Jensen Huang's AI Capex Pulse Check

Nvidia CEO Jensen Huang projected top-five hyperscaler AI capital expenditure could reach $1.3T next year, implying roughly 60% growth from an estimated $800B this year, supporting bullish demand expectations for AI infrastructure. Marvell raised its growth outlook to 50% from 45%, while CrowdStrike delivered a blowout quarter and expects ARR growth of about 41%, with net-new ARR up a record 51% year over year. The outlook was uneven across sectors: SentinelOne forecast roughly 20% growth and had cut 8% of staff, while Dick's Sporting Goods fell more than 30% in its worst trading day after deteriorating Foot Locker conditions pressured margins and guidance.

Analysis

The actionable read-through is not aggregate AI capex, but the widening dispersion between vendors with contracted design wins and vendors relying on broad enterprise demand. MRVL's custom-silicon/interconnect exposure can gain share from AVGO where hyperscalers seek a second source and negotiate economics; however, warrant-linked customer arrangements make reported backlog less economically equivalent to arm's-length demand. Favor TSM as the cleaner capacity toll-taker: it benefits whether spend lands in NVDA GPUs, AVGO custom ASICs, or MRVL connectivity, while advanced-packaging availability remains the near-term gating variable.

CRWD's growth reacceleration supports a consolidation thesis in security: autonomous attack tooling increases the value of telemetry scale, cross-module data, and a consolidated SOC workflow rather than lifting all endpoint vendors equally. That is structurally negative for S, whose lower scale limits pricing and R&D flexibility, but CRWD's valuation leaves it exposed to an outsized drawdown on any billings/retention miss or another operational incident. Over the next 1-3 months, net-new ARR, Falcon Flex adoption, and gross retention—not headline revenue—determine whether the premium can hold; over 6-18 months, AI security should concentrate share among CRWD, PANW and MSFT.

The more contrarian opportunity is DKS, where the market may be pricing a permanent impairment from Foot Locker rather than a temporary margin reset. The core banner's cash generation gives management time to rationalize stores, inventory and procurement; purchasing and logistics synergies can matter materially once promotions normalize, but this is a 12-24 month repair rather than a next-quarter rebound. APP remains uninvestable pending resolution of regulatory and platform-policy risk: a cheapened multiple does not offset the possibility that distribution or measurement practices are constrained.

Consensus still treats hyperscaler financing capacity as proof of end demand. A sustained period of negative FCF raises the hurdle for incremental projects, particularly at ORCL, whose credit spreads and lease/financing commitments are a more sensitive capex-cycle indicator than NVDA commentary. A widening in ORCL spreads, delayed data-center power interconnects, or reduced advanced-package bookings at TSM would falsify the near-term infrastructure-extension thesis.

Key Decisions for Investors

• Initiate a 3-6 month long TSM / short AVGO pair, sized beta-neutral. TSM captures heterogeneous AI compute demand while AVGO carries higher custom-ASIC concentration and customer-concession risk; reassess if CoWoS capacity additions outpace demand or AVGO discloses material non-hyperscaler AI wins.
• Maintain CRWD as a tactical long only through the next earnings print, preferably funded against S. Target 10-15% relative upside if net-new ARR and retention remain above expectations; exit on material gross-retention deterioration, Falcon Flex deceleration, or a security/availability event.
• Build DKS in tranches over 1-3 months rather than chase a one-day reversal. Underwrite a 12-24 month normalization trade with upside from Foot Locker cost/inventory remediation; stop if core DKS comps turn negative or consolidated gross margin fails to stabilize over two reporting periods.
• Avoid APP until regulatory investigations, app-distribution practices, and platform-policy exposure are independently clarified. Treat any sharp rally before those disclosures as a potential short watch item, not a fundamental recovery signal.
• Set monitoring alerts on ORCL credit spreads, TSM advanced-packaging commentary, and hyperscaler capex guidance. A coordinated deterioration is the trigger to cut AI infrastructure beta, including MRVL and FPS/PSIX, rather than extrapolate supplier guidance.