Musk Is Bearish on AI Compute: More Than 200 New Cloud Service Providers Will Die Off in Large Numbers

In order to grab computing power, Silicon Valley giants are willing to spend tens of billions of dollars; and a host of "Neoclouds" service providers without big-company backing are, through frenzied borrowing, buying tens of thousands of Nvidia GPUs, trying to play the role of "shovel sellers" in this gold rush.

However, this game of passing the buck is now approaching its endgame.

Recently, Elon Musk held an extremely hardcore conversation with well-known investor Chamath Palihapitiya on X. Musk pointedly burst the bubble of the current false prosperity in the AI computing power market: over the past few weeks, the debt market has almost completely cut off funding channels for companies that do not have substantive revenue independent of AI data centers.

What followed was Musk's extremely cold prediction: "That means those 200-plus 'Neoclouds' service providers will undergo a guppy-like die-off and consolidation. Some will survive, but not many."

At the same time, investor Chamath, who once precisely shorted the subprime crisis, added another jab, pointing out that in 2027 there will be a large amount of "stranded compute" and half-finished data centers.

The debt market's sudden brake: the elimination of 200 neoclouds

To understand this crisis, one must first understand what "Neoclouds" are.

Over the past two years, companies like CoreWeave have risen abruptly. Unlike Amazon AWS or Microsoft Azure, they do not have a massive general-purpose cloud computing ecosystem. They do only one thing: buy GPUs from Nvidia and then rent them at high prices to large model companies or AI startup teams in urgent need of computing power.

This is essentially an extremely extreme financial game of "high leverage + heavy assets." These companies have no stable cash flow support and can only rely on Wall Street debt financing to buy graphics cards that cost tens of thousands of dollars each. Their business model is extremely fragile: as long as the price of computing power leasing falls, or the chain of borrowing new debt to repay old debt breaks, the company will instantly fall into dire straits.

In the conversation, Musk pinpointed the current fatal weakness with extreme precision: the funding tap of the debt market has been turned off.

After a long period of frenzy, Wall Street financial institutions have finally begun to examine risk. When they discovered that the revenue of many AI startups simply cannot cover their high computing power bills, and that "Neoclouds" companies not only lack collateral but are also extremely dependent on high-interest debt, capital's instinct for risk aversion was fully activated.

Musk used his own company as an example: "SpaceX raised $40 billion in debt for GPUs, but we have an investment-grade credit rating and have Starlink and launch revenue. If you don't have real cash flow, don't expect to get a single cent from the debt market."

This is the cruelest reality: those more than 200 neocloud companies that lack "substantive revenue independent of AI data centers" are about to face broken funding chains. Musk used "a massive guppy-like die-off" to describe this coming wave of bankruptcies. After the capital tide recedes, all that will be left on the beach will be a mess and second-hand graphics cards that cannot be converted into cash.

The "stranded compute" of 2027: ruins after the infrastructure frenzy

In this conversation, Chamath raised an even more fatal industry pain point: the "stranded compute" crisis of 2027.

Chamath said bluntly that he is witnessing a disaster caused by bureaucracy. Many tech giants (Hyperscalers) are extremely slow in the process of building data centers. "Nvidia alone takes three months just to sign a non-disclosure agreement (NDA), and the commissioning period even requires 18 months."

This sluggish supply chain and bureaucracy, combined with physical constraints such as grid capacity and building approvals, have caused many data centers to fail to be completed on time. At the same time, in order to support their own large model businesses, the big companies will tightly hold all self-built computing power in their own hands, prioritizing internal demand and only then allocating to third parties.

This has created an extremely absurd waste of capital: many companies have spent billions of dollars buying top-tier GPUs and building massive server rooms, but by 2027, because there is no power, no network, and no application scenarios, this computing power becomes "half-finished projects," piling up in warehouses gathering dust and unable to generate any returns.

The squeeze of the giants: who is snickering, who is bleeding? In this computing power winter, not everyone is shivering.

As Chamath pointed out, this mismatch will only benefit two types of companies: The first type is non-bureaucratic giants that can quickly establish industrial-scale operations. For example, Musk's xAI, relying on a strong engineering culture, can rapidly build a supercomputing cluster with 100,000 H100s. The second type is players with extremely abundant funds and sufficient agility. For example, Microsoft and Meta have massive cash flow and can ignore short-term fluctuations in the debt market while continuing to stockpile supplies.

For the "Neoclouds" companies and large model startups caught in the middle, they face not only being unable to rent computing power, but also the extremely harsh "digital exploitation" of the giants.

Chamath revealed that big companies now do not even allow third-party model companies to connect directly. They require model companies to go through the big companies' "model router" and extract extremely high tolls from it. He even hinted that in Anthropic's previously exposed confidential S-1 document, it mentioned that "the big companies told them they had no spare computing power." This confirms that the computing power shortage and giant monopoly have reached an extremely dangerous level.

The shadow of shorting: the "double act" of Musk and Chamath?

The most interesting detail of this tweet occurred in the comments section.

When Musk and Chamath were commenting on the AI computing power bubble and expressing pessimistic expectations, Jack Gee, founder of the short-selling firm Hedgeye, fired directly below: "You two work together really well. This looks like a pretty good way to diversify risk and attack competitors."

Jack Gee's mockery was spot-on. During the 2021 SPAC (special purpose acquisition company) bubble, Chamath, relying on frenzied hype and false publicity, harmed countless retail investors and cashed out. Now, he and Musk are heavily invested in AI infrastructure while publicly talking down the "Neoclouds" companies that borrow to stockpile cards.

Their remarks objectively played a role in suppressing competitors. Once market panic intensifies and debt interest rates soar, those competitors who rely on high leverage to buy cards will fall into a dead end, while giants like SpaceX and xAI with stable cash flow and diversified financing channels can instead take advantage of the situation to bottom-fish and complete an industry oligopoly.

Over the past two years, people thought that as long as they bought Nvidia graphics cards, they could print money. But this debt-driven frenzy concealed a huge loophole in the business model. AI infrastructure is essentially a traditional business with long cycles, heavy assets, and low tolerance for error. It requires the cooperation of the power grid, the support of the supply chain, and extremely massive capital reserves.

2027 is destined to be an extremely turbulent year for the AI computing power industry. A large number of unqualified "Neoclouds" service providers will be purged, non-compliant AI startups will be worn down to death, and "stranded compute" lacking power and supporting facilities will become the most tragic relic of this bubble.

In the wave of technological revolution, those bubbles inflated only by PowerPoint and debt will eventually be ruthlessly torn apart by gravity. In the end, those who can laugh last will always be the minority who possess the strongest cash flow, the hardest-core engineering capabilities, and the calmest business minds.