Arm spent thirty-five years doing one thing: designing chip architectures, licensing them out, and collecting royalty checks. It never touched silicon. Never built a chip of its own. Just sat in the middle of the semiconductor world collecting a toll on every smartphone, every tablet, every embedded device that needed to be small and power-efficient. Ninety-seven percent gross margins. No manufacturing risk. No inventory. The best business in tech, and they never risked it.
Until March 24, 2026. That's the day Arm dropped the AGI CPU, its first-ever in-house data center chip. And they didn't ship a reference design for someone else to fab. They shipped an actual chip, with Meta as the lead customer and orders already north of two billion dollars. The royalty collector became a competitor. The architect became a builder.
The custom silicon megatrend is the defining story of this decade in semiconductors. Apple built the M-series and runs every Mac on Arm. Google built Tensor for its Pixel phones. Amazon built Graviton for AWS and saved hundreds of millions in cloud infrastructure costs. Microsoft built Cobalt for Azure. Meta is building its own Arm-based designs. NVIDIA's Grace CPU runs on Arm. Every single one of them pays Arm a license fee and a per-chip royalty. And now Arm sells them the finished chip too.
That's the part most people miss. Arm doesn't have to pick a side in the custom silicon war because it supplies every side. Apple designs its own M-series chips in-house — Arm collects the architecture license. Meta buys the AGI CPU off the shelf — Arm collects the chip sale. Amazon builds Graviton themselves — Arm collects the royalty. The moat isn't the instruction set. The moat is that the entire industry standardized on Arm, and nobody's scaling a viable alternative at the same volume.
The financials reflect that moat. Revenue hit nearly five billion dollars in the most recent fiscal year, growing better than twenty percent year over year. Gross margins sit at roughly ninety-seven percent — pure IP licensing and royalty with no inventory carrying cost and no fab depreciation. Free cash flow came in just under a billion dollars. The balance sheet is cleaner than almost any company in semiconductors: under half a billion in total debt against nearly three billion in cash. Net cash positive, no existential risk, no debt maturity wall.
The growth trajectory is where it gets interesting. Arm is targeting fifteen billion dollars in AGI CPU revenue by 2031. That's additive on top of the existing licensing and royalty business, which already prints cash with zero marginal cost. The AGI CPU market — chips purpose-built for AI inference and training workloads — is the fastest-growing segment in data center silicon. Meta signed on as anchor, but Arm is building for a world where every hyperscaler needs more inference compute than they have the engineering bandwidth to design themselves.
The stock pulled back from its highs as valuation fears crept in. One bank downgraded on multiple expansion. Another firm raised their target and argued the stock was still cheap relative to the TAM. That's what happens when a company trades at a multiple that reflects a thirty-five-year-old licensing business, then announces it's entering a capital-light chip-vertical that happens to carry better margins than anyone else's. The market doesn't know how to price a hybrid model, so it throws a wide range on the board and waits for the next earnings report to sort it out.
That earnings report lands July 29. Every hyperscaler capex number, every AGI CPU deployment update, every royalty guidance revision — it all hits in one filing. If AGI CPU orders expanded beyond Meta, the stock rerates on revenue visibility. If Meta ups their commitment, same story. If the market starts to price in the network effect of custom silicon — where every in-house chip validates the architecture and every AGI CPU sold opens a new profit line — the licensing multiple becomes a historical footnote.
This is a bet on the standardization of semiconductor design. Arm won the architecture war thirty years ago, and now they're cashing in the prize. The custom silicon trend doesn't threaten Arm. It makes Arm structurally indispensable. Every hyperscaler that builds their own Arm chip validates the instruction set for the next decade. Every AGI CPU Arm ships opens a revenue stream that didn't exist eighteen months ago. The pullback from highs is the market overcorrecting on short-term multiples. The thesis hasn't changed in the slightest.
Disclosure: The Signal holds no position in ARM. Positions may change. This is not financial advice.
