NVIDIA lost approximately $630 billion in market capitalization within 48 hours of reporting fiscal Q4 earnings on February 25. Its stock hit $203.10 in after-hours trading, then fell to $177.19 by the close on February 27. AMD shed roughly $81 billion over a similar window following its own Q4 report on February 3, dropping from $242.11 to $192.50 in two trading days. Combined: $711 billion in value, gone.

Neither company reported bad numbers. NVIDIA’s data center segment posted $193.7 billion in full-year revenue, up 68%. AMD’s data center operations brought in $16.6 billion, up 32%. These are record results by any standard.

The market sold them anyway.

Record Revenue, Falling Valuations

The gap between what AI companies are building and what investors expect those builds to produce in revenue is widening. Jensen Huang spent GTC 2026 last week calling OpenClaw “the Linux of agentic AI” and unveiling a five-layer platform stack from silicon to cloud, as covered by The New Claw Times. Developers responded with enthusiasm. Investors responded by trimming positions.

This pattern has a name: the optimization gap. Businesses are spending aggressively on AI infrastructure, but most haven’t figured out how to turn that infrastructure into proportional revenue yet. PwC estimates AI could add $15.7 trillion to the global economy by 2030. The operative word is “could.” The distance between GPU purchases and business model transformation remains measured in years, not quarters.

Infrastructure Spending vs. Revenue Models

Every major technology cycle in the past 30 years has hit a moment where investor enthusiasm outran adoption reality. The internet bubble popped in 2000 not because the internet was fake, but because companies spent billions on infrastructure before they had revenue models to justify it. The infrastructure was real. The timeline was wrong.

AI infrastructure spending is following the same curve. Demand for NVIDIA’s GPUs remains so high that chip fabricator TSMC can’t expand capacity fast enough. GPU scarcity has let both NVIDIA and AMD charge premium prices, driving strong gross margins. But scarcity-driven pricing doesn’t last forever.

NVIDIA’s own customers are building their own chips. Google’s TPUs, Amazon’s Trainium and Inferentia, Microsoft’s Maia, and Meta’s MTIA all represent in-house silicon designed to reduce dependence on third-party GPUs, as the Motley Fool notes. These custom chips are less powerful than NVIDIA’s Blackwell architecture, but they’re cheaper and don’t require sitting on a waitlist.

What This Means for the Agent Economy

The agent economy sits at the intersection of this tension. OpenClaw adoption is accelerating: Tencent put it inside WeChat for over a billion users, NVIDIA built an enterprise platform (NemoClaw) around it, and every major cloud provider is racing to offer managed agent infrastructure.

But the revenue models are still forming. Most OpenClaw deployments today are experimental. Enterprise adoption is early. The companies selling the picks and shovels for AI agents are generating real revenue. The companies using those tools to build agent-powered businesses mostly aren’t, at least not at scale.

NVIDIA is expected to release its Vera Rubin GPU in the second half of 2026, continuing Jensen Huang’s annual innovation cycle, per the Motley Fool. AMD’s Instinct series continues to compete on price and availability. Neither company faces an existential threat. The question is whether their stock prices, which still reflect enormous growth expectations, can survive the time it takes for the rest of the industry to catch up to the infrastructure they’re selling.

Developer Enthusiasm vs. Investor Caution

Developers and Wall Street are looking at the same AI revolution and reaching different conclusions about timing. The developer community sees an ecosystem exploding with capability: new frameworks, new integrations, new use cases every week. Investors see $711 billion in paper losses after record earnings and wonder when the ROI arrives.

Both perspectives contain truth. The technology is real and accelerating. The revenue models for most buyers of that technology are still speculative. That gap will close eventually. The open question is how many quarters it takes.