Back to News
Market Impact: 0.35

Raymond James: AI has turned a cyclical sector into a ‘secular boom’

NVDAAVGOAMDARMALABINTC
Artificial IntelligenceTechnology & InnovationTrade Policy & Supply ChainAnalyst InsightsAnalyst EstimatesCompany FundamentalsCorporate Guidance & Outlook
Raymond James: AI has turned a cyclical sector into a ‘secular boom’

Raymond James resumed coverage of seven semiconductor names—NVIDIA (Strong Buy), Marvell (Strong Buy), Broadcom (Outperform), AMD (Outperform), ARM (Market Perform), Astera Labs (Market Perform) and Intel (Market Perform)—saying generative AI has transformed a cyclical sector into a secular boom and that enterprise adoption and AI factory buildouts are accelerating. The note flags power-supply and packaging bottlenecks, highlights innovations like co‑packaged optics and CoWoS, models peak Blackwell GPU orders at 7.8 million units in fiscal 2027, cites Marvell targeting roughly 20% data‑center ASIC share by 2028, and cautions that Intel's loss-making foundry business remains an albatross on execution.

Analysis

Market structure: The primary winners are NVDA, AVGO and AMD — firms with software/hardware stacks or differentiated optics/ASIC IP — because AI factory scale favors incumbents and bespoke silicon; Raymond James’ 7.8M Blackwell-unit peak by FY2027 implies multi‑year GPU demand that will sustain ASPs and backlog, tightening supply for high-end wafers and packaging over 2024–2027. Losers include INTEL (foundry losses, execution risk) and smaller interconnect players without scale; pricing power will concentrate at vertically integrated suppliers and custom-ASIC providers, compressing margins for commodity GPU makers. Risk assessment: Tail risks include renewed export controls (China) that can wipe out >10–20% of TAM for certain suppliers, a macro recession that defers data‑center capex by 6–18 months, or a packaging bottleneck/utility constraint that stalls shipments for quarters. Immediate (days) risks are sentiment/earnings volatility; short term (weeks–months) is order cadence and supply shocks; long term (years) is competitive GPU parity and foundry capacity. Hidden dependencies: heavy reliance on TSMC/advanced CoWoS packaging and local power availability; catalysts are Blackwell shipments, AMD competitive GPU launches, TSMC capacity guidance and export-policy announcements. Trade implications: Tactical core long NVDA exposure with 9–18 month horizon makes sense (see decisions) plus selective AVGO/AMD overweight to capture optics/merchant GPU upside; underweight/hedge INTEL via put spreads given foundry drag. Use pair trades (long NVDA vs short INTC) to express structural share shift while neutralizing beta. Options: buy 9–12 month LEAP calls on NVDA (delta ~0.35–0.45) and buy 3–6 month put spreads on INTC to limit cost. Contrarian angles: Consensus underestimates execution/dependency risk (packaging, power, TSMC capacity) and may be overpricing perpetual upward revisions — history shows cyclical booms (GPU crypto cycle 2017–18) can reverse quickly when ASPs normalize. Overdone trades: aggressive NVDA leverage >5% position risks a sharp IV reset if Blackwell shipments slip; conversely, INTEL downside may be limited if it successfully monetizes foundry customers or secures major government contracts. Watch for >15% QoQ booking changes as a trigger to re‑rate positions.