On December 16, 2025, CoreWeave’s sharp selloff became a lightning rod for a broader market debate: is the AI infrastructure boom entering its most fragile phase, where execution hiccups and financing costs matter more than hype? New reporting and market commentary today highlights a mix of catalysts—construction delays at a key Texas data center tied to OpenAI, worries about debt-heavy growth, and a rising chorus of skeptics questioning whether the economics of renting cutting-edge Nvidia GPUs can hold up as hardware ages fast and capital gets more expensive. [1]
CoreWeave isn’t alone. The same anxiety has pressured other AI infrastructure names—especially Oracle, where investors are scrutinizing how a massive data center buildout will be funded, and Broadcom, after margin concerns helped reignite bubble talk earlier this month. [2]
Why CoreWeave has become the market’s AI infrastructure “stress test”
CoreWeave’s rise was powered by a simple promise: build an AI-first cloud optimized for training and running models, stocked with the newest Nvidia GPUs, and rent that compute to customers who don’t want (or can’t) build everything themselves. The company’s customer list has included major names in the AI ecosystem such as OpenAI, Microsoft, and Meta, and its growth has been dramatic—helping explain why the stock ran hard earlier in 2025 before sentiment flipped. [3]
But the same model that drove explosive growth also amplifies risk. CoreWeave must continuously fund new capacity and refresh hardware—and that often means tapping capital markets. As today’s coverage shows, investors are now less forgiving of leverage and delays, particularly after a steep drawdown that has made CoreWeave a symbol of the “AI trade” cooling. [4]
What’s new on December 16: Texas data center delays add fuel to the selloff
A central development in today’s news flow is a report that rainstorms in Texas earlier this year delayed construction of a major AI data center complex, pushing completion back by months. The project is described as important for CoreWeave and tied to workloads for OpenAI, and it reportedly faced not only weather issues but also design revisions. [5]
This kind of delay matters more than it might in a typical enterprise IT story. In AI infrastructure, timing is the product: customers are racing to deploy new models, and suppliers are racing to bring new megawatts online. When schedules slip, revenue recognition can slip too—exactly the kind of execution risk that markets tend to punish when a company is carrying heavy debt and spending aggressively. [6]
The debt question: growth at any cost meets the reality of capital markets
CoreWeave has leaned heavily on financing to scale. In early December, the company announced a $2 billion convertible senior notes offering that was later upsized to $2.25 billion, with the notes due in 2031 and carrying a 1.75% interest rate. The company disclosed an initial conversion price around $107.80 per share, along with other details such as capped call transactions designed to reduce dilution. [7]
To bulls, that low coupon can look like smart financial engineering—especially compared with older, much higher-cost debt. To bears, the bigger picture is the same: CoreWeave’s model requires huge upfront outlays, and the market is increasingly asking when those outlays translate into durable cash generation.
Recent investor commentary has pointed to the magnitude of spending relative to revenue. In its own analysis of CoreWeave’s financial trajectory, The Motley Fool highlighted that CoreWeave spent $1.9 billion in capex in Q3, following even higher capex in Q2, while Q3 revenue was about $1.4 billion—a mismatch that helps explain why profitability remains a looming issue even amid surging demand. [8]
The same analysis emphasized that the company has burned more than $8 billion in free cash flow over the past 12 months, underscoring why investors keep returning to one essential question: how long can the infrastructure buildout be financed before the market demands a clearer path to self-funding operations? [9]
Backlog is the bullish anchor—if CoreWeave can execute
Despite the selloff, one metric continues to underpin the bullish narrative: CoreWeave’s revenue backlog. Commentary circulating this week points to a backlog of $55.6 billion, with 40% expected over the next 24 months—suggesting substantial contracted demand that could turn today’s spend into tomorrow’s revenue. [10]
Optimists argue that this is precisely why CoreWeave exists: hyperscalers and AI labs need capacity now, and specialists can move faster than traditional cloud stacks. In a separate Motley Fool piece from December 12 (also widely syndicated), CoreWeave was framed as a rare “AI-specific cloud” player with triple-digit growth—paired with explicit acknowledgement that its heavy debt load makes it higher risk than the megacaps. [11]
“Depreciation trap”: Jim Chanos’ warning hits a nerve across AI infrastructure
A major driver of today’s anxiety isn’t just debt—it’s the accounting and economics behind the hardware itself.
In a December 16 report, famed short seller Jim Chanos argued that AI infrastructure operators may be using depreciation schedules that don’t reflect how quickly GPUs become obsolete. The critique: if companies depreciate GPUs over something like six years but competitive reality forces replacement in three to four, then “true” annual costs may be much higher than reported—pressuring profitability and potentially exposing a fragile financing stack. [12]
Chanos’ argument is simple and brutal. As quoted in that report: “If the chips last for three years, you have to depreciate a third of what you spend.” [13]
The piece also noted Chanos’ claim that rental rates for certain Nvidia “Hopper” chips have fallen materially year over year, illustrating how fast pricing power can erode when newer GPUs arrive. Even if one disputes the exact timeline, the broader point resonates: AI infrastructure is a hardware business wearing a software valuation—and hardware ages fast. [14]
Not just CoreWeave: Oracle and Broadcom show how quickly sentiment can flip
CoreWeave’s volatility is unfolding inside a wider reassessment of AI infrastructure spending.
Oracle has been at the center of this debate after its disclosures and guidance heightened investor concerns about how data center growth will be financed. MarketWatch reported that Oracle’s earnings update intensified worries about needing more debt for buildouts, and highlighted scrutiny around a $248 billion long-term lease commitment disclosed in a 10-Q, plus investor anxiety tied to reports about potential data center delays for OpenAI workloads (which Oracle denied). [15]
Investor’s Business Daily, in coverage dated December 16, similarly emphasized fears around Oracle’s aggressive infrastructure investment, pointing to $248 billion in new lease commitments and a sharp post-earnings slide as investors digested the scale of the buildout. [16]
Meanwhile, Broadcom’s recent decline has also helped reset expectations for what “AI growth” looks like once margins and product mix enter the conversation. MarketWatch noted Broadcom’s stock had extended a steep drop and described it as the company’s worst three-day slide since 2020, reinforcing how quickly the “AI premium” can compress when outlook details disappoint. [17]
Bridgewater’s “dangerous phase” warning: the AI boom leans harder on outside capital
Beyond individual names, the macro story is about capital.
Reuters reported that Bridgewater co-CIO Greg Jensen warned the AI spending boom is entering a “dangerous” phase as major firms increasingly turn to external investors to fund massive AI ambitions, with Jensen writing there’s a “reasonable probability” markets may soon find themselves in a bubble. [18]
The Reuters report cited a UBS estimate that AI data center and project financing deals surged to $125 billion through November 2025, up from $15 billion over the same period in 2024—an acceleration that helps explain why markets are suddenly more sensitive to delays, margin pressure, and any sign that monetization timelines are slipping. [19]
Dip buyers step in: Cathie Wood adds CoreWeave as the debate intensifies
Not everyone is backing away.
Barron’s reported that Cathie Wood’s ARK Invest boosted exposure to crypto-related stocks during the latest downturn—and also bought CoreWeave, describing it as a former crypto-mining company now focused on data centers, with the purchase reported at roughly $9.9 million worth of shares. [20]
This matters because it highlights the market’s split personality: one camp is de-risking AI infrastructure due to financing and execution concerns, while another sees the drawdown as a chance to buy into long-term AI demand at a lower valuation—especially for “picks-and-shovels” providers. [21]
Nvidia’s backstop agreement adds nuance to the “demand risk” debate
One detail that complicates the bear case is CoreWeave’s relationship with Nvidia.
A CoreWeave filing disclosed that under a September 2025 order form with an initial value of $6.3 billion, Nvidia is obligated to purchase residual unsold cloud computing capacity through April 13, 2032, subject to conditions. Reuters also reported the arrangement as a demand cushion for CoreWeave if customers don’t fully utilize the capacity. [22]
That kind of structure can reduce the probability of a sudden revenue cliff from underutilization. But it doesn’t remove the core debate: whether the industry will ultimately earn attractive returns after accounting for rapid hardware replacement cycles, depreciation, and the cost of capital. [23]
What happens next: the key milestones investors are watching
With the AI infrastructure sector under a microscope, the next phase for CoreWeave (and peers) is likely to be judged on execution and cash discipline more than growth headlines.
Here are the near-term signposts the market is likely to focus on:
- Data center schedule clarity: any concrete updates that reduce uncertainty around the Texas buildout timeline tied to OpenAI workloads. [24]
- Conversion of backlog into revenue: proof that backlog translates into delivered capacity and recurring cash generation—without margins collapsing. [25]
- Capex trajectory vs. revenue growth: whether spending begins to normalize relative to revenue, or whether the “build forever” dynamic persists due to rapid GPU obsolescence. [26]
- Debt and refinancing strategy: how CoreWeave manages a growing stack of obligations—including convertible notes and legacy higher-rate debt—and how credit markets price that risk. [27]
- Industry read-through from Oracle and Broadcom: continued volatility in AI infrastructure bellwethers can amplify (or ease) the pressure on smaller, more leveraged players. [28]
The bigger picture: AI is real—now investors want proof the returns are, too
Today’s CoreWeave headlines are less about whether AI demand exists and more about whether the infrastructure layer can be financed and operated profitably at scale. The combination of construction delays, debt scrutiny, and GPU depreciation concerns is forcing a market-wide reset: execution and unit economics are back in charge.
For CoreWeave, the question isn’t whether AI needs compute—it does. The question is whether a debt-funded race to deploy the newest GPUs can deliver enduring profits before capital markets, competition, and hardware aging turn the AI boom into an accounting and financing stress test. [29]
References
1. www.wsj.com, 2. www.marketwatch.com, 3. www.fool.com, 4. www.wsj.com, 5. seekingalpha.com, 6. www.wsj.com, 7. investors.coreweave.com, 8. www.fool.com, 9. www.fool.com, 10. www.fool.com, 11. www.fool.com, 12. www.benzinga.com, 13. www.benzinga.com, 14. www.benzinga.com, 15. www.marketwatch.com, 16. www.investors.com, 17. www.marketwatch.com, 18. www.reuters.com, 19. www.reuters.com, 20. www.barrons.com, 21. www.wsj.com, 22. www.sec.gov, 23. www.benzinga.com, 24. seekingalpha.com, 25. www.fool.com, 26. www.fool.com, 27. investors.coreweave.com, 28. www.marketwatch.com, 29. www.wsj.com
