The AI CPU Revolution: The $200B Market 98% of Investors Are Missing

FinanceThis Is A Historic Buying Opportunity (98% Will Miss It)

TL;DR

AI agents are shifting the data center battleground from GPUs to CPUs, creating a $200 billion market where Nvidia, ARM, Qualcomm, and AMD are competing, with the biggest opportunity lying in underpriced CPU stocks like Qualcomm that have 2+ years before delivering revenue.

Key Takeaways

1Nvidia expects $20 billion in CPU revenues this year and identifies AI CPUs as a brand new $200 billion total addressable market, with their custom Vera CPU delivering first units to Anthropic, OpenAI, SpaceX, and Oracle last quarter
2ARM's AGI CPU, delivered to multiple customers with Meta as lead partner, has over $2 billion in demand but ARM stopped reporting remaining performance obligations the same quarter it started shipping hardware, making forward visibility difficult until fiscal 2028
3Qualcomm targets over $15 billion in data center revenue by 2029 (50x growth from current $300 million), with Meta's multi-generation agreement and Azure deployment starting 2027, but the C1000 CPU doesn't enter production until Q2 2028, creating a 2-year valuation disconnect
4AMD committed $25.5 billion in equity warrants and investments to secure 14 gigawatts of deals from OpenAI (6 GW), Meta (6 GW), and Anthropic (2 GW), effectively giving away up to 20% of the company
5AI agents flip the GPU-centric model by forcing CPUs to handle unpredictable, sequential work like planning, tool calls, and data movement while GPUs handle the math, making CPU performance a critical competitive factor
6AMD's Venice CPUs ship at a ratio of 18 CPUs per 72 Instinct MI455X GPUs in Helios racks, yet almost no Wall Street analyst accounts for these expensive CPUs in their market models
7Meta and TSMC are the two companies sitting at the center of the CPU battleground, with Meta partnering across all four competitors and TSMC manufacturing every advanced CPU node

Notable Quotes

“Nvidia might be the king of AI CPUs, but the castle belongs to ARM.”
— Alex

“Nvidia's CFO said they expect around $20 billion in CPU revenues this year alone, and that AI CPUs are a brand new $200 billion total addressable market.”
— Alex

“The old way is counting units, how many chips or dollars every company sold versus their competitors. But the new way is counting tokens and watts.”
— Alex

“AMD is giving away up to 20% of the entire company to secure these deals.”
— Alex

Chapters

1. The GPU-to-CPU Shift in AI Infrastructure

GPUs break problems into parallel pieces while CPUs handle unpredictable sequential work; AI agents require constant CPU involvement for planning, tool calls, and data movement, making the CPU the new battleground—shifting from Nvidia's GPU dominance to a contested $200 billion CPU market

2. Nvidia's CPU Strategy: Grace, Vera, and ARM Dependency

Nvidia sold 2.5 million Grace CPUs but pays royalties to ARM for the Neoverse V2 core design; their custom Vera CPU with 88 in-house Olympus cores eliminates ARM dependency while maintaining code compatibility, already delivered to Anthropic, OpenAI, SpaceX, and Oracle

3. ARM's Historic Pivot: From Licensing to Hardware

ARM ended 35 years as a blueprint-only company by launching the AGI CPU with Meta as co-developer, delivering it to customers and claiming over $2 billion in demand with only 50% manufacturing capacity secured; however, ARM's price-to-sales ratio of 50 is 10x the semiconductor industry average

4. Qualcomm's 50x Data Center Gamble: $15B Target by 2029

Qualcomm targets $15 billion in data center revenue by 2029 from current $300 million, with signed deals from Meta and Microsoft Azure, but the C1000 CPU doesn't ship until Q2 2028, making it priced at zero despite massive future commitments

5. AMD's Helios Racks and Gigawatt Commitments

AMD sold 14 gigawatts of Helios rack systems (each containing 72 MI455X GPUs and 18 Venice CPUs) to OpenAI, Meta, and Anthropic, but committed $25.5 billion in warrants and equity—potentially 20% of the company—to secure these deals

6. The Hidden CPU Revenue in AMD's Helios Model

Wall Street analysts model AMD's GPU sales but ignore the 18 powerful Venice CPUs shipping per rack; these CPUs have zero competition within the integrated Helios package and are critical for agentic AI workloads

7. Valuation Reality Check and Future Metrics

ARM's forward P/S is 39 and trades 10x higher than Qualcomm (P/S under 4) despite Qualcomm having larger data center potential; the market is pricing in ARM's AGI CPU perfectly but pricing Qualcomm's C1000 at zero, creating asymmetric risk

8. Meta and TSMC: The Invisible Infrastructure Kingpins

Meta partners with ARM, Qualcomm, AMD, and Nvidia simultaneously while being ARM's AGI CPU lead partner, making Meta the best indicator of AI spending trends; all four competitors depend on TSMC's advanced nodes for production

Key People & Entities

Alex: Video creator; 8-year MIT electrical engineer and AI researcher analyzing the CPU market shift
Colette Crest: Nvidia CFO who announced $20 billion expected CPU revenues and $200 billion TAM
Lisa Su: AMD CEO who confirmed demand for accelerators and CPUs growing faster than expected
Satia Nadella: Microsoft executive who confirmed Azure will deploy Qualcomm's accelerators starting 2027
Nvidia: GPU and CPU chip designer; selling Vera custom CPUs; dominates GPU market with new $200B CPU ambitions
ARM: Architecture licenser and new hardware maker; launched AGI CPU with Meta partnership; earns royalties from Nvidia and others
Qualcomm: Smartphone and automotive chip maker entering data center CPUs; targeting $15B revenue by 2029 with signed Meta and Microsoft agreements
AMD: GPU and CPU manufacturer; selling Helios rack systems with Venice CPUs; signed 14 GW of deals

Glossary

GPU: Graphics Processing Unit; breaks problems into thousands of tiny parallel pieces for simultaneous computation, ideal for predictable math-heavy tasks like neural networks
CPU: Central Processing Unit; handles unpredictable, sequential work one problem at a time, essential for planning, decision-making, and tool calls in AI agents
AI Agent: An autonomous system that reads requests, creates step-by-step plans, makes separate tool calls, writes code, and spawns sub-agents—requiring heavy CPU involvement alongside GPU math
ARM Architecture: A CPU instruction set design licensed to chip makers like Qualcomm and Nvidia; ARM collects royalties on every chip sold using their design
Instruction Set: The language a processor understands and the list of commands software can give it; Nvidia's Vera maintains ARM compatibility to avoid rewriting existing code
Rack Scale System: A complete hardware unit sold as one integrated package (e.g., AMD's Helios) containing accelerators, CPUs, networking, and software optimized to work together
Gigawatt (GW): One billion watts of power; roughly the output of a large nuclear reactor; used to measure deployed AI infrastructure scale
Remaining Performance Obligations (RPO): Revenue a company has under contract but hasn't collected yet; provides visibility into future quarters and is often used to predict growth
Price-to-Sales Ratio (P/S): Stock price divided by revenue per share; a valuation metric comparing companies' market caps to their annual revenues
Wafers: The thin slices of semiconductor material (typically silicon) on which chips are manufactured, production capacity for which constrains chip supply

Explore