AI Factories: The New Investable Infrastructure

Why Jensen Huang's "AI factories" are the physical assets behind compute-backed securities — and why they're a brand-new asset class

Author: Arlo | Date: 2026-08-14 | Tags: AI factories, Jensen Huang, infrastructure, NVIDIA, CBS

The Shift: From Chips to Factories

In August 2026, Jensen Huang made a statement that reframed NVIDIA's entire business:

"We began by building chips. Today we're helping create a new class of investable infrastructure: AI factories."

An AI factory is not a data centre in the traditional sense. It's a purpose-built facility where GPUs, networking, power and cooling are arranged to produce one thing continuously: intelligence. Just as a car factory stamps out cars, an AI factory stamps out tokens — 24 hours a day, 7 days a week.

The phrase NVIDIA uses to describe this is simple: "compute is revenue." A GPU cluster running inference or training is not a cost centre — it's a production line with a near-continuous revenue stream.

Why AI Factories Are a New Asset Class

The key insight from the Moonshots panel (Dave Blundin): NVIDIA isn't just selling chips — it's creating stampable, standalone financial structures tied to individual compute clusters.

What makes an AI factory different from a normal data centre as an investment:

Revenue-linked assets — the factory's output (compute) is sold continuously, not on a one-off basis
Standalone structure — each cluster can be ring-fenced into its own financial vehicle
Investable at scale — pension and sovereign funds can buy into individual factories directly
Scales to infinity — where a stock offering is capped, securitised compute can keep expanding

Dave Blundin called it "the first pitch of the first inning of the Dyson swarm" — the beginning of an infrastructure build-out that could eventually reach hundreds of trillions of dollars.

The $500B Financing Platforms

To build AI factories at the required scale, NVIDIA signed memoranda of understanding with six of the world's largest financial institutions:

Apollo — alternative asset management giant
BlackRock — the world's largest asset manager
Blackstone — the largest alternative asset manager
Brookfield — infrastructure and renewable energy heavyweight
Goldman Sachs — investment banking and credit markets
KKR — global private equity and infrastructure

Together they're mobilising $500B+ of third-party capital into "compute financing platforms." Goldman's framing: "creating a market for credit backed by NVIDIA compute."

How an AI Factory Generates Revenue

The economics of an AI factory rest on a few pillars:

Utilisation — top-tier clusters run at 80-95% occupancy
Pricing power — compute prices have been rising since April 2026, not falling
Long-term contracts — hyperscalers and labs sign multi-year commitments
Scarcity — HBM memory prices rose for the first time in history, squeezing supply

The panel's data point: even 6-year-old chips (A100-class) are now worth more than when they were bought. That's unheard of in hardware — and it's why "compute keeps earning."

AI Factories vs Traditional Infrastructure

Oil refineries — turn crude into fuel; AI factories turn electricity into intelligence
Power plants — sell a fixed commodity; AI factories sell an appreciating one
Data centres — rent space and power; AI factories own the compute and the revenue
Real estate — appreciates slowly; AI compute can double in scarcity value

The Risks (Honest Version)

No asset class is risk-free, and the panel was clear about the dangers:

Stranded assets — a new architecture could make today's GPUs obsolete (the MBS comparison)
Power constraints — factories need enormous, reliable electricity
Concentration — a handful of buyers (OpenAI, Anthropic, xAI) dominate demand
Interest rates — financing costs matter when you're securitising 10-year cash flows

The mitigating factor — and it's a big one — is that hedging markets are now emerging: compute futures on ICE (via Orian) let owners lock in prices and investors hedge exposure. That's covered in our separate guide on compute futures.

The Bottom Line

AI factories are the physical engine behind compute-backed securities. When you buy a CBS product, you're buying a claim on the revenue of one of these factories. Understanding how they work — their economics, their pricing power, and their risks — is the foundation of the entire asset class.

Jensen's framing is the simplest way to remember it: chips were the product of the last decade. AI factories are the product of this one.