The CPU Rally Muse Woke Up: The Real Reason Arm Jumped 17%

This rally is bigger than one product called Muse

On September 21, Arm shares jumped 17.1%, Intel rose 12.1%, and AMD gained roughly 10%. The same day, Meta climbed more than 11% on excitement around Muse, its AI agent product. The story the market connected was simple: if AI moves from chatbots to agents, computing demand grows, and that demand may not stay confined to GPUs.

The causation deserves a closer look, though. There is no evidence that enthusiasm for Muse directly lifted Arm's revenue. Muse looks more like a catalyst that put an investment thesis already in motion on full display: agentic AI can reignite demand for CPUs.

<div class="flow"> <div class="step"><div class="n">01</div><div class="t">Agents plan</div><div class="s">They chain together multiple steps of work rather than answering a single question.</div></div> <div class="step"><div class="n">02</div><div class="t">They call tools and databases</div><div class="s">Search, memory, API calls and networking tasks multiply.</div></div> <div class="step"><div class="n">03</div><div class="t">CPU workload grows</div><div class="s">Orchestration, general-purpose computation and I/O handling all expand.</div></div> <div class="step"><div class="n">04</div><div class="t">Power efficiency becomes money</div><div class="s">The more agents run continuously, the more performance-per-watt and rack density matter.</div></div> </div>

<p class="quote">If the GPU is the AI's compute engine, the CPU is closer to the operating layer that ties thousands of tasks and data flows into an actual running service.</p>

The bigger change at Arm: an IP company starts selling CPUs directly

The most important fact in this story is not the stock move but a shift in Arm's business model. In March 2026, Arm unveiled the Arm AGI CPU, packing up to 136 Neoverse V3 cores. It is the first time in the company's history that it is offering production silicon for data centers directly, rather than just licensing designs. Meta is the lead partner and co-developer on the product.

<div class="metrics"> <div class="metric"><div class="value">136</div><div class="label">Maximum cores in the AGI CPU</div></div> <div class="metric"><div class="value">$2B+</div><div class="label">Customer demand Arm disclosed for FYE27-FYE28</div></div> <div class="metric"><div class="value">2x+</div><div class="label">Year-over-year increase in data-center royalties last quarter</div></div> </div>

That $2 billion figure should not be read as a confirmed backlog or as revenue. Arm's own language is customer demand. The company said it delivered early units to multiple customers in late July, and that it has secured manufacturing capacity to support a previously disclosed $1 billion opportunity. In other words, the market's focus is shifting from whether demand exists to how much of that $2 billion in interest actually converts into shipments and revenue.

<p class="note">There is one more correction worth making. The AGI CPU is not a product waiting for a first shipment in the fourth quarter of 2026. Arm unveiled the product in March, and said it had already delivered early units to multiple customers by the end of July.</p>

Why agentic AI matters for CPUs again

Matrix multiplication for large models still overwhelmingly needs GPUs and AI accelerators. But agents do more work between model calls. They retrieve data, read user state, call APIs, distribute work across multiple agents, and move between networking and storage.

Arm itself claims in its materials that CPU requirements could grow at least fourfold as agentic applications proliferate. That figure is a company projection, not confirmed industry data. Still, the investment logic is clear enough. As inference calls and tool calls multiply, general-purpose computing outside the accelerator grows alongside them.

Arm's edge here is power efficiency and core density. Power is already a capital constraint in AI data centers. If a CPU can process more work within the same power and rack footprint, that efficiency stops being a mere spec and becomes part of the data center's underlying economics.

But don't read this as a wholesale win for CPUs

<div class="table-wrap"> <table> <thead><tr><th>Company</th><th>What to watch in this shift</th><th>Numbers investors should track</th></tr></thead> <tbody> <tr><td><strong>Arm</strong></td><td>A new revenue stream from direct AGI CPU sales, on top of IP and royalties</td><td>AGI CPU revenue conversion, data-center royalties, manufacturing capacity</td></tr> <tr><td><strong>AMD</strong></td><td>Exposure to both EPYC server CPUs and Instinct accelerators</td><td>Server CPU share, accelerator revenue, margins</td></tr> <tr><td><strong>Intel</strong></td><td>CPU supply shortage and hopes for a recovery in x86-based demand</td><td>Server shipments, supply fulfillment, data-center margins</td></tr> <tr><td><strong>Nvidia</strong></td><td>GPUs remain central, but the company bundles Arm-based CPUs like Grace and Vera into full systems</td><td>Rack system revenue, combined CPU-GPU adoption</td></tr> <tr><td><strong>TSMC</strong></td><td>The manufacturing hub for advanced chips from Arm, AMD, Nvidia and others</td><td>Advanced-node utilization, CoWoS and packaging capacity, share of AI revenue</td></tr> </tbody> </table> </div>

Arm, Intel and AMD all rallied hard on September 21, but not every semiconductor stock moved in the same direction. That suggests the day's price action is better read not as money flowing wholesale from GPUs to CPUs, but as AI investors' field of attention widening beyond GPUs. </markdown>

Insight Times Editorial Desk