Should You Forget CoreWeave and Buy 3 Artificial Intelligence (AI) Stocks Right Now?

Key datapoint: Sandisk's data-center revenue was $440M in Q2 FY2026, up 76% YoY (company revenue $3.02B, +61% YoY) and net income $803M, +672% YoY, highlighting strong profitable AI-related demand. Palantir reported Q4 revenue of $1.4B (+70% YoY) and net income $608M (+43% YoY) but faces valuation/PR risks; CoreWeave remains unprofitable (net loss $1.16B in 2025, including $452M in Q4). Oracle posted cloud revenue $7.97B (+34% YoY) and net income $6.13B (+95% YoY), has a potential up-to-$300B OpenAI deal, but has reduced free cash flow by >$24B over three years and taken on >$100B of debt to fund data centers.

Analysis

The AI-driven data-center build-out is creating asymmetric winners: component suppliers with captive fabs and long-term ASP control (storage controller/IP owners, select NAND fabs, and GPU monopolists) will capture most incremental margin, while mid-tier system integrators face margin compression as hyperscalers push for custom stacks. Expect pricing to be lumpy — NAND capacity additions and GPU replenishment cycles create 6–12 month swings that can amplify earnings volatility for vendors without differentiated roadmap control.

On the demand side, large AI customers reallocate capital from general cloud to specialized infra, which benefits suppliers aligned with hyperscaler specs but raises customer-concentration risk for vendors dependent on a handful of large contracts. Balance-sheet dynamics matter: heavy capex plans raise near-term free-cash-flow stress and increase sensitivity to a 1–2 quarter slowdown, which can flip sentiment quickly even if medium-term demand remains intact.

From a sector-rotational angle, profitable storage franchises will attract risk-averse flows and act as de facto defensive AI plays, while outsized multiples on analytics/AI software create convex downside to any growth miss; reputational or government-contract noise can turn a high-PE story into a rapid mark-down. Liquidity and option markets are pricing asymmetric risk: implied vols are elevated for large-cap AI names, offering cheap hedges for concentrated exposure and attractive IRR for dispersion trades.

Contrarian read: the market may be underestimating the timing risk of monetizing new data-center capacity and overestimating sticky gross margins for third-party suppliers once cloud providers internalize more hardware/software layers. Positioning that buys durable cash generative parts of the stack while hedging growth-expectation risk in software/contract-heavy names captures this disconnect without betting on a full market reversal.

Key Decisions for Investors

• Long selective storage OEM (Sandisk/SNDK or WDC) — buy on a sub-10% pullback, target 12–18 month holding period. Risk: NAND oversupply or ASP collapse could wipe 20–30% in a downside scenario; reward: ownership of durable margin capture from proprietary controllers/fab access with 30–70% upside potential if AI infra demand stays on trajectory. Use 12–15% stop-loss and size at 2–4% of equity book.
• Pair trade — long NVDA / short INTC equal notional for 3–9 months. Rationale: capture continued GPU-driven margin expansion vs slower structural recovery at legacy foundry/CPU. Risk/reward: asymmetric — 1x notional each; cap NVDA upside with a 20% covered-call exit plan or scale profit-taking at 40% gain; limit loss to 15% overall on pair.
• Volatility/hedge: buy 9–12 month OTM puts on high-multiple AI/analytics names (e.g., PLTR) sized to cover 30–50% of underlying position exposure. Use these as insurance against a growth-miss or regulatory shock; budget premium as cost of hedging (expect 2–6% of position value).
• Event-driven option spread on Oracle (ORCL) — buy a 12–18 month call spread (debit) to play medium-term rebound while limiting premium at risk. Time entry after the next quarterly release if shares fall on capex/FCF headlines; risk is limited to spread premium, reward is asymmetric if cloud monetization re-accelerates within 6–12 months.