甲骨文(Oracle)在业绩超预期后飙升。买入该股票的3个理由（以及1个避免的理由）。

"BYOH upfront contracts obscure whether Oracle is actually winning margin-accretive deals or just accelerating cash while deferring capex risk to customers."

Google and OpenAI both flag margin compression risk, but neither quantifies the capex-to-FCF bridge. Oracle's $29B upfront BYOH deals look like cash acceleration theater—they front-load revenue recognition while deferring hardware capex to customers. That's FCF-positive but masks whether AI margins actually improve or just shift timing. If RPO converts at lower-than-guided gross margins due to competitive pressure, the $90B target becomes a revenue mirage with deteriorating profitability. That's the real trap, not just concentration.

"Oracle's RPO faces a 'stranded asset' risk if regional energy constraints and regulatory intervention block the physical deployment of promised AI capacity."

Anthropic is right to highlight the 'cash acceleration theater,' but both you and Grok are ignoring the regulatory risk to the RPO. If the DOJ or FTC restricts AI data center expansion due to energy grid strain, that $553B backlog isn't just a revenue mirage—it becomes a legal liability. Oracle is selling capacity that may be physically impossible to power in key regions. The real risk isn't just margin compression; it is a forced, systemic 'stranded asset' scenario.

"Counterparty credit/default risk could turn Oracle's RPO into uncollectible or impaired revenue, a material and under-discussed danger."

Google flags grid/regulatory risk — valid — but the nearer-term, underappreciated threat is counterparty credit risk inside that $553B RPO. These multi‑year, lumpy AI contracts hinge on a few high‑beta customers; if valuations tumble, customers can renegotiate, default, or enter bankruptcy, converting ‘backlog’ into contested claims and impairments. Oracle’s disclosures assume collectability; investors should stress‑test recovery rates, termination clauses, and bad‑debt scenarios.

"AI customers' overcapitalization and contract penalties minimize RPO credit risk."

OpenAI's counterparty credit risk is overstated—these AI hyperscalers like OpenAI boast $150B+ valuations, infinite funding runways from VCs/Saudis, and contracts with steep termination fees exceeding hardware costs. The $29B BYOH upfronts already pocket cash, de-risking 5%+ of RPO immediately. Default would trigger penalties, not impairments, making this backlog stickier than feared.