$XL2CSOPHYNIX(07709.HK) Computing power surplus? The faith is starting to crumble. Meta announced plans to build a cloud business and rent out its excess AI computing power.

As soon as the news broke, the U.S. stock AI sector crashed. Computing power lessor Nebius plunged over 12%, CoreWeave fell 11%; memory chips were completely wiped out, SanDisk dropped over 7%, Micron fell over 6%, Seagate dropped 7%.

But Meta rose over 6%, showing that what crashed wasn't AI, but the faith in computing power.

Because the market discovered that computing power is no longer scarce; to put it bluntly, it has gone from being a nuclear weapon to being a cabbage.

And the reason Meta is selling computing power is because even the landlord has no extra grain.

Last year, global tech giants, afraid of being left behind by the AI wave, went crazy buying GPUs. Meta alone hoarded hundreds of thousands of NVIDIA high-end GPUs. AI-related capital expenditure for 2026 is expected to reach $145 billion. The combined capital expenditure of the nine major cloud providers this year is estimated at $830 billion.

The problem is, they were so afraid of not having enough that they bought too much.

The result? After the large model training is done, daily inference doesn't use that much.

An analyst tracking the computing power growth curves of major AI labs discovered a disturbing pattern: the rate of computing power procurement far exceeds the rate of computing power consumption.

In plain language, GPUs are gathering dust in server rooms, while electricity bills are flowing away.

Can capitalists tolerate this? No, they can't.

So Meta is simply launching a cloud service to become a computing power landlord, renting out those dust-gathering cards. Even if it's cheaper, it's better than letting them rot in their hands.

Zuckerberg hinted at this back in May, saying it was definitely under consideration.

But what the market is really panicking about isn't Meta selling computing power, it's the ripple effect.

What is the business model of companies like CoreWeave? Hoard cards at low prices, rent them out at high prices.

Now that this super scalper, Meta, is personally entering the market to dump inventory, how can a price war be fought? The premium space for small and medium-sized lessors is directly shattered.

This is also what the market is panicking about tonight. This means the business model of computing power leasing is changing.

Furthermore, if the major giants have hoarded so many cards and demand falls short of expectations, HBM and high-end memory will be the first to suffer.

Finally, the faith in the AI narrative, especially the faith in computing power, is beginning to waver.

Big short-seller Michael Burry disclosed today that he has shorted the AI and chip sectors, shorted Tesla at $416, and also built a short position in NVIDIA. He said this is a judgment on the AI bubble.

Those selling the shovels may not necessarily make money, but the news that "there are too many shovels, and they're starting to be sold at a discount" is scary enough on its own.