Arm Holdings spent 35 years as the most important company in computing that nobody needed to think about. Its chip architecture is inside every smartphone, most laptops, and a substantial share of enterprise servers. Arm licensed that architecture to Apple, Qualcomm, Samsung, and more or less the whole industry, and collected royalties. Quietly. Very profitably.

In March 2026, Arm did something new: it shipped its own chip. The Arm AGI CPU, 136 cores, TSMC 3nm, designed for AI data center workloads. Meta signed on as the lead customer and co-developer. OpenAI, Cerebras, Cloudflare, and others placed orders. Lenovo, Supermicro, and ASRock started taking system orders.

Six weeks later, Arm had $20 billion in customer demand.

Manufacturing capacity is secured for $1 billion of it.

Why the Demand Makes Sense

The AGI CPU is not trying to be an NVIDIA killer. Where NVIDIA’s GPUs handle the heavy parallel computation of model training and inference, the AGI CPU is built for orchestration: managing how agents fan out across systems, routing queries, retrieving context, moving data, scheduling workloads. CPU work. A lot of it.

Arm’s thesis is that the agentic era requires roughly 4x more CPU cores per gigawatt of data center capacity than traditional AI. Thirty million CPU cores per GW in a conventional AI data center; 120 million in a fully agentic one. Infrastructure operators deploying AI at enterprise scale have consistently identified the orchestration layer as the hard part. The AGI CPU is the first chip built specifically for that problem.

The customer list suggests the market agrees. OpenAI is a launch partner. So is Cerebras. So is Rebellions. These are not companies that sign up for chip platforms as a courtesy.

The TSMC Problem

The gap between $20 billion in demand and $1 billion in secured supply comes down to one thing: TSMC’s advanced node capacity is already heavily committed. The hyperscaler custom silicon boom has been consuming leading-edge fab capacity for years, as TSMC’s own forward guidance reflects. Arm is a new entrant to that queue.

Arm’s CFO is maintaining a $1 billion revenue target, with first production revenue expected in Q4 FY2027. Converting the broader $20 billion pipeline requires securing wafer allocations, memory, packaging, and test capacity from a supply chain that is already stretched thin. The stock dropped 10 percent on earnings day despite record results. Investors, presumably, can do the math.

A Different Kind of Compute Problem

Arm projects $15 billion in AGI CPU revenue by fiscal 2031. That would make the chip business larger than the IP licensing business it spent 35 years building. It took three decades without ever shipping a chip. Now it cannot make chips fast enough.

The semiconductor supply chain has a way of humbling ambition on both sides of the demand curve. Arm made the right call on where the agentic compute stack was heading, built the right product, and found the right customers. The wait list is $19 billion long.

–
By the Control Plane Editorial Team