Michael Burry's Nvidia Warning Doesn't Go Far Enough

It is so much worse than you might think.

Photo by FlyD on Unsplash

In their latest memo to Wall Street, Nvidia name-checked Michael Burry of “Big Short” fame to contradict his AI bubble analysis and to assure the public that “they aren’t Enron.”Let’s be clear, if a multi-trillion dollar company has to address a single financial analyst by name and publicly state they aren’t Enron, that is one of the best ways to make everyone think you are the new Enron. Burry replied to this rather funny clap-back in a Substack article, pointing out that Nvidia isn’t like Enron at all but instead resembles Cisco. This is an unbelievably poignant and damning comparison. It cuts through all the bullshit and highlights the rot at the core of the AI bubble. However, I don’t believe his warning went far enough, because Nvidia has the potential to be far worse than Cisco.

Cisco & The Dot-Com Bubble

Okay, so what happened to Cisco?

Well, it was possibly the biggest loser in the dot-com bubble.

Most people focus on the unprofitable internet startups with artificial or inflated demand, like pets.com, when they talk about the dot-com bubble. But, arguably, Cisco is far more important. They were a wildly profitable and dominant provider of networking hardware. Throughout the ’90s, investors speculated that the internet could be an economic game-changer and tried to capitalise on it early, causing demand for such infrastructure to skyrocket. By 2000, between 70% and 80% of all the fibre optic cables and switchboards that made up the internet were built by Cisco. This drove profits and speculation wild, pushing Cisco’s valuation to $500 billion by March 2000, making it the most valuable company on the planet at the time.

The AI bubble mirrors this almost perfectly.

This time, AI startups like OpenAI are wildly unprofitable due to artificial and inflated demand. And this time, the infrastructure they are driving demand for isn’t Cisco’s networking but Nvidia’s AI datacentres. Like Cisco was in the ’90s, Nvidia is a mature and highly profitable company that dominates this space. Nvidia controls between 80% and 90% of the AI chip market. Hell, like Cisco once was, Nvidia is now the world’s most valuable company, with a peak valuation of $5 trillion in October 2025!

Okay, so why is that a problem?

Well, it didn’t exactly end well for Cisco.

The dot-com bubble burst. Investors realised that internet businesses like pets.com were little more than speculative ventures with functionally no customer demand. Then, interest rates went up to avoid a recession, drying up the capital these unprofitable companies needed to operate, causing a tsunami of internet bankruptcies. This proved that the hypothesised demand that fuelled this boom and speculation simply wasn’t there yet, causing investors to sell up in droves, which crashed prices.

But Cisco had supplied far too much infrastructure trying to meet this mythical demand, and much of the fibre-optic cable it had laid over the past decade remained unused for years. This unused infrastructure became known as “dark fibre”, and it was a major problem. In fact, dark fibre made up over 60% of the US’s fibre optic network in 2007.

As such, demand for new fibre optics dropped by around 70%. This cut Cisco’s revenue by over 15%, but it also destroyed any speculative value, causing its stock price to collapse by 88%! In other words, it lost more stock value than any other dot-com company. To give you an idea, Cisco lost three times more stock value during the dot-com crash than Amazon did.

Like Cisco, the infrastructure demand Nvidia is meeting isn’t real but artificial and speculative. I have written about this topic extensively, so if you want to learn more, feel free to read my back catalogue. But, if you need convincing, then don’t forget that AI improvement has ground to a halt, only 5% of OpenAI users pay for the service, and that 75% of OpenAI’s income comes from corporate companies (according to Bloomberg), despite the fact that over 95% of AI pilots using these products failed to make any noticeable gains (according to MIT). Let’s also not forget that the companies buying Nvidia’s AI chips are all wildly unprofitable AI or AI data centre companies running on nothing but debt.

Nvidia is a modern-day Cisco. The demand they are meeting is non-existent, and they are creating a huge supply-side glut that will bite them in the arse very soon. Burry is exactly right.

Or is he? Because, when you look at the details, Nvidia will face far, far worse problems than Cisco ever did!

Chip Depreciation

Burry does point out that there is a second force making this supply-side glut even worse, and that is AI chip depreciation.

As Burry points out, the entire AI industry is overstretching the lifespan of AI chips on their books and severely underestimating depreciation. Rather than having a more realistic lifecycle of up to three years, at which point more efficient and more powerful chips will have been released, making it uneconomical to keep running this ‘old’ infrastructure, they are dragging out their lifespan to six years. This means that the already wildly unprofitable business of owning and operating this AI infrastructure (read more here) is actually far, far less profitable than we all thought. When this issue comes to the fore, it could straight up pop the bubble by drying up the investment needed to buy Nvidia’s chips.

Nvidia CEO Jensen Huang has countered this argument by pointing out that their older chips are still being used for less intensive AI work, and so the same will be true for the AI chips they are shipping today. But even he doesn’t agree with himself here. He has said numerous times that their latest chips will be so efficient that even if their competitors sold their chips at zero cost, it would still make sense to buy Nvidia’s newest AI chips because it would be more economical. Huang has even stated that this means datacentres will be better off scrapping their old Nvidia chips early for the new ones, again because it will be more economical. Why? Well, Huang points out that AI is ultimately not compute-limited but instead energy-limited, as that is the harder infrastructure to build. According to Huang, this means that making AI infrastructure even a little more efficient than a competitor will yield a huge competitive edge. As such, even if the newest chips are just marginally more efficient, those that are only a few years old will need to be scrapped. .

It also means that the trillions of dollars’ worth of AI infrastructure being built will be obsolete and need to be totally replaced in just a few years.

This is worse than supply-side glut — this is a rapidly approaching stranded asset tsunami! That won’t suppress prices to a critical point; it will bankrupt Nvidia’s customers.

Cloud Compute Saturation

And this will also make a problem Nvidia faces, which Cisco never had to, far worse. Nvidia could crash its own market entirely, and Burry hasn’t really touched on this yet.

One of the main reasons Cisco only took a 15% revenue hit from the dot-com bubble bursting was that the infrastructure it had overbuilt was geo-restructured. A fibre-optic cable and networking infrastructure connecting New York and Los Angeles can only be used to send data between those two locations, and that is it. Because of this, there was still a need for fibre optic cables in the few locations that needed more bandwidth even after the bubble burst. What’s more, Cisco was well diversified, with only around 20% of its revenue coming from fibre-optic build-outs.

In comparison, Nvidia’s products are not geo-restricted. By definition, the data centres that use their chips are part of ‘cloud computing’, which means that anybody on planet Earth can use them for whatever computing requirements they have, whether it is AI or running your blog.

Furthermore, normal data centres (like most of AWS, for example) don’t have the high workload, high energy demands and insane efficiency requirements of AI data centres, which means they are currently quite profitable, despite the fact that most use comparatively old chips. In other words, they can run on both outdated and cutting-edge AI chips.

But Nvidia has put all its eggs in one basket too, as 90% of its revenue comes from selling chips to data centres, both standard and for AI, though the AI chip sales make up the vast majority of this revenue.

So, what will happen when the AI bubble bursts? Well, unlike Cisco’s dark fibre, the normal data centre market may still use Nvidia’s oversupply of AI chips. But that will, in turn, create an oversupply in the general data centre market. If we consider that AI workloads are projected to make up 70% of data centre demand by 2030, provided this bubble pops later rather than sooner, this could totally crash the general data centre market, rendering it unprofitable, and in turn, rendering demand for Nvidia’s data centre chips to near zero.

And don’t forget that by 2030, most of the AI datacentres being built today and in the next two years will be useless to the AI industry, given that it makes economic sense to ‘scrap’ them and use newer chips, as we just discussed. But, while these three-year-old data centres will be useless for AI, they will be useful for general data centre work. This means that if the AI bubble doesn’t burst soon, Nvidia will flood the general data centre market twice, once through offloading a huge amount of ‘older’ AI data centres, and then again through the AI bubble crashing, flooding the market with the newer ones.

So, it is possible that Nvidia could see close to a 90% drop in revenue when the AI bubble crashes, thanks to them completely annihilating their primary market.

Cisco saw an 88% drop in valuation when its bubble burst, causing a 15% drop in revenue. What do you think will happen to Nvidia?

Well, Nvidia is relatively financially healthy, so it can likely take this hit. However, that may change if they engage even more in the AI circular financing. But either way, Nvidia’s value is set for a monumental fall.

Summary

This is why Burry comparing Nvidia to Cisco is accurate, but the comparison doesn’t go far enough. Nvidia is poised to fall far harder than Cisco ever did. But, in doing so, it could destroy the data centre market, which will use its oversupply. Let’s not forget that Amazon’s general data centre branch, AWS, accounted for 58% of Amazon’s total profit in 2024. Likewise, general data centre operations make up a sizeable amount of Google & Microsoft’s profits. So, what happens to them when the data centre market is totally flooded? Again, Nvidia’s fall will impact others far more than Cisco ever did.

Sadly, I can’t tell when this will happen. Burry himself has admitted that the markets are acting illogically at the moment. Even those with crystal balls are left scratching their heads. But from where I stand, the question isn’t whether this will happen, but when.

---

Thanks for reading! Don’t forget to check out my YouTube channel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

---

Sources: Coffeezilla, Money Control, CNBC, Fortune, BI, Techno-Statecraft, Bruce Kushnik, BI, Cassandra Unchained, Bg2 Pod, EU News, BBC, Wall Street Zen, Nvidia, EC Mag, Bloomberg, Fortune, McKinsey, Sharetribe, Will Lockett