Citi Just Put an $800 Target on AMD, Here's the Muse Math Behind It
The note landed Tuesday morning. Atif Malik wrote it. Citi's semiconductor analyst put an $800 price target on AMD. The previous target sat at $575. The stock traded around $648 when the note hit the tape. That gap, roughly $150, represents about 24% of upside if Malik's math holds.
The stock rose 1% in premarket. It opened at another record high. Nobody on the desk looked surprised. They had seen the Muse numbers. They had watched the App Store charts. The note simply confirmed what the tape already suspected: the CPU trade has legs.
What Meta Muse Actually Does
Meta launched Muse in September 2026. The app hit the top of Apple's free chart within ten days. It accumulated 448,000 daily active users by that point, a pace that took ChatGPT nearly a year to match at the 200,000 download mark.
By day 22, Muse had crossed 5 million downloads. ChatGPT, Grok, and Claude all moved slower. The app held the top spot on the U.S. App Store free iPhone chart for 12 consecutive days.
Those are adoption numbers. They matter because Muse is not a chatbot. It runs on-device. Meta built the open-source "Glimmer" model in direct partnership with AMD. The processors let an agentic model execute efficiently on consumer hardware. The architecture avoids routing every user action through a data centre. Token costs drop. Latency drops. The workflow persists.
Muse does not just answer a question and stop. It runs. It orchestrates. It executes tasks across Instagram, Facebook, Canva, and Slack. Meta expanded those integrations in late September 2026, targeting small business owners and creators.
The app keeps working while you sleep. That behavior drives the compute story.
Why Agentic AI Needs CPUs
The AI narrative for the past three years has belonged to GPUs. Nvidia built the cathedral. Every analyst learned the language of accelerators and training clusters and FLOPS.
Agentic AI changes the syntax.
An agent runs a workflow. It might run for hours. It might run for days. Daniel Newman, CEO of Futurum Group, put it plainly to CNBC: "CPUs are actually performing the workflows while GPUs are doing the thinking".
The GPU handles inference. The CPU handles orchestration. The CPU manages state. The CPU keeps the virtual machine alive. Meta and OpenAI both rely on virtual machines to slice a single server into lightweight environments. That architecture lets one physical server handle a large number of concurrent users. Each sandbox needs dedicated cores and memory. Muse runs on AMD EPYC Turin hosts with two cores and 8GB of memory per sandbox.
OpenAI's Dots runs on an AMD EPYC 9V74 processor. It draws on nine cores. Some AMD EPYC chips ship with as many as 192 cores. The secondary-market price for the EPYC chips powering both agents sits below $3,000. Nvidia GPUs can top $30,000 each and are purchased in massive clusters.
The economics favor CPUs in this workload. The ratio of CPUs to GPUs in agentic deployments is shifting. Lisa Su said in May 2026 that the ratio is nearing 1-to-1. In agent-heavy deployments, CPUs could outnumber GPUs.
That is not a small shift. That is a structural change in how data centres get built.
The $300 Billion Number
Citi's revised CPU total addressable market model is the spine of the note. Malik raised the 2030 CPU TAM to $300 billion. The 2025 figure was $29 billion. That represents a 60% compound annual growth rate.
The prior Citi estimate sat at $237 billion by 2030. Muse pushed it higher. The agentic AI workload, always on, persistent, orchestrating, drives inference-related compute, memory, and networking consumption.
AMD sits as the primary beneficiary. Intel sits as the secondary. Malik wrote that AMD will capture the CPU renaissance because Meta is one of the largest customers of AMD's server business.
Morgan Stanley estimates that Muse alone could account for 20% of AMD's 2026 chip sales.
That is one application. One product. One adoption curve.
What AMD Already Showed
The Q2 2026 earnings report gave the thesis evidence.
AMD posted record non-GAAP earnings per share of $1.66. Consensus estimates sat at $1.61. Total revenue surged 50% year over year to a record $11.5 billion. Analysts projected approximately $11.34 billion.
The Data Center segment drove the beat. Revenue more than doubled year over year to $6.7 billion. EPYC server CPUs and Instinct AI accelerators fueled the growth. Data center revenue reached nearly 60% of total sales in the quarter ended June.
AMD projected second-half 2026 server CPU revenue to grow 80% year over year. The company expects 70% growth in 2027. Data center revenue should more than double in 2027.
Lisa Su said AMD has already increased supply during 2026. Demand remains above available capacity. She said AMD would "substantially increase" supply in 2027.
The World Labs acquisition added another layer. AMD agreed to pay $8.2 billion for the San Francisco AI model lab. Dr. Fei-Fei Li, the "Godmother of AI", leads it. The lab focuses on software for 3D environments. The deal should close by the end of 2026.
The stock has gained more than 180% in 2026. It crossed $1 trillion in market value for the first time in September 2026. AMD holds roughly 46% of the x86 CPU market.
The Risks Nobody Puts in the Headline
The $800 target is a target. It is not a fact. Malik could be right. He could be early. He could be wrong.
AMD trades at a valuation that assumes significant execution. The stock gained 32% in the month preceding the Citi note. Intel gained 21% over the same stretch. Both outpaced every tech megacap company.
Competition does not disappear. Nvidia still owns the GPU market. Meta's infrastructure chief described a deliberate three-way strategy: "There's a place for Nvidia, there's a place for AMD, and there's a place for our own custom silicon as well. We need all three".
Custom silicon from cloud providers represents a long-term threat. Arm-based chips are developing. OpenAI uses multiple CPU providers. The market is not a duopoly forever.
The agentic AI thesis also depends on adoption. Muse is popular. Dots is popular. GrokBot, Instinct, and GeminiSpark exist. The question is whether these applications become habitual. Whether they drive sustained compute demand. Whether the workloads stay CPU-bound or shift as hardware evolves.
Malik's model assumes the CPU TAM expands at 60% CAGR. That is an aggressive number. It assumes agentic AI adoption accelerates. It assumes AMD captures a significant share. It assumes Meta's commitment holds.
Those are reasonable assumptions. They are not guaranteed outcomes.
What Comes Next
AMD reports third-quarter fiscal 2026 results in the coming weeks. Vijay Rakesh at Mizuho expects "beat-and-raise results." He sees Meta's Muse and OpenAI's Dots fueling server CPU demand. Mizuho raised its AMD target to $705 from $580. Rakesh kept a neutral rating on Intel but raised his target there to $114 from $92.
BNP Paribas raised its AMD target by 60% to $960 in early October. The firm called AMD "increasingly becoming a leading provider of AI infrastructure".
Stifel raised its target to $700 from $635. Seaport Global upgraded the stock. The consensus price target among 55 analysts sits around $619, with a "Strong Buy" rating. Citi's $800 sits above the average but not by an absurd margin.
The CPU renaissance narrative has momentum. Citi articulated the mechanism. The market is pricing it in.
The next chapter arrives with Q3 earnings. Watch the server CPU revenue line. Watch the data center guidance. Watch what Lisa Su says about 2027 capacity. The stock will do what it does. The thesis will either hold or it won't.
The Takeaway
Citi's note is not a prediction. It is a framework. The framework says agentic AI changes the compute mix. The framework says CPUs matter again. The framework says AMD sits at the intersection of a product cycle and a customer relationship.
Meta Muse gave the framework a concrete catalyst. The App Store numbers gave it adoption evidence. The Q2 earnings gave it revenue proof. The World Labs acquisition gave it a forward-looking anchor.
Whether the stock rips another 25% depends on execution. On supply. On demand. On the messy, unpredictable business of selling silicon into a market that is still figuring out what it wants to be.
Malik put $800 on the page. The market will decide if he earns it.
Comments
Post a Comment