Neocloud Stocks Fall After Musk's Warning: What's Wrong?

An illustration of a falling stock chart over GPU server racks in a data center

On October 25, US time, Elon Musk fired off a pretty harsh warning to the AI computing industry. In a post on X, he wrote that more than 200 neocloud companies will face "guppy-level death and consolidation."

In other words, they'll get weeded out and merged away like little fish.

Elon Musk... I wonder how this guy finds time to be that active on social media when he's so busy... Does he have some item that slows time down just for him?? Anyway, here's his argument.

Companies that lean too hard on AI data center revenue and have no other stable income have had a very tough time raising money lately.

It probably wasn't purely because of Musk's remarks, but the next day in US premarket trading, CoreWeave and Nebius were trading about 2% to 5% lower, and selling pressure continued on IREN and other related stocks.

✦ CoreWeave Shares Fell About 33.9% in Two Weeks

US regular-session closes (USD), Oct 9 to Oct 23, 2026

546882Oct 14: -15.46%Oct 16: -11.91%Oct 22: -10.06%Oct 9Oct 16Oct 23
Related stocks also fell in premarket trading on Oct 26. Premarket moves are a separate snapshot, so they aren't included in this closing-price series.

These stocks hadn't been doing well lately anyway, and this extra drop hurt shareholders even more. CoreWeave's regular-session close fell from $82.08 on October 9 to $54.24 on October 23. That's a drop of about 33.9% in two weeks.

It fell 15.46% in a single day on October 14, 11.91% on the 16th, and another 10.06% on the 22nd.

The market was already worried about the heavy debt, the high cost of raising money, and how long it takes to earn back data center investments. Musk's warning poked at those worries even more.

〈Bloomberg on why debt worries keep driving down CoreWeave stock〉
* From the official Bloomberg Television channel.

I Get the Cloud, but What's a Neocloud?

There was a time when even the word "cloud" sounded pretty foreign to us, right? Now it's totally familiar. I'd bet everyone pays for at least one cloud service.

Starting with storage like Microsoft OneDrive, Google Drive and Apple iCloud, countless cloud services are flooding the market. Even ChatGPT and Claude, which we now use like part of daily life, are cloud services running on servers somewhere in the world.

So what is a neocloud?

✦ What Each Company Actually Does

Same group of stocks, different businesses

CoreWeaveGPU cloud specialist
Builds and runs the GPU infrastructure and cloud platform for AI training and inference, and provides computing resources to customers
NebiusAI cloud company
Rents out full-stack infrastructure, from GPU clusters to data centers and developer tools
IRENStarted in bitcoin mining, expanded into AI cloud
Secures its own power, land and data centers, and offers GPU cloud too
Applied DigitalBuilds and leases data centers
Rents space with power and cooling equipment for AI and HPC to large customers

A neocloud is a cloud operator that specializes in renting out the GPU computing needed to train and run AI models. CoreWeave and Nebius are the typical examples, and IREN and Applied Digital are also lumped in with neocloud stocks by investors.

The business structures of these companies aren't all the same. Some provide GPU cloud services directly, while others put more weight on data centers and power infrastructure.

✦ What a Neocloud Needs Before It Can Earn

Spending comes before revenue

1Secure GPUs
Buy the expensive AI GPUs
2Build the data center and cooling
Get the building and cooling equipment in place
3Connect power and network
Bring in electricity and build out the network
4Provide computing
Deliver computing resources to customers

The hardest part of this business is that you have to spend big money before you earn any revenue.

You need to secure pricey GPUs, set up data center buildings and cooling equipment, bring in power, and even build the network before you can provide computing resources to customers. Hotels work the same way.

You have to put in a lot of money up front to build the place, and revenue only trickles in once guests start using it, which is how you earn the investment back. It's similar, but the catch is the nature of the neocloud industry.

AI GPUs are very expensive, and new generations come out fast. If the pricey equipment sits idle or power supply runs late, costs and interest keep piling up while earning the investment back keeps getting pushed later.

Give it time, and the GPUs you bought become outdated before you know it, and you end up having to swap in new ones without ever earning the investment back.

✦ Google Colab: Then vs. Now

An example of equipment going out of date

A few years ago
Colab's GPUs were among the better ones, and users were very satisfied
Today
Local GPUs have caught up with, or even passed, what Colab offers

Take Google Colab, a cloud computing service, as an example. Only a few years ago, the GPUs Colab offered were among the better ones. That let people test jobs on Colab that they could never run on a local GPU, and users were very satisfied.

But local GPUs have improved a lot lately, and they've caught up with, or even passed, the GPUs Colab offers. So if you're on Colab's free plan, you're surely going to feel dissatisfied. It feels way too slow.

Google surely invested a lot to build that infrastructure. But that infrastructure has become outdated before long.

Not long ago I tried a local LLM experiment on Google Colab, got so frustrated, and just quit.

So strong AI demand alone doesn't guarantee a neocloud business will succeed. How many GPUs you've secured matters less than how much that equipment actually runs, and whether you can deliver the service you promised customers on time.

Then there's whether any cash is left after subtracting operating costs, interest and extra investment from revenue, and whether the equipment you've already invested in is going out of date too fast... Way too many things to consider.

Even if revenue grows fast, if you have to bring in big outside funding every time you expand a data center, shareholders have to give up dividends and take on the risk all over again.

〈An in-depth explainer on the economics of neoclouds〉
* From the Caleb Writes Code YouTube channel.

Why Did Musk Say His Company Is Different?

Musk claimed that SpaceX has finished raising $40 billion to buy GPUs. The difference he pointed to is where the money comes from.

SpaceX has revenue from Starlink and rocket launches, and the explanation is that it can raise funds on the strength of its corporate credit.

On the other hand, Musk's logic is that neocloud companies that depend almost entirely on AI data center revenue will struggle to keep borrowing when that revenue wobbles.

✦ SpaceX's $40 Billion Raise: Conflicting Accounts

Whether it's actually done isn't confirmed

Musk's claim
• SpaceX has finished raising $40 billion to buy GPUs
Bloomberg's report
• Citing anonymous sources, it says talks over pricing were still ongoing

That said, it's unclear whether his claim that the $40 billion raise is fully wrapped up is true. Musk wrote that the deal was done, but Bloomberg reported that, according to people familiar with the discussions, "talks over pricing were still ongoing."

We can't tell for sure whether Elon Musk's words are true, but either way, in this industry right now, the gap between "we can raise funds!" and "we actually raised the funds!" is huge.

Lately there's been a lot of talk about CoreWeave's surging bond yields, too. In Reuters' October 16 report citing LSEG data, CoreWeave's debt was about $50 billion, and some of its bonds were yielding 15% to 16% in the market.

Even with the track record CoreWeave has built, its contract backlog ahead and its sheer size, the market priced that risk very high.

✦ Short Interest as a Share of Float

As of October, relative to free-floating shares

CoreWeaveAbout 23.7%
Of free-floating shares sold short
NebiusAbout 20%
Of free-floating shares sold short

The stock market was pretty bearish too. Citing ORTEX data, Reuters reported that "Investors betting against US neocloud and data center companies have raked in $6 billion in paper profits so far this month."

How much shorting was there? CoreWeave's short interest was about 23.7% of its free-floating shares, and Nebius's was about 20%. You can feel how much conviction the short sellers had to short so heavily.

Where Does the Neocloud Industry Go From Here?

✦ Gartner's Neocloud Outlook

Through 2030

The outlook
• More than 80% of operators expected to stop operating independently by 2030
The main reasoning
• A worry that revenue may fall short of the scale of investment and financing costs

Market research firm Gartner forecasts that by 2030, more than 80% of neocloud operators will stop operating as independent companies. To put it bluntly, 80% of neocloud companies will go bust or get sold off within three to four years.

The key reasoning is that "our research indicates that enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures during this period." Of course, "ceasing independent operation" doesn't necessarily mean bankruptcy.

It also includes cases where a company is acquired or merged with another and doesn't remain an independent company.

✦ CoreWeave: Backlog vs. Spending

Signed contracts aren't cash yet

$104.2 billionRevenue backlog
End of Q2 2026, plus over $25 billion in new commitments added in early Q3
$2.575 billionQ2 2026 revenue
$9.4 billionQ2 2026 capital expenditure
Nearly four times revenue

CoreWeave reported a revenue backlog of $104.2 billion at the end of Q2 2026, and said it added more than $25 billion in new customer commitments in early Q3. That points to substantial demand and contracts.

But that doesn't mean that much cash has come in right now. Revenue can only be recognized once the service is delivered and the contract terms are met. CoreWeave's Q2 2026 revenue was about $2.575 billion, and its capital expenditure was $9.4 billion.

In other words, it's investing nearly four times its revenue. No wonder some investors are starting to ask, "Is this really okay...?"

On October 22, Nebius announced that it had obtained a $3.5 billion secured credit facility from Citi.

According to the company, the terms are a 364-day maturity and an initial interest rate of Term SOFR plus 2.75%, with the spread able to rise as high as 3.50% depending on conditions.

It also pointed to long-term contract commitments of more than $40 billion as a foundation for its funding. Investors could relax for now, but a loan with such a short maturity leaves the burden of finding funding again before long.

In the end, what the market doubts isn't AI computing demand itself. The question is which companies can turn that demand into profit while it grows, and whether they have the funds and credit to hold out until then.

Companies with long-term contracts and quality customers, and those that can bring power and data centers online on time, may have the advantage.

Conversely, companies that can't generate enough cash flow to cover high interest and big investment costs will disappear one by one.

This isn't investment advice. You know that, right? Let's each protect our own hard-earned money.

That's all for this post. See you next time.

Related posts

No comments:

No comments yet. Want to leave the first one?