There is a trade being made in corporate budgets right now, and it is rarely described as plainly as Nvidia's chief executive Jensen Huang described it at GTC 2026. His test for whether an engineer justifies their salary: their AI tool spending should be at least half their annual compensation. For a $500,000 engineer, that means $250,000 in AI costs per year. Nvidia itself is working toward a $2 billion yearly bill for its engineering workforce. The man selling the compute has a natural interest in making that ratio seem normal. But the trade he describes, money moving from payroll to AI vendors, is exactly what is happening in companies everywhere.
The question nobody asked loudly enough is whether it is working. The honest answers are coming in, and they are not encouraging.
Gartner surveyed 350 executives at companies with over $1 billion in annual revenue, all of them already running AI or automation programs. Around 80% had cut headcount. When analysts compared those cuts to financial returns, they found no connection at all. The companies reporting strong returns were not the ones that cut the most people. They were the ones that invested more in the skills and roles that let humans direct AI systems effectively. Gartner analyst Helen Poitevin put it plainly: "Workforce reductions may create budget room, but they do not create return."
Oracle cut roughly 21,000 employees while posting record revenue of $67.4 billion, a 17% increase. The savings go directly to its AI data center buildout. Amazon, Google, Meta, and Microsoft have combined 2026 capital expenditure plans running between $115 billion and $200 billion each, almost all of it for AI infrastructure. These are not struggling companies making survival moves. They are profitable businesses using layoffs as a financing mechanism.
Klarna is the case study that made this concrete. The fintech replaced around 700 customer service workers with an OpenAI-powered assistant, froze human hiring for over a year, and built the AI-first model into its pitch to public market investors. The AI handled roughly two-thirds of all customer interactions. Resolution times dropped. On paper, the efficiency metrics looked strong. Then customer satisfaction scores fell on the interactions that actually mattered: disputes, complex refunds, financially sensitive conversations. Complaints accumulated. By May 2025, CEO Sebastian Siemiatkowski told Bloomberg: "We focused too much on efficiency and cost. The result was lower quality, and that's not sustainable." Klarna is now hiring humans again through a gig-style model, targeting students and remote workers. The rehiring costs, recruiting, onboarding, and training, consumed more than the original savings projected.
Uber offers the other half of the picture. The company gave AI coding tools to 5,000 engineers in late 2025 and burned through its entire 2026 AI budget by April, four months in. Engineers were ranked on internal leaderboards by how much they used the tools, which created a direct incentive to spend. Individual monthly costs ran between $500 and $2,000 per engineer. The company's own chief operating officer said the connection between all that AI-generated code and anything customers actually experience is "not there yet." Uber now caps engineers at $1,500 a month. Walmart applied similar rationing to its internal AI assistant after usage blew past projections.
The damage lands unevenly. Stanford's 2026 AI Index found that employment for software developers aged 22 to 25 fell nearly 20% since 2024, while developers in their 30s and older saw headcount grow. AI tools are not replacing software engineering. They are replacing the specific tasks that entry-level developers were hired to do: routine code, standard tests, basic bug fixes. Senior engineers now do those tasks themselves, using AI, without handing them off. The bottom rung of the career ladder is disappearing, which means the senior engineers those companies will need in ten years have nowhere to start.
The global version of this picture is harsher still. Huang's thought experiment is built around a $500,000 engineer, a salary that applies to a small fraction of American software workers and almost nobody in markets like Malaysia, the Philippines, or Indonesia. When you apply his spending ratio to those salary levels, the AI tools cost more than the person. The ratio was set in California and does not travel.
What Gartner's data and Klarna's experience point to is the same conclusion: the returns follow the companies that spend on people who use AI, not the companies that spend on AI to remove people. The CFOs now capping token budgets after spending a year's allocation in a quarter are rediscovering something basic. The staff were not the problem.