Arm Is Now Making Its Own Chips

Arm announced it will produce and supply its own Arm AGI CPU (fabricated by TSMC on a 3nm process) with full production availability targeted in H2 this year; initial customers/samplers include Meta, OpenAI, SAP, Cerebras, Cloudflare, SK Telecom and Rebellions. Arm claims best-in-class performance-per-watt vs x86 rivals (Intel/AMD) for agentic AI workloads, positioning the firm to capture a slice of a data-center CPU market Creative Strategies forecasts rising from $25B today to $60B by 2030 (and ~ $100B including agentic AI). Upside is new direct revenue and potential large electricity-cost savings for customers; downside risk is alienating longtime licensees and triggering more direct competition with established CPU/GPU vendors.

Analysis

This is a structural strategy shift: Arm is moving from a high-margin, low-capex licensor to an active supplier pursuing a slice of a capital-intensive, winner-takes-most data center CPU market. That transition changes Arm’s marginal economics (lower gross margin per unit but higher addressable revenue) and creates a conflict-of-interest with licensees that could accelerate either consolidation around neutral suppliers or quicker vertical integration by hyperscalers.

The most immediate supply-chain second-order effect is node competition at TSMC 3nm. Arm consuming premium 3nm capacity tightens the allocation timetable for other aggressive adopters (NVIDIA, Apple’s future nodes), increasing foundry pricing power and creating short-term scarcity that favors TSMC (and hurts node-constrained incumbents). If Arm’s energy-per-inference advantage is real, customers with large fleets (Meta, cloud providers) can realize multi-hundred-million-dollar opex savings within 12–36 months, making procurement decisions driven more by TCO than single-socket peak FLOPS.

Key catalysts and failure modes are near-term and measurable: third-party benchmarks and workload-level TCO studies (1–6 months), announced design wins beyond samples (6–12 months), and TSMC capacity updates (next 12 months). Tail risks: partner retaliation (license cancellations or exclusivity switches), slower-than-advertised performance gains, and regulatory scrutiny over Arm’s dual role—any of which could erase investor optimism rapidly.

From a competitive stance, the market underestimates how quickly hyperscalers can pivot procurement to neutralize Arm’s advantage if they perceive capture risk. Conversely, if Arm secures non-exclusive, repeatable design wins at scale, the company can convert licensing leverage into a recurring hardware revenue stream that materially re-rates its growth profile over 12–36 months.

Key Decisions for Investors

• Long ARM (6–18 months): Buy ARM shares or a 12-month call spread to express upside if H2 production ramps and design wins expand; upside scenario +40–80% if adoption scales, downside -30% if licensee backlash or benchmarks disappoint. Set stop at -20% from entry.
• Pairs trade — Long ARM / Short INTC (12 months): ARM upside from agentic CPU traction vs Intel downside from share loss in CPU cycles and pricing pressure; target 2:1 reward:risk (expect ~30% gross spread move), hedge size to 25% portfolio volatility.
• Long TSM (6–12 months): Buy TSM calls to capture foundry pricing and utilization benefit from premium 3nm demand; expected asymmetric payoff +20–40% on better-than-feared node scarcity; cap exposure given geopolitical tail risk.
• Long META (12–24 months): Accumulate META on conviction that opex savings from power-efficient CPUs accelerate internal AI deployments; expect 15–30% upside if savings translate to faster product rollout, risk of execution slippage and competitive ad pressure.
• Event hedge: Buy 3–6 month out-of-the-money puts on ARM sized to 30% of position to protect against near-term partner defections or negative benchmark prints (cost as insurance against a >25% drawdown).