2 Beaten-Down AI Stocks I Can’t Stop Buying

AI infrastructure winners are being selectively repriced: Oracle reported an all-time-high contracted backlog of $638B tied to an OpenAI-led $750B/through-2030 computing-spend commitment, but the stock is pressured by customer-delivery skepticism and a surge in fiscal 2026 capex to $55.7B (from $21.2B), funded by ~$43B new debt plus ~$5B stock issuance. IBM took a 25% stock haircut after a pre-announced Q2 earnings miss, with management citing issues including a global memory shortage, yet the article highlights ongoing growth in software/consulting, a rapidly expanding AI book of business, and quantum milestone progress. Overall, the news frames the pullbacks as temporary uncertainty rather than broken fundamentals, with both names also positioned as dividend payers (Oracle ~1.4% yield; IBM ~3%).

Analysis

The cleaner expression here is not “buy the beaten-down AI names,” but “own the financing beneficiaries and be careful with names that are underwriting the capex cycle.” Oracle’s equity now behaves like a levered claim on AI infrastructure demand: if contracted demand converts on time, operating leverage is meaningful; if there is even a modest timing slip, the market will re-rate the stock on balance-sheet risk before revenue catches up. That makes the next 1-3 quarters about conversion rates, not backlog headlines.

IBM looks more like a budget-priority trade than an AI compounding story. In a slow-growth IT spend environment, services and consulting are the first lines to get de-scoped when clients need memory, servers, or GPUs, so IBM can remain fundamentally fine while still underperforming on multiple compression. The second-order beneficiary is hardware and infrastructure: NVDA, AVGO, ANET, and potentially MU get a larger share of incremental wallet share as customers push spend toward constrained compute inputs.

Contrarian view: consensus is probably too willing to treat all AI exposure as one trade. ORCL has real upside if execution is clean, but its risk is more about funding/dilution and customer concentration than business demand; IBM’s issue may be less temporary than advertised if AI projects fail to offset legacy softness. Falsifier for the bullish read is simple: if ORCL’s capex rises again without a corresponding step-up in realized revenue, or if IBM’s bookings fail to re-accelerate over the next two quarters, the market will stop rewarding the “temporary headwind” narrative.

Key Decisions for Investors

• Prefer long NVDA over ORCL for AI exposure over the next 3-6 months; NVDA captures the hardware pull-through with less balance-sheet risk. Use ORCL only if you believe revenue conversion is imminent, and size it smaller than NVDA on a risk-adjusted basis.
• Pair trade: long NVDA / short IBM into the next earnings cycle. Thesis is that incremental AI spend migrates to compute vendors while IBM remains exposed to clients deferring services; target 8-12% relative outperformance over 1-2 quarters if IT budgets stay tight.
• Watch MU and ANET as secondary beneficiaries of the capex cycle; if memory pricing or networking demand inflects higher, these names can outperform the more crowded AI software trades over 1-3 months.
• For ORCL, wait for a better entry after the next financing or guidance update. If the market rewards backlog without visible revenue conversion, fade rallies rather than chase; the risk/reward improves only after cash flow catches up.
• Set a stop/falsifier on the bullish AI-infrastructure view: if ORCL adds leverage again without revenue acceleration, or if IBM’s next two quarters show no bookings recovery, reduce exposure to both and rotate into less balance-sheet-intensive AI winners.