| Company | Nvidia Corporation — dominant supplier of GPUs for AI training and inference |
| The headline | Extraordinary reported revenue and continued strong demand for AI infrastructure |
| The emerging trend | OpenAI's new "Jalapeño" chip joins a widening field of custom silicon — Google's TPUs, Amazon's Trainium, Microsoft's Maia, and Meta's in-house chips |
| What's not yet visible | Which workloads migrate to purpose-built chips, and how much of Nvidia's share of AI-compute economics that eventually moves |
| Theme | Whether Nvidia keeps its share of the economics created by AI growth — not simply whether AI demand itself continues |
Nvidia's results are read almost entirely through one lens — is AI demand still growing? That's the easy question. The harder one, and the one the income statement doesn't answer yet, is who ends up capturing the economics of that growth five years out.
The Hook
Nvidia just posted extraordinary numbers. Revenue and AI infrastructure demand remain exceptionally strong.
If you only read the headline numbers, the story looks simple: Nvidia is winning.
What the Numbers Told Us — and What They Couldn't
What the numbers told us
- Growth is real. Demand for AI infrastructure isn't slowing down
- By every reported metric, Nvidia's business is expanding
What the numbers couldn't tell us
- Who captures the economics of that growth five years from now
- OpenAI's new "Jalapeño" chip is the latest data point in a trend that's been building quietly
- Google has TPUs, Amazon has Trainium, Microsoft has Maia, and Meta is developing its own silicon
- None of these need to replace Nvidia everywhere — only to win the workloads where purpose-built chips beat general-purpose GPUs on economics
The income statement doesn't have a line for that. It may show up years later — in margins.
The numbers weren't wrong. They were incomplete for the decision an investor had to make.
The Narrative
The easy Nvidia bear case — "AI demand disappears" — isn't the one worth taking seriously.
Demand isn't the fragile part of this story. Pricing power is.
Nvidia's valuation isn't simply pricing in AI growth. It's pricing in Nvidia capturing an outsized share of the economics created by that growth. Those are two different bets. And the market can easily mistake them for one.
Imagine AI compute demand grows 5×, but Nvidia captures a progressively smaller share of the economics created by that growth. Nvidia could still report spectacular revenue growth. That's the dangerous part. Growth is very good at hiding competitive erosion — for a while.
The Lesson
A company's reported growth rate can remain strong long after its competitive position has begun to weaken.
Revenue is a lagging indicator of pricing power, not necessarily a leading one.
"If you were underwriting Nvidia today, what would you watch more closely — revenue growth, or the customer-by-customer shift toward custom silicon that could eventually change who captures the economics of AI?"
Not every threat to a great business shows up as declining revenue.
Some show up as a shrinking share of a much bigger pie.
— Mahesh Ramanujam, FCA, DISA(ICAI) · R. Mahesh & Associates, Chennai
- Nvidia Corporation — quarterly earnings release and management commentary, August 2026
- Public reporting on OpenAI's custom AI accelerator ("Jalapeño") development
- Google Cloud — Tensor Processing Unit (TPU) program, public disclosures
- Amazon Web Services — Trainium custom silicon program, public disclosures
- Microsoft — Maia AI accelerator program, public disclosures
- Meta Platforms — public statements on in-house AI silicon development