Companies

CoreWeave Doubled Revenue and Still Lost More Money: The Middleman Problem

A bearish analysis argues CoreWeave's tripled revenue hides structural margin failure, comparing the $87B-liability AI compute middleman to WeWork and Safehold. Equipment durability is the swing factor.

CoreWeave: An Unnecessary Middleman
CoreWeave: An Unnecessary Middlemanblickpixel / pixabay
By Rebecca Stone6 min read

Why it matters

  • CoreWeave revenue nearly tripled year over year to $3.661B in 3Q26 from $1.365B, while net loss reached $1.27 per share.
  • Top two customers generated approximately 30% and 24% of revenue in the quarter ended September 30, 2026, per the 10-Q.
  • Nine-month net losses widened from -$1.87 per share in 2025 to -$3.80; operating expenses rose from $1.313B to $3.454B.
  • The analysis says $87B of liabilities sit on CoreWeave's balance sheet so hyperscalers avoid the capital investment.
  • CoreWeave depreciates technology equipment over a 6-year schedule, making chip durability the key variable in the bear thesis.

CoreWeave nearly tripled revenue to $3.661 billion in the third quarter of 2026, yet its net loss widened to $1.27 per share, and a new bearish analysis argues that this gap is structural, not transitional. The analysis, published as a short thesis on the stock (CRWV), contends that the AI infrastructure provider is an unnecessary middleman whose largest customers and key supplier could replicate its function without it — and that the company's thin margins and heavy leverage set up substantial value destruction when data center demand normalizes.

The stakes extend beyond one company. CoreWeave has become one of the most visible bets on the AI compute buildout, having assembled enormous AI-ready capacity by buying and renting data centers and filling them with high-end NVIDIA (NVDA) chips, then leasing that capacity to hyperscalers and AI labs under long-term contracts. If the thesis is right, the fastest-growing segment of the AI infrastructure market is running on a business model that only works while counterparties find it convenient.

What is the middleman argument?

The analysis frames CoreWeave's model as one that inserts itself between two parties that would otherwise deal directly. NVIDIA already sells chips to hyperscalers. Data center developers already build data centers for hyperscalers. CoreWeave buys both and re-leases the equipment and space back to Amazon, Meta, OpenAI, Anthropic and other major AI players.

"I view this sort of business model as that of a middleman," the author writes. "It inserts itself between two parties that would otherwise interact directly."

The comparison the analysis draws is deliberate and unflattering. WeWork leased office space and re-leased it to tenants; Safehold (SAFE) buys land beneath buildings and leases it back to building operators. Both companies, the analysis notes, struggled with margins "because the service they are providing could be performed by either their supplier or their customer." The office owner could have leased directly to the end tenant. The building owner could have kept the land.

The author points to a shared tell: all three companies leaned on the word "reinventing" in their marketing — WeWork's "We are reinventing the way people work," Safehold's reinvented ground lease, and now CoreWeave. The slogan, in the author's telling, marks businesses that repackage what their counterparties already know how to do.

Why do hyperscalers use CoreWeave at all?

The analysis answers this question with incentive logic rather than capability gaps. Amazon, Meta, OpenAI and Anthropic, the author writes, "clearly have the sophistication to build and operate their own AI-ready data centers" and can all buy chips from NVIDIA directly.

So why do they let CoreWeave carry the load? The author's answer: price and balance-sheet relief. "The hyperscalers and AI labs are choosing to let CRWV make the capital investment on their behalf," the analysis states. "They would rather have the $87B of liabilities sit on CoreWeave's balance sheet than their own." NVIDIA, for its part, "doesn't care whether they sell to AMZN or CRWV," the author writes — it sells to the highest bidder.

This is where the concentration data matters. CoreWeave's own 10-Q discloses: "A substantial portion of our revenue is driven by a limited number of customers. We recognized approximately 30% and 24% of our revenue from our top two customers for the three months ended September 30, 2026." A business dependent on a handful of counterparties, each of which could in-source the function, has limited pricing power — a point the analysis treats as central rather than incidental.

The author concedes the model would be stronger if CoreWeave's customers were smaller entities incapable of building their own AI compute. "Unfortunately, nearly all of CoreWeave's business is to hyperscalers and AI labs per the 10-Q," the analysis states.

Can CoreWeave scale its way to better margins?

The numbers say no, according to the analysis. Revenue nearly tripled year over year, from $1.365 billion in the third quarter of 2025 to $3.661 billion in the same quarter of 2026. Normally, margins improve as revenue grows because fixed overhead spreads across a larger base. But CoreWeave's operating expenses also nearly tripled, from $1.313 billion in 2025 to $3.454 billion in 2026.

The nine-month figures sharpen the point: net losses grew from -$1.87 per share in the nine months ended September 30, 2025 to -$3.80 in the most recent nine months. Growth, in this model, does not compound profitability.

The reason, the analysis argues, is that CoreWeave's largest expenses are variable, not fixed:

  • Depreciation expense scales linearly with more equipment owned.
  • Interest expense scales linearly with more debt.

Every incremental dollar of revenue requires building or renting more AI-ready data centers to lease to additional customers. "The 10th data center costs just as much as the 9th," the author writes. "There are extremely diminishing benefits of scale here, as CRWV is already large." The growth, the analysis adds, "was paid for by using extraordinary amounts of debt to buy rapidly depreciating equipment."

The author is careful to distinguish a young company's losses from a structural problem. CoreWeave, founded in 2017 as a crypto mining operation before pivoting to AI compute supply, would be expected to run losses at this stage. The single-quarter $1.27 per share net loss in 3Q26 is not, by itself, the evidence. The anatomy of the cost structure is.

What could prove the bear thesis wrong?

The analysis flags one variable that could change the outcome: equipment durability. Because the low or negative margins are largely driven by depreciation — a non-cash expense — CoreWeave has potential for significant cash flow. The open question, the author writes, is how much of that cash flow must be reinvested in replacement equipment.

CoreWeave depreciates "Technology equipment" over a six-year schedule, roughly in line with hyperscaler practice. The sensitivity runs both ways:

  • If equipment retains value well beyond six years, more of the cash flow becomes true earnings, and margins look materially better.
  • If the equipment loses most of its value within six years, essentially all cash flow must be spent on replacements merely to maintain revenue — confirming the margin problem.

"CoreWeave's outlook is substantially better if the equipment lasts materially longer than 6 years," the analysis states. Future profits therefore hinge on the durability and obsolescence rates of chips and other hardware.

What happens next?

The analysis closes with a direct verdict: "I think CoreWeave's business model will suffer recurring margin issues due to its key functions being replicable by tenants and suppliers. If one's counterparties can do it on their own, the main way to keep business is to compete on price." The author expects CoreWeave to follow the path of WeWork and Safehold, and extends the bearish view to the broader category: "I am bearish on CoreWeave and the rest of the Neoclouds."

For the AI infrastructure market, the thesis poses a question that outlasts any single quarter: whether the current buildout is creating durable businesses or intermediaries that exist only as long as their giant counterparties prefer renting the risk to owning it. The answer will arrive in the depreciation schedules.

Source: seekingalpha.com

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. CoreWeave Revenue Jumps 168% to $3.661 Billion, Backlog Hits $132.5 Billion
  2. CoreWeave's $104.2B Backlog Hides a Debt Bill Its Landlord Pays First
  3. CoreWeave Stock Jumps 13% After Blowout Quarter and Raised Outlook
  4. CoreWeave Reports 168% Revenue Growth While Carrying $41.6B Debt
  5. CoreWeave Downgraded: 4.6 GW of Power, $13.4B of Debt, and a 2030 Deadline

« Previous articleNext article »