Lesson 3: Compute Is Revenue

If tokens are profitable units of production (Lesson 1), then the machine that mass-produces tokens is… a factory. Jensen leans all the way into that metaphor — and it reframes how to think about IT infrastructure.

The biggest buildout in history

An "AI factory" is a data center purpose-built to manufacture intelligence at scale. And the price tags are staggering:

"$100 billion into an AI factory. It must work the first time, and it must work right away."

— Jensen Huang, NVIDIA GTC Taipei 2026

The keynote calls this "the largest infrastructure buildout in human history." Chips, racks, networking, power, and cooling all have to be designed together — because, as the keynote puts it, compute is revenue.

The line that reframes everything

"Compute is revenue now. Compute is profit. The absence of revenue and profit is loss."

— Jensen Huang

Sit with that. In the old world, a data center was a cost — overhead you tried to minimize. In Jensen's world, the data center is the product line: every token it produces can be sold. So the goal flips from "spend as little as possible" to "produce as much valuable output as possible."

And since electricity is the hard limit, the key metric becomes output per unit of power:

"Performance per watt is your revenue… The more you buy, the more you make."

— Jensen Huang

The four things that decide if it pays off

Buried in the keynote is a genuinely useful framework for judging any big tech investment. Jensen says four things determine whether a multi-billion-dollar factory earns its keep:

Time to first token — how fast it comes online and starts producing. (In business terms: time to value.)
Throughput per watt — how much useful output per unit of power. (Efficiency.)
Reliability — how rarely it breaks (he calls it "mean time between interrupts"). (Uptime.)
Lifetime of the asset — how long it stays useful before it's obsolete.

"If the life of the asset is long, the TCO is low. This is the difference."

— Jensen Huang

That acronym — TCO, total cost of ownership — is straight out of an MIS textbook. The price on the invoice is only part of the story; what matters is cost over the asset's whole life, including power, downtime, and how soon it's obsolete.

The reframe for you: time-to-value, efficiency, reliability, and total cost of ownership are how professionals judge any technology investment — from a $100B AI factory down to which laptop your team buys.

From a chip company to an infrastructure company

"Our customers and our partners don't want to buy computers. They want to build AI factories… NVIDIA has really become an infrastructure company."

— Jensen Huang

This is a classic strategy move: sell the outcome (a working, profitable factory), not the component (a chip). Customers don't want GPUs; they want revenue-generating capacity. Sound familiar? It's the outcome-over-technology idea again, at industrial scale.

IT infrastructure, capacity planning, and TCO are bread-and-butter MIS — and this keynote shows them operating at civilization scale. Two ideas to carry: (1) compute is now a strategic constraint. Just as cash or talent can limit a company, access to affordable compute can now decide who wins. (2) "Cheapest sticker price" is a trap. The right lens is value and total cost over the asset's life — a point MIS has argued for decades, now worth literal billions.

Quick check

Why does Jensen say "performance per watt is your revenue"?

A factory has a fixed power budget. If every token is sellable, then squeezing more output from each watt directly increases how much you can produce and sell. Power efficiency literally becomes a revenue lever.

🧠 Think like an MIS analyst

The "cheaper chips" temptation

Your company is building AI capacity. A vendor offers chips that cost 30% less up front but deliver less output per watt and a shorter useful life. Finance loves the lower sticker price. What do you tell them?

You reframe from sticker price to total cost of ownership and revenue per watt. Jensen's exact warning: "choosing the wrong architecture just because the chips are cheaper doesn't translate." If your power budget is fixed, lower output-per-watt means fewer sellable tokens for the same electricity — you've capped your revenue to save on the invoice. Add a shorter useful life (higher TCO) and the "cheap" option can be far more expensive over three years. You'd model it out: output × price over the asset's life, minus power, downtime, and replacement. Decide on value, not the sticker.

✍️ Your turn

Think of a tech purchase you've made (phone, laptop, subscription). Walk it through the four lenses: time-to-value, efficiency, reliability, and lifetime. Did the cheapest option win?

"Compute is a strategic constraint." Can you think of a company or country where access to computing power could decide who wins? Why?

Key takeaway: When AI output is sellable, the data center becomes a factory and "compute is revenue." Judge that infrastructure (and any tech investment) by time-to-value, efficiency, reliability, and total cost of ownership — not the sticker price. And sell the outcome (capacity), not the component (a chip).

← Previous2 · Meet the AI Agent Next →4 · Platforms & Ecosystems