AI Isn’t Cheaper Than Hiring People. Nvidia Just Said So Out Loud.
A senior Nvidia VP said the cost of compute at his company is "far beyond" the cost of employees. That's worth reading twice — especially this week.
Bryan Catanzaro, VP of applied deep learning at Nvidia, told Axios on April 28: “For my team, the cost of compute is far beyond the costs of the employees.” Not a hedge. Not a future projection. An explicit, present-tense statement that AI infrastructure has, at his company, already crossed the threshold where it costs more than the humans using it. This is Nvidia. The company whose chips power the AI buildout. The vendor is telling you the product is expensive.
The dominant story in the tech industry right now is that companies are cutting headcount because AI has made the remaining workers productive enough to make the math work. That story is probably true in certain pockets — and it will probably be more true in three years than it is today. But as the complete picture of what’s happening right now, it doesn’t hold. What Catanzaro described is a company where AI compute has become a line item that rivals the entire payroll — and this is at Nvidia, the company selling the hardware that generates those costs. If the vendor of the product is telling you it’s more expensive than your workforce, you should factor that into the narrative you’ve been repeating about AI-driven efficiency. The vendor has no incentive to make the product sound unaffordable. He said it anyway.
The layoff numbers make this harder to ignore. The tech sector has recorded over 92,000 layoffs in 2026 across nearly 100 companies, and the stated rationale in most cases involves some version of AI-driven restructuring. Companies are paying simultaneously to reduce headcount and to scale AI infrastructure. The underlying bet is that the second cost will eventually justify the first — that compute will get cheaper, productivity will compound, and the math will eventually work out. That may be correct. But right now, for a significant portion of these organizations, both bills are coming due at the same time, and the AI spend is the one that’s growing.
A McKinsey analysis projects $5.2 trillion in AI data center capital expenditures alone by 2030, with traditional IT infrastructure adding another $1.5 trillion on top of that — roughly $7 trillion in total compute buildout by the end of the decade. That number clarifies something: the assumption that AI is cheaper than labor is not a present-tense fact. It’s a bet on a future cost curve. Cutting 92,000 jobs on a projection is a different thing than cutting them on arithmetic that’s already resolved. The first is a wager. The industry has mostly been describing it as the second.
The assumption that AI is cheaper than labor is not a present-tense fact. It’s a bet on a future cost curve.
There’s also a structural asymmetry that most coverage misses. Nvidia sells the infrastructure that generates these costs — they make money whether or not the AI spend eventually pays off for their customers. For the companies actually running the compute budgets, the math is considerably less forgiving. Every dollar going to a data center lease or an API bill is a dollar not yet producing revenue. The cost curve on compute does come down over time — competition from AMD, Google, and custom silicon projects at the major hyperscalers are all designed to reduce Nvidia dependency specifically. But “cheaper eventually” is a different business model than “profitable now,” and right now most companies are running the first model while presenting it in earnings calls as the second.
Jensen Huang’s remarks at a Stanford panel on April 20 add another dimension: AI agents are more likely to “harass and micromanage” workers than to replace them outright. So Nvidia’s VP says AI costs more than employees. Nvidia’s CEO says AI isn’t taking your job — it’s going to be a very persistent supervisor. Both statements are probably more accurate than the version of this story that’s been running in most headlines, which assumes AI is simultaneously cheap, effective, and reducing headcount. Right now it appears to be none of those three things cleanly, and two of Nvidia’s most senior executives are among the better sources for that fact.
None of this is an argument that AI spend is irrational, or that the cost curves won’t come down. They probably will. The trajectory of compute costs over the last decade points in that direction, and the companies making these bets are making them on something more than pure faith. What this is an argument against is the confidence with which companies are treating “AI replaces workers because AI is cheaper” as a settled proposition when one of the industry’s most important insiders is saying, plainly, that it isn’t — at least not yet, and not at Nvidia. Catanzaro said the quiet part out loud. The industry’s accounting should catch up.
Sources: Fortune · Crunchbase · McKinsey · Fortune / Jensen Huang
