In recent years we’ve seen how the race to artificial intelligence has turned GPUs One of the most coveted assets Technology industry. But getting those chips was just the beginning: before the next generation of hardware arrived, you had to finance them, install them, connect them, and find enough business to recoup the huge investment. This gear has allowed companies that we barely knew a few years ago to grow at high speeds. This may now become your greatest weakness. Elon Musk believes that a significant portion of them will not survive as we know them today.
Traders put figures and recipients on that warning this weekend. The message was immediately echoed in the stock market: CoreWave, Nebius, IREN and Applied Digital closed lower on Monday, Musk wrote in a Reuters report.
Elon Musk’s published message on X about NeoCloud Photo: Elon Musk at X (click to view original publication)
To understand what’s at stake, it’s best to start with the word. Neoclouds They are cloud providers Primarily designed around GPU computing and AI loads, particularly model training and execution. Its offering is more specialized than giants like AWS, Azure or Google Cloud, which also offer storage, database, security and many business services. When a company needs thousands of GPUs working in tandem, these companies try to provide exactly that capacity without forcing them to build an entire infrastructure.
Corewave and Nebius, two companies we’ve already looked at, are part of a category where we also find names like Lambda and Cruso. Their opportunity appeared when the rise of AI triggered the need for GPU power, and many companies looked for specialized options to get it. Instead of competing with hyperscalers on every possible service, these providers focused on meeting that new demand. The bet has done enough to form its own sector, driven by clients who need to train and drive increasingly demanding models.
Neocloud Business
The trickiest part of the model begins here. As Gartner explains, the neocloud business premise is spending cash to acquire less computing power and only work as long as revenue growth continues. To grow, these providers must commit huge amounts of capital before knowing how long it will take to recoup that money: buying GPUs, expanding data centers, securing power and building the infrastructure needed to put that capacity into service. Much of this expansion was financed with debtFinancial leases or structures are backed by assets and contracts. The risk is that the clock has been ticking since day one: today’s highly valued GPU may lose appeal when a new generation arrives, when the payoff begins to expire.
CoreWeave lets us see that paradox quite clearly. According to documentation filed with the SEC, the company’s turnover in the second quarter of 2026 was $2,575 million, up 112% from a year earlier, while its revenue backlog (future revenue already contracted) reached $104.2 billion at the end of June. But this growth coexisted with a much heavier financial structure: it registered a net loss of 626 million dollars, supported 640 million in net interest expenses and accumulated about 35.6 billion in debt. There is business, and a lot of it, but a huge bill to maintain it.
Musk uses SpaceX as a counterexample. The company has reached a binding commitment of approximately $40 billion to finance GPUs and other equipment through leasing, an operation whose final closing remains subject to certain conditions. SpaceX can access that capital because it has investment grade and revenue streams, according to the trader AI is out of the data centerLike Starlink and its launch business. Major rating agencies have given it an investment-grade rating this year. This distinction is key to their argument: A company can assume a large investment in computing if it has other sources of income that can sustain it.
Here it is convenient to distinguish mask warnings from what appears in specific cases. Just a few days ago, Nebius secured a $3.5 billion line with Citi, and Lambda closed guaranteed financing of $926 million in August. Both cases show that capital is available for certain operators, especially when there are long-term contracts, clients with a good credit profile or assets capable of supporting the loan. In this context, pressure may increase for those companies that rely more on continuous growth and refinancing.
It is also not reasonable to read Mask’s prediction as a diagnosis of an unbiased observer. SpaceX acquired xAI in February, operating a massive infrastructure similar to Colossus, and signed a deal to give Anthropic access to Colossus 1. which gives it direct knowledge of the market, But there is also interest in this. Importantly, his thesis is not alone: Gartner also estimates that more than 80% of neoclouds may cease to operate independently before 2030. This can translate into acquisitions, mergers or market exits, consolidation scenarios that are also contemplated by other analyzes of the sector.
The effect, moreover, is not to end the neoclouds themselves. AI infrastructure financing is spreading through loans through banks, funds and other credit vehicles, asset-backed loans and deals related to data centers and computing power, a nexus that S&P Global has already warned about. This does not mean that a wave of closures will necessarily cause an industry-wide crisis. The meaning is simpler: that the demand for artificial intelligence may continue to grow, and even so, those who go into debt to build the infrastructure that supports it are not guaranteed to survive.
Picture | Gage Skidmore (CC BY-SA 4.0, edited by Xataka for cover) | Screenshot @elonmusk (X)
In Xataka | Anthropogenic AI has just taken an important step in the industry: opening its account to the public
