Is Nvidia Buying Its Own Revenue?

Circular AI

A fun trick from the dot-com era was circular financing, where a company would invest in a startup, and that startup would then use the money to buy the first company's products, creating the magical illusion of revenue growth. It is a little weird to see this ghost reappear in the AI boom, but at a much, much larger scale.

This week, Nvidia, the chipmaker at the heart of the AI revolution, announced it will invest as much as $100 billion in OpenAI, its single most important customer. The official purpose is to help the ChatGPT maker fund its massive data center expansion. The deal, however, has immediately revived concerns that the AI gold rush is being propped up by financial engineering. Here is Bloomberg:

"The action will clearly fuel 'circular' concerns," Stacy Rasgon, an analyst with Bernstein Research, wrote in an investor note after the deal was announced.

Those concerns have followed Nvidia, to varying degrees, for much of the AI boom. The chipmaker participated in more than 50 different venture investment deals for AI companies in 2024, and is on pace to top that number this year, according to data from PitchBook. Some of these companies, which include AI model makers and cloud providers, then use that capital to buy Nvidia’s expensive graphics processing units.

The deal will "likely fuel these worries much hotter than what we have seen previously, and (perhaps justifiably) raise concerns over the rationale behind the action," Rasgon wrote.

The question of whether this is technically circular financing is an interesting one. Nvidia insists the funds are not for "direct purchases." But the capital is explicitly earmarked to help OpenAI build data centers equipped with Nvidia's chips. Even if the funds aren't used for a direct purchase, the investment provides OpenAI with the financial capacity it needs to remain Nvidia's best customer. The effect, critics argue, is the same: Nvidia is ensuring demand for its own products by bankrolling the buyer.

The strategic logic, of course, is sound. OpenAI is a money-losing business burning through billions of dollars a year. Its ability to afford the next generation of even more expensive Nvidia chips is not guaranteed. By investing, Nvidia is effectively de-risking its own customer base. As one analyst aptly put it, "It's kind of like having your parents co-sign on your first mortgage."

The risk, however, is that this is classic bubble behavior that masks the true, organic demand for AI infrastructure. The deal makes it incredibly difficult for investors to discern how much of the AI boom is driven by genuine, sustainable demand and how much is being propped up by a circular flow of capital between the industry's two most important players. The arrangement may be a visionary move to secure the future of AI, but it carries the distinct echo of past bubbles, where the line between investment and self-dealing became dangerously blurred.

Catching Up

If you are a giant, century-old company and a flashy competitor is eating your lunch with some new technology, there is a standard playbook. You hold a press conference, you show off a prototype that does the new thing, you use all the right buzzwords, and you make sure to mention the competitor by name so everyone knows you are in the same league. It is a performance, a signal to investors and customers that you are not, in fact, about to become obsolete.

And so this week we have Nissan, a pioneer of the electric car that somehow got left behind, putting on a very good show in Tokyo. The company demonstrated its new AI-powered self-driving system, and its executives were not shy about who they were aiming for. "Our rival is Tesla," said the head of Nissan's autonomous driving division.

The demonstration itself was impressive. Here's Nikkei Asia on the experience of riding through the chaotic Ginza district:

Starting, stopping and proceeding through crosswalks were all smooth and effortless, like riding with a veteran driver. There was so little discomfort that this reporter even began to feel sleepy.

This is, you have to admit, a compelling vision for the future of commuting.¹ The car, a prototype Ariya SUV, uses a combination of cameras, LiDAR, and a trendy "end-to-end" AI to navigate. This is a big deal because it means Nissan is moving away from old, rigid, rule-based systems and embracing the learning, data-driven approach that powers Tesla's Autopilot.

But the interesting part is not that Nissan is copying Tesla, but how it is choosing not to copy Tesla. While it is adopting Tesla's AI-first software philosophy, it is pointedly rejecting its hardware dogma. Musk has famously dismissed LiDAR—the laser sensors used by almost everyone else, including Waymo—as a "fool's errand." Nissan's prototype, however, is full of the stuff.

This is the follower's dilemma in a nutshell. You know you have to embrace the new paradigm (AI), but do you go all-in on the leader's specific, radical, and possibly flawed strategy (vision-only)? Or do you hedge? Nissan is hedging. It wants the learning flexibility of a neural network, but with the comfort and redundancy of lasers. It's a conservative, very legacy-automaker thing to do.

The problem, of course, is the timeline. Nissan says this "world-class" system will be ready for mass-produced cars in "fiscal 2027." By then, Tesla will have had three more years to collect data from its millions of cars on the road, further honing its own AI. Three years is a very long time in the AI business. Still, it was a good show.

¹ Though one hopes the final product is not so smooth that the person in the driver's seat, who is still legally responsible for the car, also begins to feel sleepy. That seems less than ideal.

The Scoreboard

• Semiconductor: Nvidia Has a Problem: Too Much Money (WSJ)
• AI: Alibaba launches Qwen3-Max AI model with more than 1 trillion parameters (Reuters)
• Software: Dedicated mobile apps for vibe coding have so far failed to gain traction (TechCrunch)