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

Tech firms are cutting jobs even as AI spending surges, with Meta planning 8,000 layoffs and 6,000 fewer hires, Microsoft offering a large voluntary buyout, and more than 92,000 tech layoffs reported in 2026 so far. The article argues that AI is still often more expensive than human labor because of compute, energy, and infrastructure costs, with a MIT study saying automation is economically viable in only 23% of vision-based roles. It suggests the current AI adoption wave is creating a short-term mismatch between rising spending and unclear productivity gains.

Analysis

The key market implication is not that AI is already substituting labor, but that management teams are treating AI spend like a strategic arms race even when unit economics are still unfavorable. That creates a near-term margin paradox: companies can justify layoffs as optics and reallocation, while the actual earnings drag shifts into cloud, inference, networking, and power rather than payroll. The second-order winner is the infrastructure layer, not the software layer, because every failed or overused deployment increases compute intensity and raises switching costs for hyperscaler ecosystems.

This also argues for a longer-than-consensus digestion period before AI translates into broad enterprise productivity. If adoption is still early, the next 6-12 months should feature more pilot sprawl, more vendor experimentation, and more budget reallocation away from discretionary headcount replacement into consumption-based infrastructure. That is bullish for the picks-and-shovels stack, but bearish for pure-application narratives where monetization depends on visible labor displacement that may not arrive on schedule.

The risk to the current trade is that expectations are already embedded in capex-heavy winners, while the labor-saving story remains underwhelming. If usage-based pricing replaces flat subscriptions, gross margins for AI software could compress before utilization scales, creating a gap between revenue growth and cash flow. The contrarian read is that layoffs may be a signal of management discipline and cost discipline, not confidence in AI ROI, which means the market may be overpricing the pace of corporate replacement demand over the next 2-3 quarters.

Key Decisions for Investors

• Overweight NVDA on 3-6 month horizon, but hedge with a basket short in high-multiple AI application names: the market is still underpricing the persistence of infrastructure intensity even if software ROI disappoints. Prefer call spreads over outright stock to limit valuation risk.
• Long MSFT vs short UBER on a 1-2 quarter view: MSFT can monetize AI through platform lock-in and enterprise distribution, while UBER faces a longer payback period for AI tooling with less pricing power and more budget pressure.
• Use META downside protection into the next earnings cycle: the stock can still rerate lower if investors conclude layoffs are expense optics rather than durable margin expansion. Consider put spreads rather than naked shorts because any capex discipline headline can trigger reflexive squeezes.
• Pair long NVDA / short MS as a relative-value hedge within AI spend: AI infrastructure beneficiaries should outperform capital-markets sensitivity if the market rotates toward tangible picks-and-shovels exposure. Time frame: 1-3 months.
• Do not chase the broad 'AI productivity' basket here; instead wait for evidence of usage-based pricing and sustained inference cost declines over the next 6-12 months before adding software winners. The current setup favors accumulators of compute demand, not beneficiaries of labor substitution.