Art of Speculation
Chinese-Language AI Fundamentals & Macro, Concentrated Investing
CoreWeave's demand is not the problem; execution and financing are. Q2 revenue rose 112% to $2.58B, slightly above estimates, with $1.51B of adjusted EBITDA at a 59% margin. Backlog reached $104.2B, up 246%, before $25B+ of new commitments early in Q3. Only 1.5GW of its roughly 3.7GW of contracted power is active, so how fast capacity comes online and turns into revenue matters most, while $9.4B of capex and heavy losses remain the drawbacks. x.com
CoreWeave is the first cloud provider to deploy Nvidia's Rubin NVL72 platform, an early lead that could lift its AI cloud revenue growth, pricing power and market share. The deal also validates its ties with Nvidia and Dell. Unusually large buying of $117 calls looked like institutions entering, and the chart broke its downtrend line after holding its 200-day averages. As a swing trade over the next month, the targets are 128, then 145; a gap-up on news like this tends to run for several days. x.com x.com x.com
CoreWeave's pricing power is overlooked amid financing fears, making it potentially the most misunderstood AI stock right now. It had over $104B in Q2 backlog and signed $25B in net new commitments over six weeks. Short-term contracts approach $40M per MW annually versus about $7M on active capacity, but that rate cannot be applied fleetwide. A $4.2B convertible carried a 2.875% coupon despite near-15% unsecured bond yields; $35B debt, dilution and capacity-conversion risks remain. x.com
CoreWeave's stock may already reflect much of its real financing risk amid extreme pessimism. Its unsecured bonds yield distressed rates and CDS spreads have surged, while Q2 net interest expense was $640M and 2026 capex guidance is $35–39B. Q2 revenue was $2.58B with about $104B of backlog, but whether that demand yields shareholder value after financing, depreciation and reinvestment remains uncertain; attractive financing is central to the case. x.com
CoreWeave's risk is buildout speed: each new build needs capital before the last pays for itself. Its analysis shows $35.6B of loans lasting about nine months beyond customer contracts; $14.6B is due in 2031-32 as final contract cash arrives. IREN's Microsoft project loan matches its contract, but $6.3B of convertible bonds mature in 2031-33, mostly after contracted cash ends; low shares mean cash repayment. Slower AI demand growth could let both earn back projects before funding more. x.com
Rising GPU rental rates and Anthropic's compute margins challenge claims of collapsing demand for CoreWeave's capacity. The WSJ reported Anthropic's first-half revenue of $16.3B against $9B in compute and infrastructure spending; Bloomberg reported its Q3 compute margin above 40%. Long contracts and fully amortizing loans cushion demand declines but cap upside. Despite the challenging neocloud model, CoreWeave's risk/reward is by far the best and its quality well ahead of peers. x.com x.com … Show more
CoreWeave's signed backlog of $99.4 billion is more than double its $45 billion market cap, yet the market trades it on headlines and competition fears rather than contracts. 36% of that backlog converts to revenue within two years, backed by 3.5GW of contracted power. Nebius is similar, with about $50 billion of backlog against a $52 billion market cap. The biggest companies are paying tens of billions because neoclouds cannot build fast enough. x.com
CoreWeave's heavy debt and negative free cash flow heighten the risk that neocloud contract growth will not fund expansion. It has $104.2B in backlog against $35.6B in debt and negative $5.74B quarterly free cash flow; Nebius has over $40B in contracts but $5.7B quarterly capex and customer-concentration risk. DigitalOcean's diversified cloud model reduces reliance on debt-funded GPU expansion, while TeraWulf avoids direct GPU depreciation but faces capital-intensive construction. x.com
CoreWeave's growth story is shifting from proving demand exists to proving infrastructure delivery can keep pace with customer commitments. Guidance calls for $12-13B of 2026 revenue, over 1.7 GW of year-end active power and $18-19B of annualized exit revenue, needing about 0.7 GW of net additions from a 2.5 GW contracted but inactive pool. Each megawatt needs about $34M of investment, 82% of it technology equipment, bringing heavy depreciation, ongoing financing and short replacement cycles. … Show more
CoreWeave's growing backlog, active power and GPU pricing make its declining shares look cheaper, though debt and customer cancellations remain risks. Blossom reports backlog rose from $99B to $129.2B, active power from 1GW to 1.5GW and GPU-offering prices by 25%. Its debt burden remains a valid concern. Unusual Whales reports Microsoft pulled out of an Nscale deal, illustrating the risk of hyperscalers leaving neocloud contracts. x.com x.com x.com
CoreWeave's growing revenue and large backlog have not translated into positive fundamentals. Its trailing free cash flow has fallen to -$13.65 billion, net income to -$1.93 billion and EPS to -$3.66; the stock may still work out, but the uncertainty warrants caution. x.com
CoreWeave's strong short-term liquidity and half its backlog set to convert by 2028 mean it won't go bust. At just 1.3x 2027 sales and probably 1x exit ARR, the only way the stock could be lower in 12 months is bankruptcy. x.com
