Worldwide AI spending will total $2.52 trillion in 2026, up 44% from 2025, according to Gartner. AI infrastructure spending alone is projected to hit $487 billion this year, representing 53% year-on-year growth, according to IDC.
There’s capital swimming toward a set direction, with near-universal enterprise adoption. What’s less clear is whether the returns are following.
Uber
Uber rolled out Claude Code to its engineering organisation in December 2025. By March 2026, 84% of engineers were classified as agentic coding users, up from 32% in February.
The company burned through its entire 2026 Claude Code and Cursor budget in four months, with monthly API costs per engineer running between $500 and $2,000.
Except they’re not sure it’s resulted in anything. Referring to the connection between rising token consumption and measurable improvements, Uber’s COO Andrew Macdonald said on a podcast recently, “That link is not there yet”.
He referenced what some are calling “tokenmaxxing“—high AI usage volume producing no proportional output gains.
For context, Uber spent $951 million on R&D in Q1 2026 alone, up 17% year-on-year. 95% of its engineers use AI tools monthly, with 70% of its committed code now AI-generated, according to a LinkedIn post by CTO Praveen Neppalli Naga.
And they still can’t draw a direct line from that activity to useful product improvements. That looks like a planning problem, not a technology problem.
ROI
Uber isn’t the exception. MIT’s 2025 State of AI in Business report found a 95% failure rate for enterprise generative AI projects, defined as not having shown measurable financial returns within six months.
The average large enterprise now spends over $110 million per year on AI. Yet according to a 2025 BCG study, only 5% of 1,250 surveyed enterprises were achieving substantial ROI from those investments at scale.
Zapier’s Enterprise AI Benefits survey found that while 97% of US enterprises use AI, only half report that the benefits are widespread across teams. 73% have a formal ROI tracking process, yet 91% still struggle to measure results.
Meanwhile, 61% of the 3,700 senior business leaders in Kyndryl’s 2025 Readiness Report feel more pressure to prove ROI on their AI investments now than a year ago.
But perhaps it’s all to be expected. Gartner classifies AI as being in the “Trough of Disillusionment” throughout 2026, noting that “the improved predictability of ROI must occur before AI can truly be scaled up by the enterprise.” Maybe this is all part of the correction.
Why firing people to fund AI tokens is a bad trade
Nearly 55,000 job cuts were directly attributed to AI in 2025, according to Challenger, Gray & Christmas (out of more than 1.17 million layoffs by October 2025). Amazon targeted approximately 30,000 corporate workers for layoffs in October 2025 while simultaneously committing over $100 billion in capex, much of it to AI infrastructure.
They weren’t alone in the layoffs—dozens of other companies from Atlassian to Cisco and Coinbase decimated their workforces too.
But the pattern has started reversing: a recent survey found that 29% of 2,000 hiring managers reopened positions previously eliminated by AI.
55% said they planned to increase the number of contract or temporary workers within the first half of 2026; 60% said the same for full-time workers.
Meanwhile, Gartner predicts half of companies cutting customer service staff due to AI will rehire for similar roles by 2027.
The math around all of this is uncomfortable. Severance costs money, and a damaged employer brand makes the next recruitment round more expensive.
Institutional knowledge doesn’t transfer cleanly, and when you need to bring people back six months later, the market has moved and you’re negotiating from a weaker position.
| Stage | Cost type |
| Redundancy payouts and severance | Immediate cash outflow |
| Damaged employer brand | Higher future recruitment cost |
| Lost institutional knowledge | Operational drag, 12-24 months |
| Rehiring at a market premium | Often 1.5-2x original salary |
| Ramp-up time for new hires | Productivity deficit during onboarding |
There’s also a conceptual problem with the underlying logic around AI-related layoffs. Companies that cite AI productivity gains to justify headcount cuts are making two contradictory claims at once.
If AI has amplified what your team can produce, the rational move is to direct those amplified capabilities at bigger problems: new products, new markets, deeper customer work.
Cutting the team that just became more capable is not an efficiency decision. It’s an admission there was no real plan for the productivity increase.
Where your AI budget needs to go
The 5% of enterprises achieving measurable AI returns share a consistent approach: they treat training and enablement as infrastructure, not a perk, and they don’t buy AI licences expecting adoption to follow.
An AI budget that accounts only for compute, tooling, and licences is incomplete. The full budget needs lines for:
Training—sustained, role-specific development in how AI tools fit specific workflows. A one-hour onboarding session isn’t training. The assumption that employees will self-teach is the primary reason most rollouts plateau.
Hiring—if your AI strategy requires new capabilities (prompt engineering, workflow design, AI governance, output review), those capabilities live in people. Token spend doesn’t produce them.
Workflow optimisation—AI performs well inside well-designed processes and poorly inside broken ones. Redesigning those processes requires human judgment and time.
Governance and guardrails—the regulatory and reputational exposure from ungoverned AI output at enterprise scale has real life consequences (and examples). Prevention costs less than remediation.
Output quality control—AI generates, but humans complete and verify. Review isn’t optional.
AI token spend without these foundations produces expensive, hard-to-measure activity—precisely what Uber’s COO described. The companies building durable AI capability are investing in both sides of the equation.
“AI adoption is fundamentally shaped by the readiness of both human capital and organisational processes, not merely by financial investment.” — John-David Lovelock, Distinguished VP Analyst at Gartner.