Compute Costs Exceed Employee Pay, Nvidia Executive Says: AI Is Still More Expensive Than Human Workers
Headline and Lead
Silicon Valley layoffs suggest a rapid shift from human workers to machines. Field evidence tells a different story.
Bryan Catanzaro, vice president of applied deep learning at Nvidia, was blunt in an interview with Axios.
For his team, compute costs are now “far beyond the costs of the employee.”
That admission comes from one of the most powerful companies in the global AI supply chain.
So why are tech giants laying off staff while spending hundreds of billions more on AI?
This report, based on Fortune and supplementary sources, examines short-term economics, investment policy, and future pricing.
When Layoffs Don’t Match Real AI Costs
Meta disclosed plans earlier this year to cut roughly ten percent of its workforce, about eight thousand people.
The company also paused hiring for six thousand open roles.
Meta leaders said the goal was to run the company more efficiently and fund new investments.
Microsoft offered its largest voluntary buyout program in company history to thousands of employees.
Around the same time, Uber said it had already exhausted its 2026 budget for AI coding tools by April.
Praveen Neppalli Naga, Uber’s chief technology officer, told The Information that costs had “exploded” beyond what finance teams expected.
The company launched an internal leaderboard to encourage use of Claude Code.
Microsoft, according to The Verge, canceled most direct Claude Code licenses for employees.
The corporate replacement will be GitHub Copilot CLI.
AI tools became popular so fast that spending outpaced financial planning.
According to Layoffs.fyi, more than 118,000 tech workers were laid off in 2026.
That figure has nearly matched all of 2025, about 120,000, in half a year.
The contradiction is stark: fewer human workers alongside rising bills for intelligent software.
Scientific Evidence: AI Is Not Cost-Effective in Most Roles Yet
A MIT study in 2024 found AI automation was cost-effective in only 23 percent of vision-related roles.
In the remaining 77 percent, continuing with human labor was cheaper.
That finding aligns with the experience of large-scale operations leaders.
Hardware, energy, and data center maintenance push the final bill higher.
Some engineers have also reported serious failures from autonomous agents.
One well-known case involved database and network damage after “overuse” of an agent.
Layoffs alone cannot explain a full story of workforce replacement.
In many organizations, AI remains an expensive complement, not a cheap substitute.
The $740 Billion Big Tech Capex Wave
Morgan Stanley estimates tech giants have announced $740 billion in capex so far in 2026.
That figure is 69 percent higher than in 2025.
McKinsey projects AI spending could reach $5.2 trillion by 2030.
In an accelerated scenario, the number could approach $7.9 trillion.
$1.6 trillion would go to data centers and $3.3 trillion to IT equipment.
Tropic reported in December 2025 that AI software pricing rose 20 to 37 percent in one year.
Flat subscription models often fail to cover operational costs for heavy users.
Keith Lee, professor of AI and finance at the Swiss Institute of Artificial Intelligence, sees a “short-term mismatch.”
Companies are evaluating AI as a complementary tool, not cheap labor replacement.
That view is likely to persist until cost structures stabilize.
When Will AI and Labor Reach Economic Balance?
Gartner predicts inference costs for a one-trillion-parameter language model will fall more than 90 percent in four years.
Better model design, hardware supply, and infrastructure optimization will ease cost pressure.
Lee expects pricing to shift from flat subscriptions toward usage-based models.
But cheaper is not enough; AI must become predictable and reliable at enterprise scale.
Reducing hallucinations and human oversight needs are prerequisites for broader adoption.
Federal Reserve data shows about 18 percent of U.S. firms had adopted AI by late 2025.
That rate grew 68 percent from September 2025.
Lee stresses: “It’s not just about being cheaper than humans; it’s about being cheaper and predictable at scale.”
The Yale Budget Lab has not yet confirmed widespread job displacement driven by AI.
Current layoffs reflect financial pressure, organizational redesign, and bets on AI’s future.
HR leaders should model scenarios where demand for specific skills returns.
Cutting staff without a skills roadmap can multiply rehiring costs later.
Global Scene: U.S. Leads Adoption, but Cost Pressure Is Worldwide
American companies are adopting AI faster than other major economies. Read our tech economy analysis.
Federal Reserve figures show adoption growth has been striking in recent months.
Reports of costly AI are not limited to one company or one industry.
Amazon, Meta, Microsoft, and Google are experiencing layoffs and heavy capex at the same time.
Amazon engineers have referenced a $200 billion data center budget in internal discussions.
That figure alongside tens of thousands of job cuts raises questions of social fairness.
Should society bear the transition cost of AI when short-term returns remain unclear?
Governments in Europe and Asia are subsidizing AI infrastructure to stay competitive.
Energy and water pressure from data centers has put sustainability at the center of public policy.
The equation is not only human versus machine; it is also today versus tomorrow.
Message for Decision Makers: Investing Without the Math Is Risky
For middle and senior managers, the first lesson is that AI hype cannot replace total cost of ownership analysis.
License fees, API usage, on-premise GPUs, energy, maintenance, and oversight staff belong on one sheet.
The second lesson: mandatory adoption without spending caps can burn an annual budget in months.
Uber is a live example of a leaderboard without cost controls.
The third lesson: short-term layoffs may free cash but can destroy critical skills.
Vista Equity Partners has warned against cutting intern programs in the name of AI alone.
Successful organizations deploy AI in defined workflows with clear KPIs and usage ceilings.
They keep human-in-the-loop processes until reliability is sufficient.
For readers in economics and technology, this period is a recalibration of expectations.
Artificial intelligence is revolutionary, but revolutions are not cheap on day one.
The Fortune report reminds us that even at Nvidia, where GPUs are sold, internal compute costs more than human staff.
That reality overturns the simple narrative that machines are cheap and people are expensive.
The future may look different; today, the math must be written honestly.
No comment