US-China AI race: one goal, two very different approaches By Investing.com

BofA forecasts AI-related capex will add roughly 0.4 percentage points to US GDP growth this year, underpinning a near-term investment cycle centered on data centers and specialized silicon. Key exporters—Taiwan (analysts keep an 8% GDP forecast for 2026), Mexico and Korea—are highlighted as primary beneficiaries as global supply chains redirect capital. The note flags an ongoing US–China race (innovation vs. state-led scaling) and warns AI's productivity upside will hinge on adoption and a significant 'skills challenge', with upside risks concentrated in near-term hyperscaler spending and potential fading contribution in 2027.

Analysis

The immediate alpha sits in the capital-goods nodes that are hardest to scale quickly: extreme-UV lithography, test & inspection, and specialty substrates. Bottlenecks here transmit non-linearly — a 6–9 month semiconductor equipment backlog can force hyperscalers to stagger deployments, converting a single-year capex bump into a multi-year revenue stream for suppliers and OSATs. Expect revenue cadence and margins to bifurcate between firms that can convert lead times into pricing power and those that compete on throughput.

Geopolitics and raw-materials form the key second-order trade-offs. Any incremental tightening of controls on advanced-node exports or a disruption in shipping routes will accelerate onshoring into Mexico/Korea/Taiwan corridors, advantaging logistics, local assembly, and power/energy suppliers while compressing margins for long-distance OEMs. Conversely, falling energy or copper prices would compress hyperscaler opex-benefit math and slow refresh cycles, flipping returns across the chain over 6–18 months.

Idiosyncratic tail risks are concentrated in demand sequencing and policy: a 30–50% pullback in hyperscaler AI budgets or broad new export restrictions could erase equipment visibility inside a single quarter; alternatively, multi-year commitments from top cloud providers extend the cycle into 2027–28. Monitor order books, wafer starts, and hyperscaler booking disclosures as high-frequency signals of cycle persistence.

Contrarian read: markets are underpricing the duration of the equipment cycle. Hyperscalers’ multi-year model refresh and redundancy plans mean incremental orders are likely to be lumpy but persist beyond 2026, creating asymmetric upside for capital-equipment, substrate, and commodities exposure versus the headline AI-apparel winners that depend on faster software productivity wins.

Key Decisions for Investors

• Buy ASML (ASML) and Lam Research (LRCX) as primary equipment exposure — target a 12–36 month hold. Position size 2–4% each; use 10–15% stop if order-book momentum reverses. Reward: double-digit EPS re-rating if backlog converts; Risk: export-control or technology-substitution event that delays orders.
• Long NVDA via 2027 LEAP call spread (buy Jan-2027 $450 call / sell Jan-2027 $600 call) to capture continued data-center GPU scarcity with defined downside — allocate 1–2% of portfolio. Expect 2–3x payoff if model training demand persists; downside limited to premium if AI budgets roll off.
• Pair trade: long KLA/LRCX (equipment/test) and short Intel (INTC) to express structural capex tailwind vs legacy logic-cycle commoditization. Rebalance quarterly; close if semiconductor capex indicators decline by >20% YoY.
• Commodity hedge — buy Freeport-McMoRan (FCX) or 12-month copper futures to play rising copper and specialty metals demand from data centers and electrification. Timeframe 6–24 months; hedge with 20–30% put protection to guard against demand-shock scenarios.