Is AI Really in a Bubble? What the Numbers Actually Show

Question

The bearish narrative around AI is everywhere—phantom data centers, inflated growth projections, and whispers that the whole thing is overblown hype. But let’s cut through the noise with actual math.

The $5 Trillion Question

JPMorgan ran the numbers: if global AI infrastructure hits $5 trillion by 2030, that stack would need to generate ~$650 billion in annual revenue just to deliver a 10% return. That’s 150%+ of Apple’s yearly sales and 30x OpenAI’s current revenue.

Sound impossible? Maybe. Or maybe it’s just how transformative revolutions actually work.

Why Traditional Analysis Gets This Wrong

Here’s the key insight: AI isn’t another additive tech like mobile or cloud.

Jensen Huang (NVIDIA) nailed it recently—older software was pre-compiled and static. AI is different. It needs to generate intelligence in real time, contextually aware, constantly producing new tokens. That’s computationally expensive. Massively expensive. Which means:

AI needs factories, not just software
Those factories need hundreds of billions in capex
Goldman Sachs and BofA project AI infrastructure spending crosses $1 trillion by 2028
The compute power required keeps accelerating, not decreasing

Wall Street analysts have consistently underestimated NVIDIA’s trajectory. They’re projecting $275B in FY’27 sales—but the actual math suggests the company will be vastly larger. Even early bulls were too conservative.

Two Phases Coming

Phase 1 (Now): Generative AI and Agentic AI adding hundreds of basis points to GDP.

Phase 2 (Next): Physical AI—autonomous vehicles, humanoid robots, smart cities, automated factories. The compute requirements? Basically unquantifiable right now. These billions of edge devices need redundant, robust compute infrastructure that dwarfs what we’re building today.

The Real Money Moves

Long-term investors like Baillie Gifford (who spotted Tesla early) now have NVIDIA as their largest position. They think in decades, not quarters.

Sure, some sectors cooled—Meta, Oracle, CoreWeave saw corrections. But that’s not a bubble popping. That’s a reset. The fundamental demand for NVIDIA GPUs remains insatiable.

Bottom Line

Right now? The AI revolution is still underhyped. We’re nowhere near the euphoria stage where dumb money floods in. Yes, stay vigilant. Yes, filter out the garbage analysis. But the infrastructure spending wave is just getting started, and it won’t stop when the hype cycle does—because AI actually works and companies are racing to implement it.

Watch for NVIDIA’s next earnings beat. Wall Street will need to raise estimates again. Because they always underestimate.