Nvidia Just Proved ‘AI Efficiency’ Is Fake as Layoffs Rise: ‘The Cost of Compute Is Far Beyond the Cost of Employees’

For the past couple of years, we’ve all been spoonfed the same narrative: AI is making companies leaner, faster, and more efficient. Investors were pretty keen to jump on the bandwagon, which is why valuations have skyrocketed and everybody in big tech is in the middle of a global arms race for AI infrastructure.

Unfortunately, this idea that AI is better for a company’s bottom line has also been used to rationalize mass layoffs. Everybody inherently assumes that it’s cheaper to install AI bots than it is to pay white collar workers, and so Wall Street has been rewarding the companies that are simultaneously cutting headcount and ramping up AI spending. 

But a leading AI executive has just flipped the script.

In a recent interview with Axios, Nvidia VP of applied deep learning Bryan Catanzaro (NVDA) made a surprisingly blunt admission: In a lot of cases, the cost of AI compute is actually greater than the cost of human employees. Just let that sink in for a minute, because it goes against everything investors have been told for the last two years.

It also makes you wonder about the economics sitting at the center of the global AI boom — because if the numbers don’t actually align, then a lot of the assumptions underpinning this market cycle may turn out to be catastrophically wrong.

Researchers Have Been Warning About this for Years

This all makes sense at face-value. After all, why pay for training, salaries, office costs, or vacation days when agentic AI can automate everything?

Well, now we have our answer: Executives and investors had absolutely no idea how expensive AI could be when it’s running at scale.

It turns out that it takes a ridiculous amount of computing power to train and deploy advanced AI models. But on top of that, you need obscene amounts of electricity, water and cooling infrastructure, specialized chips, and expensive networking equipment. And once all that stuff is in place, the costs compound quickly.

We’re all talking about it now because Nvidia’s own team has plainly admitted AI is super expensive. But researchers at MIT published hard numbers on this over two years ago — and they reached the same stark conclusion.

Their 2024 study found that AI automation is only economically viable in around 23% of jobs where visual processing plays a major role. In the other 77% of cases, it was actually cheaper for humans to do the work themselves. Why?

Unlike people, AI doesn’t switch off and go home at 5 o’clock. It continues to draw power all night, and GPUs depreciate incredibly quickly. Data centers need constant maintenance and regular upgrades that can cost millions and millions of dollars.

That’s why some companies are burning through their AI budgets like wildfire. We’re not even done with Q2, and Uber’s chief technology officer has already blown through his full AI budget for the year because of how high token costs have gotten.

Translation: Technical capability doesn’t automatically translate into economic viability. Every single query that gets processed by a large language model costs money. If you multiply that by the billions of daily interactions AI is facilitating, that “cheap digital worker” narrative really dies a death.

This is where we reach the intriguing problem that’s deeply embedded in the AI boom. 

If AI Efficiency Is Fake, Why Are Workers Still Getting Laid Off?

We’re not just living through a golden age of AI in 2026. We’re also enduring a dark age of mass layoffs. It feels like another round of redundancies gets announced every couple of days.

Oracle (ORCL) is eliminating up to 30,000 employees so it can redirect resources toward AI infrastructure, Jack Dorsey cut nearly half of his workforce over at Blocks (XYZ), and Atlassian has reduced its global headcount by 10% to reposition its software for the “AI era.” Coinbase (COIN) has laid off 700 employees so AI agents can start running teams, Amazon (AMZN) has eliminated some 14,000 corporate roles, Cisco (CSCO) has let go of 4,000 workers — you get the idea.

But there’s definitely more to all this than meets the eye.

CEOs are currently facing a perfect storm of unfair market expectations, corporate signaling, and competitive pressures. Shareholders want executives to be able to showcase how they’re aggressively embracing AI to prove they’re keeping up with the rest of the pack.

As a result, we’ve hit a situation in which the C-level is so afraid of getting behind the curve that they’ve lost sight of long-term costs. Layoffs are a key part of this theater performance. Cutting workers while investing in AI seems like a shrewd and forward-looking decision to investors. 

But that’s created a deadly dynamic where huge corporations are diving head-first into AI initiatives before the numbers are fully proven. From the outside looking in, it seems like there’s not going to be many net automation gains here. A lot of these layoffs don’t need to happen — or at the very least, they aren't happening because of AI — which has led to accusations of “AI washing."

To be honest, it’s probably going to get worse before it gets any better. AI is genuinely disrupting labor markets, and so it's the perfect scapegoat for any business that’s desperate to make cuts or justify its high valuation in 2026. 

But the issue investors need to be paying attention to right now is sustainability. Eventually, all of these companies will have to justify their rampant AI spend with real economic returns. User growth and engagement metrics won’t cut it, and executives are going to need to demonstrate actual profit.

Even AI executives are ready to admit that markets are underestimating how expensive this transition period is turning out to be.  

That doesn’t mean the AI bubble is going to pop any time soon. But it does mean that the path forward is going to be a lot more volatile and expensive than markets could have expected — because right now, none of the companies replacing workers are gaining anything by investing in AI. The only winners here are the companies selling them compute.