Why I'm Buying Growth Stocks While Everyone Else Is Panic-Selling Tech

Hyperscalers are set to spend about $700 billion on AI infrastructure this year, supporting demand even though chip rental rates are down ~70% from peak. Alphabet reports 7–8 year‑old TPUs at 100% utilization and CoreWeave says 5‑year‑old GPUs remain fully booked, and TSMC has materially increased capex to expand advanced-chip capacity, signaling durable long-term economics. Despite a post‑Iran war rotation into value and energy, the author argues AI infrastructure spending and chip demand remain robust and favors buying growth names such as Alphabet, Amazon, Broadcom, and AMD.

Analysis

Winners will be those that capture protocol-level captures and recurring revenue on the AI stack rather than one-time silicon sales. Hardware vendors that provide networking, interconnect, and IP blocks embedded into hyperscaler custom chips (broadly, networking ASIC/IP providers) stand to convert a fleeting accelerator cycle into multi-year annuity streams because switching costs and validation timelines favor incumbents. Conversely, commodity CPU/GPU sellers without sticky software or integration moats face margin compression as hyperscalers internalize more of the stack.

TSMC’s multi-year capacity build creates a lumpy equipment and construction cadence that amplifies cyclicality: a 12–36 month lead time means demand signals today will only clear into supply years out, creating asymmetric downside if cloud bookings slow. That same cadence gives investors discrete early-warning indicators (foundry booking revisions, capital equipment order flows, hyperscaler committed capacity disclosures) which should be treated as high-information catalysts. Rental-rate compression at the infrastructure layer is not a teardown of economics but a margin reallocation — winners will be the owners of the software/managed layers and high-throughput interconnects.

Near-term risks cluster around geopolitics and macro-driven compute demand: an escalation that disrupts East Asian supply chains or a sharp downturn in enterprise IT spend could flip the cycle within 3–6 months. Longer-term structural risks include custom-accelerator substitution and inference commoditization, which will shift profit pools from GPU-centric vendors to vertically integrated cloud owners and networking/IP providers over 2–5 years. Monitor foundry utilization, hyperscaler capex guidance, and hybrid cloud pricing spreads as primary reversal signals.

Key Decisions for Investors

• Long TSM (18–24 months): core position (3–4% NAV). Thesis: capture foundry pricing power and equipment cadence while monitoring wafer bookings quarterly. Target +40% if utilization stays elevated; downside -20% if hyperscalers pull back. Use 6–12 month trailing stop or hedge with 18-month put spread to cap loss.
• Pair trade — Long AVGO / Short INTC (6–12 months): equal notional positions (1.5% NAV each). Rationale: AVGO monetizes networking/custom-chip adoption; INTC remains exposed to execution and legacy node mix. Expect 20–30% relative outperformance; cap risk by buying 6–9 month ITM calls on AVGO and 6–9 month OTM puts on INTC as a defined-cost structure.
• Directional options — Long AMD Jan-<next-year> 2027 calls (12–18 months): allocate 1% NAV to 2–3 call spreads. Aim to capture agentic-AI CPU cycle with 3:1 asymmetric payoff while limiting premium erosion. Hedge by selling shorter-dated calls if implied vols spike.
• Risk-management — Buy 3–6 month protective puts (5–10% OTM) on GOOGL/MSFT/AMZN covering 1–2% of NAV if hyperscaler exposure >5% each. Cost of insurance ~1–2% but preserves optionality against sudden demand shock or geopolitical disruption that would cascade through AI capex.