One analyst is calling for Nvidia to nearly double revenue in 2027 while the rest of Wall Street wonders if AI buildout has already peaked.
The Summary
- Gene Munster projects Nvidia revenue will grow 90% in calendar year 2027, calling the company's trajectory "breathtaking"
- This forecast implies continued enterprise acceleration in AI infrastructure spending, not a slowdown
- If accurate, it signals the agent economy is still in infrastructure phase, not deployment phase
The Signal
Munster's 90% revenue growth call for 2027 is either prescient or delusional, and which one matters for anyone betting on the agent economy. Nvidia just posted another quarter of absurd numbers, and the street's been waiting for gravity to reassert itself. Munster thinks gravity doesn't apply here.
The math is straightforward. If Nvidia does $120 billion in 2026 revenue (conservative estimate given current run rate), 90% growth puts 2027 at $228 billion. That's not a company. That's a rounding error away from the GDP of Portugal. For context, Nvidia did about $61 billion in fiscal 2024. Munster is calling for a near-quadrupling in three years.
"What's been happening with Nvidia is breathtaking."
Two ways this pencils out:
- The foundation model arms race intensifies as companies rush to build proprietary intelligence
- Enterprise AI deployment finally moves from pilot programs to production infrastructure at scale
- A new category of AI hardware demand emerges that we're not pricing in yet
The bear case writes itself. AI spending is already massive. Hyperscalers are building their own chips. Model efficiency improvements mean you need less compute for the same output. Every venture capitalist with a podcast thinks we're in an infrastructure bubble that makes the 2000 fiber optic buildout look rational.
But Munster's not betting on more of the same. He's betting that 2027 is when agents stop being demos and start being deployed. When every enterprise SaaS company isn't just adding a chatbot, but rebuilding their product as an agent platform. When the shift from "AI features" to "AI-native architecture" becomes unavoidable for anyone who wants to stay competitive.
The implication splits across market structure and capability. On structure: if Nvidia can actually pull 90% growth in 2027, it means no one else has caught up on AI chips. Not AMD, not the hyperscaler custom silicon, not the startups. Nvidia's moat isn't just chips, it's CUDA and the entire software stack developers actually know how to use. On capability: it means the models and agents being built in 2026 are demanding enough to require that much more compute.
The Implication
Watch Nvidia's data center revenue mix in the next two quarters. If the growth is still coming from foundation model training, Munster might be early. If it's shifting to inference and enterprise deployment, he might be right on time. The difference matters because training buildout is lumpy and can pause. Inference buildout is what happens when AI goes from lab to production, and that's the durable revenue stream that justifies a $228 billion run rate.
For builders: plan your compute budget accordingly. If Munster's right, GPU availability doesn't get easier in 2027. It gets harder. The companies locking in capacity now will ship. The ones waiting for prices to come down will wait.