Last night's CPI data was unremarkable, but the market is already searching for the next sector to carry the AI narrative

Last night's CPI data came in right in line—3.4%, matching market expectations, without stirring up too much turbulence on the macro front.

But last night's U.S. stock market had no shortage of highlights. The biggest focus was this: with the storage chip narrative cooling off in the short term, which sector is next in line to carry the AI story? Beyond optical communications, which we've discussed before, several companies' earnings reports last night are telling the market this: Neocloud (next-generation AI cloud) is emerging as a strong candidate.

1. Doubling Revenue, and a Backlog Expanding at Triple Speed

On the U.S. stock market last night, two focal names in the Neocloud sector—CoreWeave (CRWV) and Nebius (NBIS)—delivered impressive earnings, closing up 19% and 34%, respectively.

First up, CoreWeave. Quarterly revenue hit $2.58 billion, doubling year-over-year. Even more striking is the order backlog: it has swelled to $104.2 billion, expanding 2.5x in twelve months. And with Q3 just getting underway, new contracts have already added another $25 billion.

Nebius' growth was even more pronounced: quarterly revenue of $582 million, up 454% year-over-year. Its AI cloud business has reached an annualized run rate of $3 billion. On the operational side, EBITDA turned positive for the first time.

Put the two earnings reports side by side, and they point to the same conclusion: existing compute capacity is being snapped up entirely by customers, and even so, supply still can't keep up with demand.

2. The Moat in This Business Isn't Chips—It's Electricity

Notably, the way these new cloud providers are building their moats is shifting from "hoarding chips" to "hoarding electricity."

The logic isn't hard to follow: when GPUs are scarce, paying a premium on orders always gets the job done. But substations, transmission capacity, and grid interconnection permits—none of these can be fast-tracked with money alone. The chip shortage is a commercial problem; the electricity shortage is a physics problem.

Nebius' recent deal pricing is the perfect illustration: for short-term, fast-delivery compute, pricing has reached $40–50 million per megawatt, while long-term contract pricing sits at just $20–25 million per megawatt—short-term orders are priced at roughly double the rate of long-term contracts.

Even more interesting is the company's sales strategy. For 2027 capacity, customers are willing to buy out entire tranches right now, but management is deliberately holding back a portion and not signing. It's not that they can't sell—it's that the scarcer electricity becomes, the more valuable that capacity becomes down the road. The pricing power needs to stay in their own hands.

3. Burry's Short Thesis, Taken Apart Point by Point

These two earnings reports also happened to dismantle a short seller's case.

According to public reports, short seller Michael Burry had previously questioned whether some vendors were depreciating GPUs over five to six years, when the actual usable lifespan of GPUs might only be two to three years—suggesting the sector's profits could be overstated.

But the details disclosed by both companies on their earnings calls almost systematically refuted that premise.

First, assets haven't depreciated. Nebius held its first public auction for compute capacity, and the hammer price came in 15% higher than the company's previous all-time high quote. An asset that, per the short thesis, "only has two to three years of residual value left"—and yet buyers are bidding it up.

Second, the lifespan far exceeds the assumption. CoreWeave confirmed signing a new A100 compute contract at favorable pricing, with the term extending all the way to 2029. This is a chip that launched in 2020 and is already three generations behind—by the short model, it should have been written off long ago, yet in reality, it's locked in a nine-year commercial contract. The company also noted that pricing for older products is no worse—and in some cases better—than it was years ago, with the legacy fleet essentially fully booked.

One analyst described the current compute crunch this way: customers can no longer afford to be picky—if they can't get their hands on the latest high-end cards, they'll take the older models just the same.

4. What the Shorts Missed: Renewal Revenue Is Almost Pure Profit

If the first two points only overturned the "lifespan assumption," the third point strikes directly at the short's valuation model.

These legacy clusters being renewed had already paid off all asset-side loans during their initial contract terms, and depreciation had been fully booked. In other words, for every dollar of revenue from renewals, the only costs against it are electricity and minimal operating expenses—there's almost nothing else eating into it.

An A100 contract signed through 2029 flashes two cards at once: the commercial lifespan of older chips is far longer than the shorts imagined, and these assets—counted as "zero" in the short model—are steadily churning out cash that's nearly pure profit. Burry wasn't just wrong about the depreciation period as a single parameter; he completely missed the fattest second revenue stream in the entire business model.

5. Final Thoughts

Connecting the dots across recent market action, the power transition within the AI trade is now quite clear: the storage narrative peaked first, optical communications picked up the baton on earnings momentum, and now Neocloud has added a triple confirmation of "capacity sold out, electricity pricing power, and a bankrupted short thesis."

For investors, the stock-picking criteria going forward can be compressed into three questions: Do they have access to electricity? Is order visibility high? Can legacy assets keep generating cash? Those who check all three boxes are the hard assets of the new narrative; the pseudo-compute plays that only know how to tell stories and burn capital will be weeded out faster in this "audit-style" market cycle.

In the second half of the AI trade, having a loud voice doesn't count for much—you need to hold the real goods.

Risk Disclosure: This article is contributed by a third-party author and reflects only the author's personal views, and does not represent BIT's position or opinions. The market data, company financial information, and analytical judgments referenced herein are sourced from public channels. BIT makes no guarantees as to their accuracy, completeness, or timeliness, and shall not be held liable for any losses arising from the use of this content. This article is for informational purposes only and does not constitute any investment advice, solicitation, or recommendation to purchase or sell any financial product. Markets involve risk, and investment requires caution. Historical performance does not guarantee future returns. Investors should make independent judgments based on their own risk tolerance and take full responsibility for their investment decisions.