Nvidia posts $68B quarter, up 73% YoY, but stock sells off 5% at open
Feb 26, 2026
Key Points
- Nvidia posts $68.1 billion in quarterly revenue, up 73% year-over-year, but stock sells off 5% at open despite beating consensus and raising guidance.
- Power grid capacity, not chip supply, emerges as the structural ceiling for AI demand growth as data centers approach infrastructure limits.
- Jensen Huang argues AI agents will adopt existing enterprise software rather than replace it, countering the SaaS apocalypse narrative.
Summary
Nvidia posted $68.1 billion in quarterly revenue, up 73% year-over-year and 20% quarter-on-quarter, beating consensus by nearly 3%. The stock popped 3% after earnings but sold off 5% at market open, marking its worst day since April 2024 when DeepSeek sparked fears that cheaper inference models would kill demand for Nvidia chips.
The selloff contradicts the fundamentals. Nvidia's gross margin, revenue guidance, and supply position all exceeded analyst expectations. The company disclosed $21.4 billion in inventory (up from $19.8 billion) and $95.2 billion in locked supply commitments with chip manufacturers, signaling confidence it can meet demand for several quarters ahead.
The real concern appears structural rather than cyclical. Semi-Analysis flagged a potential hard ceiling that may emerge not in silicon but in power infrastructure. Data centers currently consume under 1% of U.S. electricity. If AI deployment accelerates faster than power plants can come online, a timing mismatch between chip availability and grid capacity could flatten demand growth. Investors may be pricing this risk in.
Jensen Huang pushed back on the SaaS apocalypse narrative during an earnings call with CNBC, arguing that AI agents will use existing software tools rather than rebuild them from scratch. Training a model to recreate a calculator when a SaaS calculator exists wastes tokens and money. This dynamic has already played out in databases, where certain platforms have seen demand spike because agents pull them off the shelf. Software firms will adopt agentic AI to boost their own efficiency, not be displaced by it.
Chip supply is locked. Energy is the open question. Margin compression from cheaper models proved real but not fatal. Demand for frontier-grade models remains exceptional, with very little demand for third-tier options across any model tier. Customers want only the most performant option for each specific task.