This Should Pop The AI Bubble

Oversupply indicators and insider warnings.

Photo by Lanju Fotografie on Unsplash

When will this damn AI bubble burst? How many clear-as-day signals does the market need shoved into its face for this silly charade to fall apart? Without a functional crystal ball, we simply won’t know until it is too late. But the frustration of watching the market aggressively mash the accelerator rather than the brake in response to this impending car crash is palpable. Those paying attention, who know just how insanely over-inflated this industry is, are nervously twitching at even the slightest sign that the bubble will pop. But, over the past week, we have seen some major movements that absolutely should have popped this bubble.

How does a bubble burst?

Well, a bubble is driven by speculative demand, which fuels unfounded investment that pushes an asset’s price beyond its true value and spurs a huge increase in supply as investors rush to meet this demand. A bubble bursts when this speculative demand fails to materialise, and investors realise there is a supply-side glut that undermines the value of their investments. So, they flee and sell their positions to reduce their losses, only for this mass sell-off to crash the price of these overvalued assets.

From tulip mania to the dot-com bubble to even the US housing bubble, this dynamic has played out time and time again.

So, if you don’t have a crystal ball, the next best way to try to predict when a bubble will burst is to notice when the lack of speculative demand starts to cause serious issues, a pivot, or even a walkback from those most exposed to the bubble. That is a giant red flag that the pop is rapidly incoming.

Again, I cannot predict when this bubble will pop more than I can predict when Trump will actually follow through on a peace deal with Iran. This world is a chaotic mess after all.

But we have seen three of these giant red flags over the past few weeks.

As I wrote about a few weeks ago, Elon Musk’s xAI has rented out almost all of its AI compute capacity at its Colossus data centre to its competitors, Anthropic and Google. Anthropic agreed to rent 222,000 GPUs for $1.25 billion a month, and Google has agreed to rent 110,000 GPUs for $920 million a month. Bear in mind the highest verifiable figure of operational GPUs at xAI’s only data centre is 230,000 as of September 2025. So, these deals could represent xAI renting out the vast majority of its own compute.

These deals make no sense. Musk believes that AI needs way more compute capacity, which is why he plans to launch a million orbital data centres. So why sell pretty much all of xAI’s compute capacity to xAI’s competitors?

The only way these deals make sense is if the demand for xAI’s Grok and the utilisation of this data centre are so painfully low that it is not just more profitable but more advantageous in the long run to sell this compute directly to xAI’s main competitors.

Now, yes, Grok is objectively and measurably less capable than Claude, and it carries far more legal baggage, too. So, is this lack of demand just down to a lack of competitiveness? Well, I’d argue that if AI is going to be the general-use tool it promises to be, then Grok is nearly as capable as every other LLM AI out there. On top of that, xAI operates using much smaller infrastructure than Anthropic or Google, which both have multiple GW of compute capacity to hand. xAI was set up to only ever be a niche player, and it still couldn’t generate enough demand to justify its comparatively microscopic operation.

Ultimately, this is a major sign that the speculated demand for AI simply isn’t there and that those who bet big on it are desperately trying to create an escape route, even if it gives competitors an advantage.

And it isn’t just xAI doing this. So is the Zuck!

Meta, formerly known as Facebook, plans to sell its ‘excess’ AI compute after failing to stir up enough demand for its models. This is even more catastrophic than xAI’s pivot to renting out compute because Meta has really bet the farm on their AI being wildly popular. Meta has its own immense AI data centres and has also signed countless deals to buy and rent AI compute from a variety of companies. They signed a $100 billion agreement with AMD to build 6GW of AI compute in February, signed a $21 billion agreement with Coreweave to rent compute in April, have a $27 billion contract to rent AI compute from Nebius over the next five years, and raised $30 billion in corporate bond debt last October, explicitly to expand its own data centres.

In other words, Meta has done a Metaverse — bet big on a stupid idea that no one wants. Quite frankly, Zuck becoming bearish on a business venture should be seen as a bubble-pop indicator; the man is brilliant at betting large and late and stumbling right as the proverbial rug is pulled.

It is kind of obvious that Zuckerberg severely overestimated demand for his AI. But is that due to low industry demand or just this product being bad? After all, Meta’s latest model is far behind even Grok.

Well, I don’t think it matters either way. Lenders and Meta shareholders were happy to support Meta’s AI expenditure, believing it would generate demand for internal models. This pivot still shows a catastrophic lack of demand. Moreover, renting the vast amount of compute on the open market will surely flood it, crashing the price and choking AI circular funding (with more on that in a second). But also, rent this to whom? If xAI and Meta need to rent out, the only people buying are Anthropic, Google and OpenAI. They already have their own infrastructure, and who is to say they don’t have a fatal lack of demand? After all, companies are making a huge U-turn on AI adoption right now as they realise it doesn’t do what it says on the tin. Also, if there are only a handful of buyers of AI compute, doesn’t that prove that demand hasn’t materialised? No matter how you cut it, Meta giving up and selling their AI infrastructure is a sign that industry-wide demand simply doesn’t exist.

As a side note, this heavily suggests that current AI models are not compute-constrained, despite what the industry has parroted for a long time. If xAI and Meta can turn around and rent out huge portions of their AI infrastructure without massively impacting their current AI operations, it shows just how horrifically overbuilt and overinvested their AI infrastructure is, which, again, suggests a colossal supply-side glut.

There is another reason why Anthropic and OpenAI potentially renting this ‘excess’ compute could pop the bubble. The AI bubble is actually a bubble within a bubble. You see, inside the overall AI bubble is a computer hardware and compute leasing bubble, propped up by circular financing. The main AI labs, namely Anthropic and OpenAI, are not just wildly unprofitable, but they also receive much of their funding from chip makers and AI compute providers. For example, when Nvidia, Microsoft or Amazon invests in OpenAI, it isn’t in cash but in GPU lease agreements. This enables these providers to effectively buy their own products via an AI company, which enables them to register it as revenue, boosting their revenue. They also get a share of the AI lab in return, and by ‘buying’ ever smaller portions of the AI lab each time they do this, they can raise the price of the shares, making their previous investments more valuable, once again raising their revenue. Sasha Yanshin has a great video on this topic if you would like to know more.

So, Anthropic and OpenAI can’t really afford to rent all this GPU capacity, and that will get in the way of this circular financing. And they desperately need this circular financing to stave off bankruptcy. Not to mention that if either one of them goes bust, that will pop the bubble too.

Because OpenAI and Anthropic are still private, we can’t tell if they are experiencing the same demand problems as xAI and Meta. But there is one thing that suggests they are and that investors are already trying to exit and mitigate losses: Oracle.

OpenAI has agreed to buy $300 billion of AI capacity from Oracle over the next five years. Oracle itself has taken on a truly horrendous amount of debt to build out this infrastructure for OpenAI. In other words, Oracle has bet everything on the speculative demand for AI existing, because if it doesn’t, they could be drowned by the debt they have taken on for these deals. So, you’d think if investors were confident this AI demand was coming, that Oracle’s shares would be doing well. Well, they aren’t. Oracle’s share price has dropped by more than 40% in the last month.

If you ask me, those are two massive red flags: that the speculative demand for AI simply isn’t there and that investors are potentially starting to recognise this fact and pull out.

Given how insanely overleveraged and overvalued this industry is, these should be the pins that pop the bubble. But that assumes the market and investors are behaving rationally, and quite simply, they aren’t. Instead, it feels like all these billionaires are playing a giant game of chicken, seeing how far they can push this scheme, with the first to exit taking all the gain and everyone else taking the loss.

---

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. 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!