Arm to Sell Its Own Chips, Eyeing Sales Goal of $15 Billion

Arm will begin selling its own AGI CPU chips and expects the new chip business to generate about $15 billion annually within five years, driving total company revenue to roughly $25 billion (approximately 5x current sales). The AGI CPU (up to 136 cores, 300W) will be produced by TSMC with Meta as the first major customer; off-the-shelf systems are available now and broader volume shipments are expected in H2. Shares rose as much as 7.8% (SoftBank +8.6% in Tokyo) on the outlook; management projects EPS of ~$9 in five years versus analyst estimates of ~$1.75 this fiscal year (ex-items), noting current gross margin of 98% but expected lower margins on direct chip sales offset by much larger gross profit dollars per unit.

Analysis

Arm’s shift from pure-IP to wafer-backed SKU sales materially changes the profit pool: per-unit gross dollars will rise while corporate gross margins will mechanically compress as capital intensity and COGS scale. The second-order winners are entities that capture incremental wafer/assembly/test volume and server-system integration economics — think leading-edge foundries and hyperscaler-friendly OEMs — while long-time licensees face a new choice set that can depress their willingness to pay for differentiated Arm designs.

Operational execution — yield, power/perf delivery in real data centers, and multi-supplier qualification — is the single biggest near-term risk and the main gating factor for the street’s 3–5 year revenue assumptions. A failed thermal/efficiency story, lower-than-expected wafer supply or a public hyperscaler benchmark miss would compress valuations rapidly; conversely, stable production yields and broad OEM uptake would force incumbents (x86 vendors) into accelerated price/perf counters, compressing their margin pool over multiple quarters.

From a strategic-franchise view the market is likely under-pricing the customer-conflict angle: ARM must now balance volume-driven economics against the licensing flywheel that made the IP business defensible. That creates a durable but contested niche where Arm can extract higher per-socket dollars only if it avoids alienating its largest architecture licensees — a governance and commercial-design problem as much as a product one, with clear upside if they thread that needle and clear downside if they don’t.

Key Decisions for Investors

• Long ARM LEAPs (18–36 months): buy ARM Jan 2028 OTM call spreads to capture multi-year upside from ASP expansion while limiting theta. Risk/reward: asymmetric upside if scale-up works; capped loss = premium (~100%+ potential upside vs limited downside).
• Pair trade: long TSM (3–12 months) + long SMCI (6–12 months) / short INTC (6–12 months). Rationale: foundry + OEM systems capture incremental volume; Intel is most exposed to share loss and margin hit. Risk/reward: structural upside as datacenter box growth drives wafer demand; risk = foundry capacity reallocation or Intel tactical pricing.
• Event hedge: sell short-dated ARM calls (1–3 months) against long-dated calls (calendar spread) to fund LEAP exposure ahead of initial volume release. This monetizes near-term optimism and protects from a short-term post-launch disappointment.
• Selective long META (6–12 months) exposure via equity or call buys sized as conviction trade: benefits from early access but not sole beneficiary. Risk/reward: upside if integration improves density and lowers OPEX; downside if customer in-house alternatives scale faster.