CoreWeave

Your GPUs depreciate over 6 years but a new generation comes out every 18–24 months. How is that not an accounting fiction?

CoreWeave extended computing equipment useful life from 5 to 6 years effective January 1, 2023. The company's 10-K says this reflects "continuous advancements in hardware performance, software optimization, and data center design improvements." In FY2023, the change reduced expenses by $20M. By FY2025 with $30.6B net PP&E, a 1-year useful-life extension is worth hundreds of millions in annual depreciation savings. The GPU generations accelerating simultaneously creates a direct contradiction between accounting life (6 years) and economic life (3 years max for peak performance). (CoreWeave FY2025 10-K, Note 1 — PRIMARY)

You have $25 billion in debt. Your revenue run-rate is $8 billion. How do you service this without a continuous equity dilution cycle?

Interest expense was $536M in Q1 2026 alone — annualized $2.1B. Revenue $2,078M in Q1 2026. The company raised $1,985M in a private placement in Q1 2026 (issuing 23M new shares). Revenue growth is +112% YoY, so the hope is that $8B annualized revenue grows to $15B+ by FY2027 as OpenAI/Meta contracts ramp. But the debt also keeps growing: $8B (Dec 2024) → $21.6B (Dec 2025) → $25.1B (Mar 2026). The dilution is constant. Net loss was $740M in Q1 2026 alone. (CoreWeave Q1 2026 10-Q — PRIMARY)

Microsoft is 67% of your revenue. If Azure builds its own GPU clusters, you're done. What's your plan?

Microsoft's 67% share (FY2025) has declined to ~45% of Q1 2026 revenue as OpenAI and Meta ramp. The $60.7B RPO includes new long-term contracts (OpenAI $6.5B through 2031, Meta $14.2B through 2031) that diversify away from single-customer dependence. But top 2 customers still = 65% of Q1 2026 revenue (Customer A 45%, Customer B 20%), and 98% of revenue sits under committed contracts (a contracted-vs-on-demand figure, not a concentration figure). Microsoft IS building Azure GPU infrastructure — the question is whether CoreWeave's specialized dense-GPU architecture remains superior for model training even as Azure scales. (CoreWeave Q1 2026 10-Q; FY2025 10-K — PRIMARY)

Nvidia is your supplier, your shareholder, and your backstop customer. Isn't that the definition of a conflict of interest in any restructuring?

Yes. Nvidia is: (1) GPU supplier — CoreWeave is almost entirely Nvidia-dependent; (2) >5% beneficial owner at IPO (S-1 — PRIMARY); (3) customer — paid CoreWeave ~$320M through 2024 (REPORTED / NOT PRIMARY-CONFIRMED — not surfaced verbatim in the 424B4; confirm vs. the S-1); (4) capacity backstop — obligated to purchase unsold capacity through 2032, ~$6.3B initially (REPORTED / NOT PRIMARY-CONFIRMED — not surfaced verbatim in the 424B4; confirm vs. the S-1). In a distressed scenario, Nvidia's interests as creditor (wanting its GPU assets back), supplier (wanting future GPU orders), and equity holder (wanting share value preserved) directly conflict. This is undisclosed in a single risk factor in the S-1 but the systemic conflict is exactly what Burry's thesis targets.

What happens to the GPU collateral if Blackwell makes H100s economically worthless in three years?

NOT SOURCED from filings. The GPU LTV (loan-to-value) covenants in the DDTL facilities are referenced in the risk factors ("if we are unable to maintain certain financial ratios and other conditions") but the specific LTV triggers are NOT disclosed in the public filings reviewed here. If H100 GPUs lose >50% of economic value by 2027 (plausible given Blackwell ramp), the collateral backing the DDTL 1.0 and 2.0 facilities ($5.9B as of Mar 2026) may trigger LTV covenant violations — requiring either additional collateral posting or accelerated repayment. (Risk factor reference: CoreWeave FY2025 10-K; specific LTV thresholds NOT SOURCED.)