U.S. CPI came in at 3.4% overnight, right in line with market expectations, and did little to change the broader macro setup. The more active debate in equities centered on something else: where the AI trade goes next.
With the storage-chip theme losing momentum in the near term, investors are looking for the next segment that can carry AI-related positioning. Beyond optical communications, which had already been part of that discussion, the latest earnings from several companies are pushing another candidate forward: Neocloud, or next-generation AI cloud providers.
CoreWeave and Nebius put up strong numbers
Two closely watched names in the Neocloud group, CoreWeave (CRWV) and Nebius (NBIS), released earnings that drew a strong market response. Their shares closed up 19% and 34%, respectively.
CoreWeave reported quarterly revenue of $2.58 billion, double the level from a year earlier. Its backlog reached $104.2 billion, up by two and a half times over 12 months. The company also added another $25 billion in newly signed contracts just as the third quarter began.
Nebius posted even faster growth. Quarterly revenue reached $582 million, up 454% year over year. Its AI cloud business is now running at an annualized pace of $3 billion, and EBITDA turned positive for the first time.
Taken together, the two reports point to the same conclusion: existing compute capacity has already been heavily booked by customers, and supply still is not enough to meet demand.
The moat is shifting from chips to power
The article argues that the way these new cloud providers build barriers to entry is changing. The old emphasis was on stockpiling GPUs. The new emphasis is on securing power.
The reasoning is straightforward. When GPUs are scarce, companies can still try to solve the problem by paying more. Power infrastructure is different. Substations, transmission capacity, and grid interconnection permits cannot be created quickly just by spending more money. In that framing, a chip shortage is a commercial problem; a power shortage is a physical one.
Nebius’ recent pricing is presented as a clear example. Short-cycle, fast-delivery compute sold for $40 million to $50 million per megawatt, while long-term contract pricing was only $20 million to $25 million per megawatt. On that basis, short-duration orders were priced at roughly double long-term deals.
The article also points to sales strategy. Customers are already willing to buy out 2027 capacity, but management has deliberately held some of it back instead of signing everything now. The point, it says, is not weak demand. The point is that as power becomes scarcer, capacity reserved for later may command better pricing, allowing the company to keep control over price setting.
Short arguments on depreciation are being tested
The piece cites public reporting saying that short seller Michael Burry had previously questioned whether some companies were overstating profits by depreciating GPUs over five to six years even though their practical useful life might be only two to three years.
Details disclosed on the companies’ conference calls are presented as a challenge to that thesis.
First, the assets are not showing the kind of value decline that thesis would imply. Nebius, in its first public auction of compute capacity, saw a final sale price that was 15% above the company’s previous highest quoted price. If those assets were really down to only two or three years of residual value, the article argues, buyers would not be bidding them higher.
Second, useful life appears longer than bears assumed. CoreWeave confirmed a newly signed A100 compute contract that extends through 2029 and was described as having attractive pricing. The A100 was introduced in 2020 and, according to the article, is already three generations behind. Even so, it has now been tied to a commercial contract stretching across nine years. The company also said pricing for older products is not weaker than it was several years ago and may even be stronger, while most of the older fleet is already booked.
The article quotes an analyst’s description of the current compute shortage: customers are no longer in a position to be selective, and if they cannot secure the newest cards, they are still willing to take older ones.
Renewal revenue is a key part of the model
The article goes one step further and says the short case may also be missing a major source of earnings power: renewal revenue from older clusters.
By its account, many of these older clusters had already repaid their asset-side loans during the initial contract term, and their depreciation had already been fully recognized. That means each renewal dollar is matched mainly against power costs and limited operating expenses.
Using the A100 contract that runs through 2029 as an example, the article says the disclosure reveals two things at once. Older chips are remaining commercially useful longer than bears expected, and assets that might have been treated as effectively worth zero in a short model are still producing cash flow that comes close to pure profit.
In that reading, the problem for the short thesis is not only the depreciation assumption. It is also the omission of a second, highly profitable revenue phase within the business model.
AI leadership inside the market keeps rotating
The article concludes that leadership within the AI trade has become clearer over recent sessions. Storage expectations topped out first. Optical communications then took over on earnings momentum. Now Neocloud is presenting what the article calls a three-part validation: sold-out capacity, pricing power tied to electricity, and fresh evidence working against the core short narrative.
For investors, it reduces the selection framework to three questions: whether a company has access to power, whether order visibility is strong, and whether older assets can keep generating cash. Names that meet those conditions are presented as the sturdier part of the new theme, while companies that rely on narrative alone and continue burning capital may face a harsher test as the market focuses more closely on the numbers.
The original article also included a risk disclaimer saying the content reflects only the third-party author’s personal views, does not represent BIT’s position, and should not be treated as investment advice.