vnitkar

Everyone that owns neoclouds such as $WULF $IREN $NBIS $CIFR $APLD should read this: The market is CONFUSING a $CRVW problem with an AI demand problem. AI infrastructure stocks are getting CRUSHED, but the evidence doesn't point to collapsing demand. It points to a financing problem at CoreWeave. And the market is treating very different businesses as if they're all the same. That's the part I try to pay attention to. > GPU rental prices are rising. > Hyperscalers are still spending hundreds of billions on AI infrastructure. > The biggest AI labs are scaling rapidly, and access to power remains one of the biggest bottlenecks in the entire buildout. None of that means every company in the sector is a good investment. Far from it. But there's a big difference between a demand problem and a financing problem. Right now, the market seems to be mixing the two together. Demand for compute is still strong So, let's start with what's actually happening in the market: @matthew_sigel pointed out that Blackwell Ultra (B300) GPU rental prices had risen for 17 consecutive days, reaching a new all-time high based on verified transactions from U.S. neoclouds. That's a pretty important signal. High-end compute isn't suddenly flooding the market. It's still scarce. teapaw.org/matthew_sigel/status/2… Then there's the spending: @unusual_whales highlighted that $META, $AMZN, $MSFT and $GOOG are expected to spend around $750 billion on capex this year, mostly tied to AI infrastructure. That's more than 80% above last year's roughly $410 billion. Think about the scale of that investment. teapaw.org/unusual_whales/status/… The AI labs are scaling, too. Anthropic's reported revenue figures are substantial, OpenAI continues to operate at a massive scale, and demand for cloud compute remains strong. Even older GPUs like the H100 and A100 continue to command meaningful rental prices. None of this guarantees that today's spending levels will continue indefinitely. Capex can slow, pricing can change, and new capacity will eventually come online. But the evidence doesn't point to an industry where customers have suddenly stopped wanting compute. @Kalshi also highlighted Anthropic's claim that AI could push U.S. GDP toward $44 trillion by 2030. Whether that exact forecast proves correct is another question. The important point is that the labs are still building toward a future that requires ENORMOUS amounts of computing power. teapaw.org/kalshi/status/21114018… @saxena_puru made a point worth keeping in mind. Hotter inflation brought rate concerns back into focus, and investors started looking much harder at the cost of financing AI infrastructure. But the demand picture is different from the financing picture. GPU rental prices remain elevated. The market is repricing the cost and risk of building out capacity, not necessarily the need for that capacity in the first place. teapaw.org/saxena_puru/status/211… CoreWeave is where the real problem sits. And this is where things get more complicated. CoreWeave isn't just building infrastructure and collecting rent. Its model depends heavily on financing expensive GPU capacity, locking in customer contracts, and earning enough on those contracts to cover its capital costs. That creates a very different risk profile when interest rates and credit spreads move against you. The recent All-In Podcast discussion with Ben Horowitz, Chamath Palihapitiya and David Sacks highlighted the financing risks at CoreWeave. I think the distinction they raised is important: strong demand for AI compute doesn't automatically translate into attractive returns for every company financing the infrastructure. Here's how I see the situation: The figures discussed in the recent All-In Podcast episode paint a pretty uncomfortable picture: Debt and obligations: Roughly $35 billion to $50 billion, depending on what is included. Additional financing: Around $35 billion potentially needed to fund its existing commitments. Borrowing costs: Previously around 8.5%, with some bond yields reportedly climbing toward 15%. Customer concentration: Roughly 75% of revenue reportedly comes from a relatively small group of customers. Contract structure: Fixed-price agreements can leave CoreWeave earning less than it could in a rising spot market, while its financing costs keep moving. These figures need to be read in context. Debt, leases and future funding requirements aren't interchangeable, and bond yields or credit default swap spreads aren't direct predictions of what will happen. But the underlying issue is clear: CoreWeave needs substantial capital to deliver on commitments it has already made, and that capital is becoming more expensive. You can have strong demand and still have a bad business model if the economics don't work. That's the distinction investors need to understand. (You can watch the whole episode here, I highly recommend it. youtu.be/_b3LfMflIGs) Chamath laid out the financing risk during the discussion. If a business relies on borrowing heavily and earning a spread between its financing costs and what customers pay, a sudden increase in credit spreads can wreck the economics. Ben Horowitz drew a comparison with his experience at LoudCloud in the early 2000s. The company committed to expensive bandwidth when prices were high, only to see those prices collapse later. Leveraged businesses can get trapped when the economics change faster than their contracts. Today's AI infrastructure market has different demand drivers and physical constraints. But the lesson about locking in capital commitments is worth taking seriously. David Sacks pushed back on the broader pessimism. His argument was that investors are dealing with an industry targeting enormous future outcomes, where short-term changes in financing conditions can trigger outsized market reactions. He also made an important distinction between different financing structures. Asset-backed financing tied to GPUs and investment-grade customers isn't the same as unsecured corporate debt. And a business with shorter payback periods or more flexible contracts has a different risk profile from one locked into longer commitments. Both sides can be right here. AI compute demand can continue growing rapidly while CoreWeave's financing structure creates real problems for shareholders. That's exactly why I try to look beyond the headline. And this is where I think the current selloff gets interesting. Not every company in the sector owns GPUs, takes the same technology risk, or relies on the same financing model. Take $WULF as a example: CEO @PaulBPrager has repeatedly stressed the difference between companies like CoreWeave and businesses focused on providing powered data center infrastructure. TeraWulf's model centres on developing and leasing infrastructure, with long-term agreements and credit support associated with counterparties such as Google. The company is positioning itself around powered real estate rather than taking exactly the same GPU ownership and technology risk as a pure neocloud. That doesn't make it risk-free. Construction costs, financing, delivery timelines, customer concentration and the conversion of contracted revenue into actual cash flow all matter. But the risks are different. teapaw.org/paulbprager/status/211… @anni_sen has been making a similar argument. The market has been punishing a broad group of companies for concerns that are particularly acute at CoreWeave. But owning land, securing power and delivering data center capacity isn't the same business as buying expensive GPUs, financing them with debt and taking the risk that those assets become less competitive over time. Consider the differences: CoreWeave: GPU ownership, financing costs, customer concentration and fixed-contract economics. TeraWulf and Cipher: Infrastructure development, power availability, construction execution and lease economics. Applied Digital: Data center development, customer contracts, financing requirements and delivery execution. Nebius: AI cloud infrastructure, GPU investment, customer demand and the capital required to scale. These companies still have substantial risks. Long-term contracts don't automatically guarantee attractive returns, and secured power doesn't guarantee that a project will be delivered on budget or on time. But their balance sheets, funding requirements, contract structures and exposure to GPU depreciation need to be evaluated individually. teapaw.org/anni_sen/status/211082… teapaw.org/anni_sen/status/211151… That's why I don't think a broad selloff automatically tells us the whole sector is broken. The important question is how that company turns AI demand into shareholder value. Power is still the bottleneck. There's another part of this debate that often gets lost in the noise. AI infrastructure doesn't appear overnight. You need electricity, grid connections, land, permits, data centers, cooling systems and the equipment to run them. Those constraints take time to solve. That doesn't mean every company with access to power will win. Power has to be available where customers need it, infrastructure has to be delivered, and the economics have to justify the capital invested. But it does mean the industry can't simply switch on unlimited supply because GPU demand is strong. The physical buildout remains a major constraint. And that is one reason I wouldn't interpret pressure on CoreWeave's financing model as proof that the broader AI infrastructure buildout is about to unwind. My takeaway: I think investors are dealing with two separate questions. 1/: Is demand for AI compute still growing? The evidence currently points to yes. GPU rental prices remain elevated, hyperscalers continue to invest heavily, and the AI labs are scaling. 2/: Will every company building AI infrastructure turn that demand into attractive returns for shareholders? Absolutely not. CoreWeave's debt, funding requirements and contract economics deserve serious scrutiny. Strong demand doesn't eliminate financing risk, and a company can grow rapidly while still delivering disappointing returns to shareholders. But that doesn't mean every neocloud, data center developer or power-focused infrastructure company has the same problem. For me, the job is to separate the companies with fragile financing structures from those with more sustainable economics, credible counterparties and a realistic path to generating cash flow. The selloff may create opportunities. It may also expose businesses that were never as attractive as their growth rates suggested. The key is to distinguish between the two. A financing problem at one company is not, by itself, evidence that the AI demand thesis is broken. -BP Please note: This is my personal take and not financial advice.