Compare

The template only earns its keep if you actually put two cases next to each other.

This is the reason every teardown answers the same questions in the same order. Put a luxury house beside a console platform and the differences stop being vibes — one is refusing sales to protect a price, the other is selling below cost to collect a toll, and both are defending a moat.

Case ACase B

Technology · NASDAQ: NVDA

Nvidia

Designs the chips that train and run nearly every large artificial-intelligence model, has them manufactured by TSMC, and gives away the software that makes them impossible to replace.

Technology · NASDAQ: AMZN

Amazon

Runs a near-breakeven retail operation at enormous scale, and earns essentially all of its profit from renting out the two things that operation forced it to build — computing capacity and shelf placement.

The thesis in one line

Nvidia does not sell chips. It sells the only door into a room that eighteen years of programmers have already filled.

The store is not the business. The store is the customer acquisition cost for the businesses.

Verdict

Watch it

Own it

Moat

Wide

Switching costs · Network effects · Scale economics · Process power

Wide

Scale economics · Network effects · Process power · Switching costs

Porter's five forces

Competitive rivalry

Threat of new entrants

Threat of substitutes

Buyer power

Supplier power

Competitive rivalry

Threat of new entrants

Threat of substitutes

Buyer power

Supplier power

Headline figures

Revenue: $130.5B
Gross margin: 75.0%
Data centre share of revenue: ~88%
Founding capital: $40,000

Net sales: $638B
AWS share of operating income: ~58%
Retail operating margin: ~5.4%
Third-party share of units: ~60%

Unit economics

One H100 accelerator (illustrative — Nvidia does not disclose per-unit cost, and these are outside teardown estimates)

The silicon is a few thousand dollars and the software that makes it worth thirty thousand is free. The margin is where the software is paid for — which is why company-wide gross margin sits at 75% and why every competitor with a cheaper chip still loses.

One $50 third-party item sold through Amazon

Amazon captures roughly 38% of the sale price and carries none of the inventory risk. The seller took the risk, paid for the warehouse, and then paid again to be found in a search of Amazon's own catalogue.

What would change her mind

A down year that is not a disaster: data centre revenue falling 20–30% while gross margin holds above 65% and the customers' own chips take less share than feared. That would show the moat works when demand stops doing the work, and I would buy the recovery. The other direction: if the hyperscalers' capital spending plans are cut while Nvidia's customer-financing commitments grow, the revenue was partly Nvidia's own money and the multiple has no floor.

If AWS revenue growth falls below 15% for two consecutive quarters while segment operating margin also declines, the cloud business is being commoditised rather than merely contested, and the capex is being spent defending share rather than buying growth. That combination — decelerating growth and compressing margin at the same time — breaks the thesis outright. A single weak quarter on either measure alone does not.

1 mechanism in common

platform flywheel