An Nvidia Exec Just Said AI Costs More Than His Employees

If anyone has the incentive to tell you AI is cheap, it's Nvidia — the company that sells the chips. So when Bryan Catanzaro, Nvidia's VP of Applied Deep Learning, told Axios plainly that "for my team, the cost of compute is far beyond the costs of the employees," that's worth sitting with. It's not a skeptic's talking point. It's an insider's admission.

The math most leaders haven't run

That spending isn't hypothetical — it's already reshaping budgets. Uber's CTO, Praveen Neppalli Naga, put it bluntly when describing the company's pivot to AI coding tools: "I'm back to the drawing board because the budget I thought I would need is blown away already." That's a technology leader at a well-resourced company getting surprised by AI costs. Most mid-market organizations have far less budget cushion to absorb that kind of surprise.

The counterintuitive research: humans are still cheaper, most of the time

Here's the part that should reshape how you think about your own AI roadmap. An MIT CSAIL study found that AI automation is economically viable in only 23% of roles where vision is a primary part of the work. In the other 77% of cases, it's still cheaper for humans to do the job.

That number cuts against the popular narrative in both directions. It's not "AI is unstoppable and will replace everyone economically viable to replace." It's also not "AI is all hype." It's something more useful and more specific: AI is economically superior in a real but bounded set of use cases — and blanket AI deployment across every function, without that analysis, is how organizations end up with Uber's problem instead of a competitive advantage.

"What we're seeing is a short-term mismatch." — Keith Lee, AI and finance professor, Swiss Institute of Artificial Intelligence, on why AI costs remain less efficient than human labor once hardware and energy costs are counted

The layoffs that got ahead of the math

This mismatch hasn't stopped the workforce reductions. Meta laid off 10% of its workforce — roughly 8,000 employees — and scrapped 6,000 open positions in April 2026. Across nearly 100 companies, more than 92,000 tech layoffs have happened in 2026 alone, per Layoffs.fyi. And yet the Yale Budget Lab reports no clear, widespread evidence yet that AI is actually driving measurable productivity gains at the scale that would justify those cuts.

That's the gap that should concern every leadership team: workforce decisions are being made ahead of the economic data that would justify them. Some of that will be validated in hindsight. Some of it won't — and the organizations that ran the numbers first will be the ones that know which is which before it's a crisis.

What this means before you sign the next AI contract

Model the total cost, not the license fee. Compute costs, integration overhead, and ongoing model costs routinely dwarf the sticker price of an AI tool subscription. Run the same discipline you'd apply to any capital investment — not just a SaaS renewal.

Identify where AI is genuinely cheaper before you deploy it. The MIT CSAIL finding is a useful gut-check: in the majority of roles, the economics still favor keeping a human in the seat. Deploy AI where the cost-benefit case is real, not where it's assumed.

Don't let cost pressure substitute for a change plan. As I wrote in Why 42% of AI Projects Never Make It to Production, most AI initiatives fail on methodology, not technology. A rushed AI deployment driven purely by cost-cutting pressure is exactly the pattern that produces abandoned projects and re-hiring six months later.

Build the ROI case before the layoff decision, not after. If workforce reductions are being justified by AI capability that hasn't been validated yet, that's a governance and change management gap — not just a finance one.