
CoreWeave Doubled Revenue and Still Lost More Money
CoreWeave's revenue nearly tripled to $3.661 billion, yet nine-month losses widened to $3.80 per share. A bear case calls it a middleman whose costs scale with sales.
- By
- Tom Whitfield
- Filed
- Channel
- AI & Compute
- Read
- 4 min read
CoreWeave's revenue nearly tripled year over year to $3.661 billion in the third quarter of fiscal 2026, yet net losses widened from -$1.87 per share to -$3.80 over the trailing nine months. That pairing — fast growth and deepening losses — is the core of a new bear thesis that compares the neocloud provider to WeWork and Safehold, two companies that also promised to reinvent an existing market and struggled to hold margins.
What is the bear argument?
The analysis, published as an opinion piece, frames CoreWeave (CRWV) as an unnecessary middleman. The company buys or rents data centers, fills them with high-end NVIDIA (NVDA) chips, and leases that AI-ready computing capacity to hyperscalers and other large users through long-term contracts.
The problem, the author argues, is that both sides of that chain can already deal with each other directly. NVIDIA sells chips to hyperscalers today, and data center developers already build facilities for them. CoreWeave inserts itself between parties that do not need an intermediary.
The comparison to WeWork and Safehold is structural rather than rhetorical. WeWork leased office space and re-leased it to tenants; Safehold buys land under buildings and leases it back to building operators. Both struggled because their suppliers or customers could replicate the service themselves. In WeWork's case, office owners could lease directly to the end tenant; with Safehold, building owners could simply keep their land.
Why do counterparties tolerate the middleman?
The answer, in the author's view, is price. WeWork won leases by paying more than tenants would. Safehold bought land at a 3%-5% cap rate, far above its value to the seller. CoreWeave, the argument runs, wins business by accepting very low margins.
The customers reinforce this reading. Amazon, Meta, OpenAI, Anthropic and other major AI players clearly have the sophistication to build and operate their own AI-ready data centers, the piece notes. NVIDIA sells to the highest bidder and does not care whether that is Amazon or CoreWeave. The hyperscalers are choosing to let CoreWeave make the capital investment on their behalf — they would rather have the $87 billion of liabilities sit on CoreWeave's balance sheet than their own.
Customer concentration sharpens the risk. CoreWeave's own 10-Q filing states: "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."
Why scale does not fix the margins?
Young companies often run net losses, and the author concedes that the $1.27 per share net loss in 3Q26 alone is not proof of a broken model. The deeper issue is cost structure. Revenue rose from $1.365 billion to $3.661 billion year over year, but operating expenses rose almost in lockstep, from $1.313 billion to $3.454 billion.
Normally, overhead spreads across a larger revenue base and margins improve. CoreWeave's two largest expenses are variable, not fixed: depreciation scales linearly with more equipment owned, and interest expense scales linearly with more debt. The tenth data center costs as much as the ninth. Growth has been paid for with extraordinary amounts of debt used to buy rapidly depreciating equipment.
What could prove the bear case wrong?
The thesis carries a stated risk. Depreciation is a non-cash charge, so CoreWeave can generate significant cashflow even while reporting losses. The open question is how much of that cashflow must be reinvested in replacement equipment.
CoreWeave depreciates its technology equipment over a six-year schedule, roughly in line with hyperscaler practice. If the chips actually retain value well beyond six years, a larger share of cashflow becomes true earnings and the margin picture improves substantially. If the equipment loses most of its value within the depreciation window, essentially all cashflow goes to replacement just to maintain revenue.
The bottom line from the source: when counterparties can perform a middleman's function themselves, the only durable way to keep the business is to compete on price. Unless hardware lasts far longer than expected, the author expects CoreWeave — and the neocloud sector broadly — to follow the path of WeWork and Safehold.
Source: seekingalpha.com
More from Tom Whitfield
Show full bio
Staff writer covering companies, deals and strategy in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,054 articles
Related articles
coreweave-s-87b-balance-sheet-the-middleman-that-hyperscalers-don-t-need
CoreWeave's $87B Balance Sheet: The Middleman That Hyperscalers Don't Need
nvidia-posts-59-69b-quarter-puts-6-3b-into-coreweave-deed950e
NVIDIA Posts $59.69B Quarter, Puts $6.3B Into CoreWeave
coreweave-s-troubles-rattle-confidence-in-the-ai-infrastructure-boom-b2702459
CoreWeave's Troubles Rattle Confidence in the AI Infrastructure Boom
coreweave-s-troubles-are-rattling-the-ai-boom-markets-notice-6fcd00f2
CoreWeave's Troubles Are Rattling the AI Boom, Markets Notice



