Is this the end of the AI cycle?

[cobito]

I've been closely following the financial situation of the AI industry as best I can for nearly a year now. I don't own any stocks or have any economic interest beyond wanting to know a few months in advance when the hardware market might start to normalize. I also lack knowledge in economics or finance. This thread is not about financial advice. It's purely my personal observations from the perspective of an outsider on all these matters.

I've been wanting to buy some components to upgrade my machines for nearly a year now. In fact, I built my last machine exactly a year ago, and that's when the crisis began. This post is exclusively about that.

On another note, my general opinion on AI is that it's a revolutionary technology changing the paradigm of many economic sectors and will be as disruptive, if not more so, than the popularization of the personal computer or the expansion of the Internet. However, I don't think the way it's being commercially exploited is right, hoarding all kinds of resources by any means necessary just to meet revenue and profitability targets. It's unsustainable and doesn't seem ethical to me. This is just my personal opinion.

Now, getting down to business: essentially, I've identified three types of companies involved in this industry. From lowest level to highest: hardware manufacturers (e.g., Nvidia), hyperscalers (e.g., Microsoft), and service providers (e.g., OpenAI).

They all feed into each other: some hardware manufacturers provide hardware, hyperscalers provide computing power, and service providers provide their latest models. Everything is announced as billion-dollar exchanges, but it's money that just circulates within the same ecosystem. And this money comes from two sources: revenue from traditional services (e.g., Microsoft cloud services) and through loans. It's an ever-growing pool of money. And the thing is, this money comes from equity stakes, commitments, and loans.

In this circle, the only absolute winners are the hardware manufacturers who have decided to stay out of the circular economy, as they provide the non-negotiable raw materials. And not only that, but hardware has become a strategic resource and even a financial asset (at least, that's what Nvidia is trying to make it). They've already won no matter what happens. All they have to do is weather the storm that's coming their way.

Over the past two or three years, AI has dominated tech news to such an extent that during every quarterly earnings report by major hyperscalers (Google, Meta, Microsoft, Amazon, etc.), whenever a significant increase in AI infrastructure spending was announced, those companies' stocks would see historic surges. Investors blindly believed in the future profitability of AI and accepted any amount of spending without any guarantee of return.

The thing is, that future has already arrived. We're in the future of two or three years ago, and the tide has turned.

The AI pullbacks began this summer, but back then there was a very good excuse: valuations were so sensitive that even minor bad news would trigger major corrections. So it was dismissed as temporary and insignificant.

But over the past two weeks, a few events have occurred that are starting to suggest that (as Jeremy Irons' character said in Margin Call), the music is beginning to stop. For me, these developments have prompted me to write this post, even at the risk of coming across as a pessimist in a few days or weeks:

Financial Sector

• Banks are now trying to offload debt related to AI data centers [1]
• CoreWeave's credit default swaps have reached crisis levels as investors lose confidence in AI-related debt [2]

In other words, on one hand, major financial institutions like JPMorgan and Morgan Stanley are saying that the AI debt they underwrote has exceeded the risk threshold they consider acceptable, and they have decided to place that debt in some form (watch out for the "opportunities" our banks might offer soon). And on the other hand, there is corporate debt tied to AI data centers that already, as of today, carries a high risk of default.

Service Providers

Regarding AI growth and profitability, the two major service providers have announced concerning figures:

• OpenAI's annualized revenue (not profits) has been $50 billion, instead of the expected $70 billion [3]
• Anthropic's revenue growth is slowing down at a faster rate than expected [4]
• There are increasingly more voices saying that Anthropic has overpriced itself heading into its IPO.
• Character.ai has gone bankrupt. It's an AI service that predates OpenAI [5]

In other words: growth expectations (not profitability, which hasn't even been reached yet) are proving highly disappointing.

Paying Customers

Paid AI services have been heavily subsidized by the providers themselves from the start. No one yet knows the true cost of the services being offered. But in software development, AI is already a relatively mature technology, and this summer Microsoft updated its GitHub Copilot pricing, multiplying the cost for heavy users. It's unclear whether this price hike still covers costs, but after the change, several companies have started heavily limiting AI usage. The most pronounced case is Accenture, a company that has been touting the benefits of AI from the beginning:

• Accenture limits AI usage to $29/month per employee. [6]

This translates to the fact that when the true cost of AI starts being paid, major customers are reluctant to cover such expenses. Revenue holds steady, but consumption drops.

Hardware Manufacturers

Well, here comes what personally interests me. What's happening with hardware manufacturers? And not just any hardware, but the kind I need most (RAM and hard drives). Before continuing, a couple of relevant news items:

• Toshiba is going to double its mechanical hard drive storage production by March 2027 (the increase has already begun) [7]
• CXMT has stepped on the gas with next-generation memory production [8]

So, in summary, we have a context where:

• The AI industry is finding it increasingly difficult to take on debt.
• AI revenue growth is proving much weaker than expected.
• There is no solid evidence that any provider has reached profitability.
• A relevant player has failed to survive the high costs.
• Major paying customers who could prop up the whole operation are starting to heavily limit AI usage.
• The market has pushed hardware manufacturers to increase production one way or another.

And then I look at the stock valuation trends of the major publicly traded hardware manufacturers: Nvidia, Intel, AMD, Micron, SK Hynix, SanDisk, Seagate, and Western Digital. What I see is that their current valuations are the same as they were 6 months ago. This might not seem very revealing. But what is revealing is the pace and continuity of their decline. Unlike last summer when dips were excused as profit-taking, there are no buying opportunities being seen now. Day after day, investors are saying the price is too high. Predominantly, bargains aren't appearing; instead, overvalued stocks are, and some companies are already worth less than half of what they were a few months ago.

And when will we know if we're facing a temporary hiccup (and this whole post is nonsense) or if a cycle shift has already begun? Because all I've done here is speculate. Well, confirmation could come on October 21 (this Wednesday), 28, and 29. On those days, Meta, Alphabet, Microsoft, and Amazon will release their quarterly earnings along with their AI infrastructure spending forecasts. If they say growth will continue to accelerate: we're looking at unaffordable hardware for a while. If they say growth will moderate: we might see slight price drops. And if they say they'll cut spending: the crisis is set, and prices could suffer historic declines.

If we are truly facing a cycle shift, several months will still pass before we see moderate hardware price drops, because there is pent-up demand and signed contracts that guarantee demand to some extent from major data centers.

No one knows what's going to happen. What is certain is that there is a lot of fear, nervousness, and distrust.

[1] https://startupfortune.com/banks-are-now-trying-to-offload-the-ai-data-center-debt-they-just-underwrote/
[2] https://www.cnbc.com/2026/10/15/coreweave-credit-default-swaps-hit-distress-levels-as-investors-sour-on-ai-debt.html
[3] https://es-us.finanzas.yahoo.com/noticias/ingresos-anualizados-openai-50-000-173506788.html
[4] https://www.reuters.com/business/finance/investors-temper-anthropic-ipo-valuation-expectations-over-revenue-growth-2026-10-15/
[5] https://www.nbcnews.com/tech/tech-news/characterai-files-bankruptcy-sell-assets-disney-20-million-rcna603174
[6] https://thenextweb.com/news/accenture-claude-29-a-month-cap
[7] https://www.trendforce.com/news/2026/10/02/news-toshiba-to-double-ai-data-center-hdd-capacity-by-fy2027-with-380m-expansion-as-ai-storage-demand-surges/
[8] https://www.bloomberglinea.com/tecnologia/la-empresa-china-cxmt-afirma-haber-superado-a-samsung-y-sk-hynix-en-el-desarrollo-de-chips-de-memoria/

[pos_yo]

@cobito The prices of memory and storage make absolutely no sense. We're talking about components that don't use the absolute cutting-edge manufacturing methods, so production costs haven't risen anywhere near as much as the final price has. It's not normal that right now a Crucial DDR5 (32G 6000mt/s CL30) kit is going for 500€ on Amazon while a Ryzen 7 9800X3D is cheaper, at 426€. I've never seen a memory kit that should be the standard priced higher than a high-end CPU. It makes no sense...

[Yorus]

I've been following this topic for a while on some YouTube channels about gaming hardware, like El Rincón de Varo, and it's often discussed that we still have 3 or 4 years of this situation, or even with price hikes. All of this will depend on what you're saying, because it's obvious there's a bubble and everyone wants to take advantage of it. Over time, the market will stabilize: some companies will fail while others will solidify their position... There simply isn't room for everyone, especially when they try to recoup their investments and customers aren't willing to put up with it.

[cobito]

@Yorus said:

I've been following this topic for a while on some YouTube channels about gaming hardware, like Rincón de Varo, where they say we still have 3 or 4 years of this, or even with price increases. All of this will depend on what you mention, because it's obvious there's a bubble and everyone wants to cash in, but over time the market will stabilize and some companies will fall while others will solidify their position... there isn't room for everyone, especially when they try to recoup their investment and customers aren't willing to play along.

I've also read the 3 or 4 years thing around here. And actually, it's a fairly credible forecast assuming there are no shocks. In the end, it's just a matter of waiting for AI data center growth to start stabilizing and for hardware manufacturers to adapt their production to the new demand.

But there are certain aspects of all this that remind me a bit of the 2000s housing bubble. And when news comes out about loans with default risks in an industry where everyone has been fighting to issue credit and handing out blank checks without restraint, it all starts to click.

The conservative take is to say this will last 3 or 4 years. There's little risk in that prediction, and whoever makes it can always claim they just didn't account for unpredictable event X if they're wrong.

The reports from BBVA's Research Department are quite prestigious when it comes to forecasting the trajectory of the Spanish economy. In 2007, the bank had already predicted that the rise in housing prices was slowing down and forecasted 1.4% growth for 2008 and stagnant prices for 2009. The reality is that there were 4-5% drops in both 2008 and 2009. Furthermore, the report claimed things like the market would experience a "soft correction" that would have a "limited effect" on economic growth, or that employment and families were out of harm's way. I don't need to say how "comical" all that sounds in hindsight.

I'm not saying the same thing will happen now: if the AI bubble suffers a major burst, I highly doubt it will directly impact the Spanish economy (let alone as severely as in 2008). Indirectly, we'll see. What I'm saying is that it only takes a couple of shocks to completely upend all these forecasts.

And when I see a veteran AI company go bankrupt, AI corporate debt swaps soaring, and services disappointing with their numbers among other things, all while the narrative of hardware shortages and multi-year contracts for entire production runs continues, and investors are simultaneously offloading positions just days before Meta and others release earnings, the whole situation starts to smell like trouble.

[Yorus]

@cobito If this blows up, in Spain there are two ways to look at it. On one hand, it will affect us less since actual investment is lower and maybe people haven't gone quite as crazy. In the US, it could be a major crash. On the other hand, the US companies that do push forward will hold a huge advantage over us (referring to the European Union), which isn't exactly cause for celebration, because it's still the future, but it needs its correction just like the dot-coms did.