into the tools and aren't transparent. Enterprise AI deployment audits consistently find that retry logic, context window management, and retrieval augmentation add between 40% and 60% on top of the token costs most teams are tracking. The bill that surprises most enterprises isn't the one from launching AI. It's the one from running it every day.

GitHub Copilot, the dominant AI developer tool, is moving its entire pricing model to usage-based billing on 1 June 2026. One developer reported their projected monthly cost rising from around €67 in April to roughly €966 under the new model. That's just one lone developer.

Token consumption across enterprise has increased 13x since January 2025. That's an insane increase in little over a year. And, you guessed it... budgets haven't followed the 13x increase during the same time.

FIGURE 2 - The pricing paradox

The HR and TA Exposure

Of course, this isn't just a tech story, it's a human capital story too. After all, the AI is here to supposedly replace expensive human labour, or at very least make it more efficient.

The assumption baked into most enterprise AI investment decisions - and certainly into most HR and TA technology roadmaps for 2026 - was that AI would reduce costs. That is the pitch that was made and the pitch that was believed. Josh Bersin's April 2026 analysis projects 30 to 40% of existing HR roles can be automated, a finding that was used in many organisations to justify both the AI investment and the headcount reduction that preceded it.

What nobody modelled with any rigour was the consumption cost of actually running those AI systems at scale. The pilots were cheap but normal usage isn't. And the gap between the two is why so many CIOs and CFOs are now in a state of shock.

The data is now highlighting the gap directly. Gartner's May 2026 analysis found that roughly 80% of organisations report workforce reductions tied to AI, yet the same organisations cannot demonstrate ROI from those reductions. Gartner also forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs and unclear business value. Separately, the Harvard Business Review summary of Gartner's 2026 research found that only 1 in 50 AI investments delivers transformational value, and only 1 in 5 delivers measurable ROI at all.

Meta, which cut 8,000 employees earlier this year and closed 6,000 open roles specifically to free up budget for AI, is now discovering - as every enterprise will eventually discover - that the AI infrastructure it freed that budget to fund is itself an ongoing, usage-based, non-linear cost. You do not buy AI capability once. You rent it by the token, forever, at a price that rises with every additional use case you deploy it for.

SPONSORED

Employer branding now has a measurement standard. The Talent Gravity Standard is a six-driver framework for quantifying employer attractiveness and the gap between brand promise and employee experience.

Read the Standard

For HR specifically, the exposure arrives through three channels...

The first is direct: the AI tools already deployed in recruitment stacks. ATS integrations, AI screening tools, automated outreach platforms, conversational AI for candidate experience - all of these are token-consuming systems. 87% of companies now use AI in some part of their recruitment process. Most of those contracts were signed during the 2024–2025 experimentation wave. Most enterprise AI contracts were signed when AI pricing was the wild west of flat tiers and generous allowances. Independent ATS pricing research published in February 2026 found that AI-powered ATS platforms charge an average of 68% more than non-AI equivalents, and that the gap between the advertised "starting" price of recruitment software and what teams actually pay in production runs to roughly 247%. Renewal season is arriving and the pricing conversations are going to be different this time.

What should HR teams do - slash the AI spend and rehire the humans or double down on the AI? I know where I'd put my money.

The second is indirect: the budget competition. Every dollar that goes to AI infrastructure is a dollar that doesn't go to talent acquisition, employer brand, learning and development, or any other HR function that requires sustained investment. When Uber's COO explicitly framed the question as a trade-off between AI token spend and headcount, he was naming a choice that CEOs across every industry are running the numbers on. HR teams that have already absorbed headcount reductions justified by AI efficiency will now face a second pressure: the AI itself costs a lot more than projected.

The third is structural: the pivot cost. Only 24% of TA leaders expect headcount increases in 2026. Only 30% expect their TA budget to grow at all. Yet 59% plan to increase TA technology spending. The function is shrinking its people and growing its tooling at the same time, on the assumption that the tooling will deliver the savings (and do the work). When the bill shock forces a budget correction, and it will, the functions that have been allowed to atrophy will face the most painful rebuilds.

FIGURE 3 - Three channels of exposure

The Pitch and the Price

I think it's worth being direct about the connection between the replacement narrative and the cost crisis, because the two are not separate phenomena - they are two acts of the same story.

Act one: the token salesmen sell enterprises on a vision of AI replacing human labour at scale. The efficiency gains are real. The headcount savings are real. The investment is justified. High on AI hype CEOs approve it all. HR teams shrink as a result and recruiting pipelines get paused. The employer brand goes silent.

Act two: the usage bills arrive. Uber's COO, confronted with a full-year budget exhausted in four months, admitted he could not draw a clear line between AI token consumption and useful consumer features shipped. "That link is not there yet," he said. "It's very hard to draw a line between one of those stats and 'Okay now we're actually producing like 25% more useful consumer features.'" The ROI promised in AI sales pitches is proving harder to demonstrate than the very real consumption generating the invoices.

What Uber disclosed - and what dozens of organisations will soon disclose over the next 12 to 24 months (mark my words) - is that the token economy is not a one-time cost of transformation. It is a permanent, escalating, consumption-based operating expense. Gartner forecasts AI agent software spending will reach nearly $207 billion in 2026, up more than 139% from $86.4 billion in 2025. That trajectory does not flatten, it compounds.

Microsoft, which poured $80 billion into AI data centres and embedded Copilot across its entire product line, is now quietly pulling back in several enterprise scenarios - not because the AI doesn't work, but because the bills started to arrive. In a number of real-world deployments, running AI costs more than employing the humans it was supposed to replace. And, as AI models get more sophisticated (and expensive) and usage increases, so will those costs.

If huge swaths of the workforce aren't replaced and the huge cost savings aren't realised in the way they were sold... this could all backfire really badly.

SPONSORED

HR has CIPD. Marketing has CIM. Employer branding has nothing. Until now. The IEBP is being established to give the discipline the professional infrastructure it has always lacked.

Learn what we're building

What Comes Next for Corporate HR?

The bill shock will force a reckoning that the replacement narrative and AI hype-train delayed. And when the reckoning arrives - not as a theoretical risk but as a budget line that the CFO is freaking out over - HR, TA and EB teams will face a moment of uncomfortable clarity and deep regret.

The organisations that eliminated human capability on the promise that AI would substitute for it will discover that the substitution is partial, the costs are ongoing, and the talent they let go has moved on. Rebuilding takes time and budget. And rebuilding requires an employer brand proposition that has, in many cases, been sorely neglected for two years on the assumption that the hiring freeze was permanent. The macro labour pattern underneath this story is not new. It is the same paradox that has shaped every previous wave of automation, and it has rarely produced the lasting headcount savings corporations expect from it.

The organisations that maintained some version of their talent infrastructure - the employer brand, the pipeline, the in-house TA capability - will be better positioned to respond when the reckoning arrives. Not necessarily because they were smarter, but because they were more cautious about betting it all on a pitch that was built by people whose business model requires consumption to grow without limit.

Sam Altman called it a meter because he was describing a utility. Utilities have bills. And the bill, for a great many organisations that moved fast and didn't read the contract or do the maths carefully, is going to be considerably larger than the efficiency savings they were promised when they signed it.

HR didn't create this situation. But HR will be the schmucks that are left to solve it.

TAKEAWAYS

SOURCES

Have something to say that you can't say on the record? It belongs in The Exit Interview.

Jump to the form