The Market's Most Feared Month Is Here. These 3 AI Stocks Have Survived It Before.

Oracle shares surged 35.95% in one session (Sept. 10, 2025) despite an earnings miss, driven by remaining performance obligations of $455B (+359% YoY), but the article flags $122B of debt and that $638B of backlog is concentrated among financially strained AI customers such as OpenAI. Broadcom jumped ~20% after reporting fiscal Q3 revenue up 22% to a record $16B, with AI chip revenue up 63% to $5.2B and free cash flow up 47% to $7B, boosted by a new custom chip customer identified as Anthropic. Overall, the piece argues September’s outsized moves are more tied to verifiable contract/customer disclosures and supply-constrained pricing power than to generic AI hype.

Analysis

The market is starting to re-rate AI winners away from narrative and toward contract quality. That favors AVGO more than the broader AI basket because custom silicon has a visible path to revenue conversion and better gross margin durability than merchant GPUs; each incremental design win can matter disproportionately when the customer set is small. By contrast, ORCL’s upside is more fragile because backlog only becomes value if it converts into cash without forcing balance-sheet stress or customer-credit deterioration.

The second-order issue is financing. If hyperscaler and AI-startup capex is already consuming most operating cash flow, then backlog announcements may be less bullish than they look: they can signal demand, but also an ecosystem increasingly dependent on vendor financing, prepayments, or aggressive customer concentration. That is a medium-term risk for ORCL and, to a lesser extent, for AVGO if its new customers are still early in monetization; the market may be capitalizing years of revenue that are not yet de-risked.

MU is a different setup: tighter supply in HBM can support pricing into the next couple of quarters, but memory cycles usually break when supply response catches up or AI accelerator demand normalizes. The contrarian view is that the current enthusiasm may be underestimating the speed at which pricing power can fade if major cloud buyers shift capex timing; if so, MU is the most cyclical way to express a positive AI view and the easiest one to fade on any sign of inventory build or ASP compression. September seasonality matters less than the next earnings confirmations and any revision in forward backlog-to-revenue conversion.

Key Decisions for Investors

• Long AVGO vs short SOXX for the next 1-3 months: better exposure to contract-backed AI monetization and less multiple risk than the broader semiconductor basket; stop if AI revenue guidance is cut or new-customer contributions disappoint.
• Sell ORCL on strength or buy put spreads into earnings: the setup is asymmetric to the downside if backlog quality, customer concentration, or debt servicing pressure is questioned; thesis breaks if management lifts cloud revenue conversion and FCF guidance materially.
• Long MU only on pullbacks after earnings confirmation: best risk/reward is for a 3-6 month trade on HBM pricing, but trim aggressively if channel checks show supply catch-up or if memory ASPs flatten.
• Pair trade: long AVGO / short ORCL to express 'verified demand vs narrative backlog' over the next quarter; risk is a surprise contract disclosure or a balance-sheet-friendly refinancing that reassesses ORCL.