The AI industry is becoming increasingly concentrated and circular.

Earlier this month, Nvidia and Microsoft announced a joint $15B investment in Anthropic.

But the money comes with fine print: it will help fund Anthropic’s $30B commitment to buy compute from Microsoft — which will run on, shockingly, Nvidia systems.

This announcement is the latest in a web of deals linking various tech companies in the AI space. These circular deals have become common as companies partner with each other to push forward AI development.

But investors are starting to scrutinize these deals as market valuations get frothy from AI euphoria.

Nvidia is the foundation of the AI industry.

The company isn’t just selling critical chips for LLM training and development, it’s also financing the companies that buy them.

In a modern spin on vendor financing, Nvidia provides upfront equity capital to help fund the build-out of AI infrastructure. This infrastructure will ultimately run on Nvidia hardware.

Some of the company’s major investment announcements include:

These deals function like discounts. Nvidia provides capital upfront, and that money will ultimately be used to buy Nvidia products. In return, Nvidia locks in future demand and gets equity upside.

The $100B OpenAI deal illustrates this playbook. The investment will help fund OpenAI's data center build-out, which will require an estimated four to five million Nvidia chips.²

Training and deploying AI models requires absurd amounts of computing power.

As developer of the most widely used AI model, OpenAI is racing to secure enough compute capacity to support its exploding growth.

To this end, the company has signed several long term contracts for data center infrastructure, including:

These contracts have made OpenAI one of the largest counterparties across the industry.

When Oracle, Microsoft and other cloud providers invest in data centers, they rely on future revenues from OpenAI and other customers to repay those costs.

In a downside scenario, losses could pile up if the AI products and services powered by these data centers don’t generate enough revenue.

Bain & Company estimates that it will cost around $2 trillion to fund the necessary data center infrastructure. Even if companies can cover much of that through AI-driven cost savings, they still expect at least $800 billion will need to come from new revenue sources.⁵

AI is not the first capex megacycle.

The railroad boom in the late 19th century and the fiber optic wave during the internet bubble both triggered massive investment, followed by overcapacity and crashes. But over time, the technologies still transformed the economy.

For Nvidia and OpenAI, trillions in deals and contracts place the pair at the center of a circular AI ecosystem.

The successful scaling and mainstream adoption of AI depends on both companies continuing to grow, and turning the flood of investment into real revenue over the next several years.

But the risk is that this revenue never shows up.

👇 Check out the attached chartbook for a visual breakdown of AI’s circular ecosystem.

The information contained herein should not be construed as legal, investment, accounting or other professional services advice on any subject. ILLIQUID INSIGHTS LLC, its affiliates, officers, directors, partners and employees expressly disclaim all liability in respect to actions taken or not taken based on any or all the contents of this publication. Furthermore, opinions expressed by participants herein are not the opinions of ILLIQUID INSIGHTS LLC. Copyright © 2025 ILLIQUID INSIGHTS LLC. All rights reserved.