‘The cost of compute is far beyond the costs of the employees’: Nvidia exec says right now AI is more expensive than paying human workers | Fortune

Meta introduced final week in a memo that it plans to put off 10% of its workforce, about 8,000 workers, in addition to scrap plans to rent for six,000 open positions. It’s a part of an effort to “run the corporate extra effectively and to permit us to offset the opposite investments we’re making,” in line with the memo. Microsoft has supplied hundreds of its personal workers a voluntary buyout, the most important the corporate has ever supplied.

Different tech headers, nevertheless, counsel that proper now, AI isn’t saving firms cash on labor; it’s truly costing them greater than the people they at present make use of.

“For my workforce, the price of compute is much past the prices of the workers,” Bryan Catanzaro, vice chairman of utilized deep studying at Nvidia, not too long ago informed Axios.

An MIT research from 2024 backs up Catanzaro’s expertise. Analyzing the technical necessities of AI fashions wanted to carry out jobs at a human stage, researchers discovered that AI automation can be economically viable in solely 23% of roles the place imaginative and prescient is a main a part of the work. Within the remaining 77% of the time, it was cheaper for people to proceed their work.

On different cases, AI has confirmed to be fallible, with one engineer saying an AI agent destroyed his database and community because of what he known as “overuse.” 

Regardless of no clear proof on AI bettering productiveness and, in line with the Yale Price range Lab, no widespread knowledge to assist the thought of AI displacing jobs, Large Tech corporations have continued to pour cash into AI, saying $740 billion in capital expenditures this yr to date, in line with Morgan Stanley, a 69% enhance from 2025. The magnitude of spending has triggered some firms to rethink their finances altogether.

“I’m again to the drafting board as a result of the finances I assumed I would wish is blown away already,” Uber chief know-how officer Praveen Neppalli Naga informed The Data earlier this month, referring to the rideshare large’s pivot to AI coding instruments, akin to Anthropic’s Claude Code.

This enhance in spending has coincided with extra layoffs within the tech sector. In accordance with knowledge from Layoffs.fyi, there have been greater than 92,000 layoffs in tech in 2026 to date throughout practically 100 firms. The speed of those workforce reductions is already far outpacing final yr, which noticed about 120,000 layoffs over the yr.

The continued AI spending and layoffs, at the same time as human labor stays cheaper, expose a significant discrepancy within the economics of AI, mentioned Keith Lee, an AI and finance professor on the Swiss Institute of Synthetic Intelligence’s Gordon College of Enterprise.

“What we’re seeing is a short-term mismatch,” Lee informed Fortune.

The AI-labor price stability

In accordance with Lee, the price of utilizing AI has remained much less environment friendly than human labor attributable to {hardware} and vitality prices elevating working prices for suppliers. At its present tempo, AI expenditures might attain $5.2 trillion by 2023, with $1.6 trillion from knowledge heart spending and $3.3 trillion from IT tools, in line with McKinsey knowledge. Spending might surge to $7.9 trillion by 2030 at an accelerated tempo. In the meantime, charges for AI software program have elevated by 20% to 37% over the previous yr, spending administration agency Tropic famous in December.

AI firms may additionally be shedding cash because of their flat subscription mannequin, Lee famous, with fastened subscription charges failing to cowl working prices for heavy AI customers.

“In consequence, some corporations are starting to re-evaluate AI not as a transparent cost-saving substitute for labor, however as a complementary software—no less than till the fee construction stabilizes,” he mentioned.

Whereas AI might price greater than human labor right this moment, there might be warning indicators of a tipping level towards AI’s financial viability. For one, Lee indicated, the price of utilizing AI will change into considerably decrease, with performing inference—how AI analyzes knowledge—for a big language mannequin with 1 trillion parameters plummeting by greater than 90% over the following 4 years, in line with a report final month from analyst agency Gartner. AI infrastructure will possible enhance, and mannequin designs and {hardware} provide will observe. AI firms may also possible change how they worth their instruments, switching from a flat subscription to usage-based pricing, Lee predicted. 

However the way forward for AI’s financial viability may also rely on if the know-how proves its value. It should show itself dependable, with fewer hallucinations and a lowered want for human oversight, successfully integrating into an organization’s infrastructure, in line with Lee. Federal Reserve knowledge reveals about 18% of firms had adopted AI instruments as of the top of 2025, a 68% progress within the adoption charge since September 2025.

“It’s not nearly AI changing into cheaper than people,” Lee mentioned. “It’s about changing into each cheaper and extra predictable at scale.”