Anthropic weighs building its own AI chips- Reuters By Investing.com

Intel extended its weekly gain to over 33% on news of a Google AI collaboration. Anthropic is reportedly considering designing its own AI chips amid a shortage of processors, though plans are in early stages and it may still opt to buy chips. The company recently signed a long-term deal with Google and Broadcom for Google's Tensor Processing Units; Meta and OpenAI are reported to be having similar in-house chip discussions.

Analysis

When large model operators move toward bespoke accelerators or increased long-term chip commitments, the highest-probability second-order winners are data-center infra suppliers — high-bandwidth memory, switch/ASIC vendors, and packaging/interposer specialists — because incremental AI capacity typically multiplies demand for these components by 2-4x per unit of compute. Conversely, any shift to in-house ASICs raises order volatility for general-purpose accelerator suppliers and can accelerate consolidation around a smaller set of foundries, squeezing mid-tier fabless vendors over a 12–36 month window.

The timeline for meaningful impact is multi-year: expect prototypes and pilot racks within 12–24 months, production ramps in 24–48 months. Key observable catalysts that will move markets sooner are foundry allocation announcements, taped-out test chips, and hyperscaler purchase orders for memory/interconnect — each can re-rate suppliers by 10–30% ahead of broad adoption. Tail risks include failed tape-outs, access limits to advanced nodes, or regulatory export controls that could both delay projects and transiently boost incumbent GPU demand as customers backfill supply.

Market pricing often underestimates capital intensity and time-to-scale; investors who assume immediate third-party demand growth are overpaying for near-term optionality. A pragmatic stance is to size exposure to infra-exposed, sticky-revenue names while trimming exposure to firms whose growth depends on one-off model provisioning. Maintain tight stop-losses and use spreads to finance directional exposure because binary execution outcomes (success/failure of designs) create fat tails in returns.

Key Decisions for Investors

• Buy AVGO 3–9 month call spread (one staggered leg) to capture likely near-term bump in data‑center networking/HBM switch demand; target 20–30% upside, max loss ~6–8% of notional (2:1 reward:risk).
• Buy GOOGL 6–12 month 25–35% OTM call spread to hedge for faster-than-expected cloud/Tensor monetization from strategic AI partnerships; size <3% portfolio, aim for 3:1 payoff if adoption accelerates.
• Accumulate INTC on >10% pullback as a tactical 3–9 month swing trade (small position): potential upside from packaging/CPU/FPGA orders but high execution risk—limit to 1–2% portfolio and use 15% stop-loss (target 40% rally on re-rating).
• Pair trade (6–12 months): long AVGO / short META equal notional. Rationale: AVGO benefits from infra spend concentration while META faces margin pressure from capex and internal silicon investments. Target asymmetric payoff ~2.5:1, cut if spread narrows by 10%.