Intelligence is foundation
Podcast Subscribe
Voices & Thought Leaders Wednesday, 25 March 2026

Arm's Not Licensing Anymore - They're Selling Chips

Share: LinkedIn
Arm's Not Licensing Anymore - They're Selling Chips

Arm announced their own CPU this week. Not a design to license - an actual chip you can buy. For a company that's spent 34 years licensing IP to other manufacturers, this is a structural shift, not a product line extension.

Ben Thompson's analysis at Stratechery cuts through the obvious competitive narrative. Yes, Arm's clients are building their own designs. But the deeper driver is how computing itself is changing. The old model - license the architecture, let others manufacture - worked brilliantly for mobile. It's less obvious that it works for what's coming next.

Why the Licensing Model Had Limits

Arm's business was elegant. They designed processor architectures. Companies like Apple, Qualcomm, and Samsung licensed those designs, customised them, and built chips. Arm took royalties on every unit sold. Low capital risk, high margins, massive scale.

That model assumed clients wanted general-purpose designs they could adapt. But AI workloads aren't general-purpose. They're highly specific - optimised for matrix multiplication, low-precision arithmetic, massive parallelism. The architecture matters less than the system integration. Memory bandwidth, interconnects, power delivery - the chip is part of a larger system, not the product itself.

When your clients are building systems, not just chips, licensing an architecture becomes less valuable. They need control over the full stack. Arm's royalty model doesn't capture that value. So Arm's moving up the stack too - selling complete solutions, not just blueprints.

What Arm's Actually Selling

Arm's new CPU isn't competing with Intel or AMD in the traditional sense. It's targeting cloud infrastructure - data centres running AI inference at scale. The pitch is efficiency. Arm's architecture uses less power per operation than x86. In a data centre, power is the largest operational cost after hardware. Small efficiency gains multiply across millions of servers.

But selling chips directly changes Arm's relationship with their ecosystem. Apple and Qualcomm are now potential competitors, not just customers. That's risky. Arm's betting the market's big enough for both models - licensing for some use cases, direct sales for others. Whether that's true depends on how their clients react.

Thompson's insight is that Arm's constraints forced this move. Licensing revenue is capped by the number of chips clients build. Direct sales let Arm capture more value per unit, but require manufacturing partnerships, supply chain management, and customer support. Different business, different risks.

The Bigger Pattern - Integration Over Modularity

Arm's shift reflects a broader trend. For decades, computing got more modular. Separate companies handled design, manufacturing, software, integration. That modularity enabled innovation - you could swap components without rebuilding the whole system.

AI workloads are reversing that trend. Performance comes from tight integration between silicon, software, and system design. Apple's M-series chips work because Apple controls the hardware, the OS, and the applications. NVIDIA's dominance in AI comes from their integrated stack - chips, drivers, CUDA, libraries. You can't just drop in a competitor's GPU and expect the same results.

Arm's realising that selling architecture in isolation isn't enough. They need to sell systems. That's a harder business - more capital intensive, more competitive, more exposed to manufacturing risk. But it's where the value is moving.

What This Means for Builders

If you're building on cloud infrastructure, Arm-based instances are already available from AWS, Google, and Azure. They're cheaper to run for certain workloads - web servers, containerised apps, inference tasks. The cost difference is real, but the tooling and ecosystem are still maturing. You'll hit more edge cases than with x86.

For hardware companies, Arm's move is a signal. The era of clean separation between design and manufacturing is eroding. Vertical integration is coming back, and it's being driven by the economics of AI workloads. If your business model assumes modularity, that assumption is getting riskier.

Arm's not abandoning licensing. They're adding a second model. But the fact they felt they had to tells you something about where the market's heading. Computing is consolidating again. The question is who controls the stack.

Read Ben Thompson's full analysis at Stratechery

More Featured Insights

Builders & Makers
Seven Reasons AI Projects Fail - From Someone Who's Watched It Happen
Robotics & Automation
Amazon Buys a Humanoid Company - And Nobody's Surprised

Video Sources

Theo (t3.gg)
AI has a subsidization problem
Ania Kubów
Deploying AI Models with Hugging Face - Hands-On Course
Two Minute Papers
DeepSeek Just Fixed One Of The Biggest Problems With AI
Matthew Berman
Hard Takeoff has started

Today's Sources

DEV.to AI
Why AI Projects Fail: Lessons From 20+ Enterprise Implementations
DEV.to AI
AI for Small Business: 10 No-Code Automations You Can Implement Today
Towards Data Science
My Models Failed. That's How I Became a Better Data Scientist
The Robot Report
Amazon acquires humanoid developer Fauna Robotics
The Robot Report
TI partners with NVIDIA to accelerate robot deployments
The Robot Report
Russ Tedrake to unveil stealth AI startup at Robotics Summit
Hackaday Robotics
3D Printed Robot Arm Built For Learning Purposes
ROS Discourse
From 'move and pray' to movement feedback
ROS Discourse
FusionCore, ROS 2 Jazzy sensor fusion package
Ben Thompson Stratechery
Arm Launches Own CPU, Arm's Motivation, Constraints and Systems
Latent Space
Why There Is No 'AlphaFold for Materials' - AI for Materials Discovery
Latent Space
[AINews] Apple's War on Slop

About the Curator

Richard Bland
Richard Bland
Founder, Marbl Codes

27+ years in software development, curating the tech news that matters.

Subscribe RSS Feed
View Full Digest Today's Intelligence
Free Daily Briefing

Start Every Morning Smarter

Luma curates the most important AI, quantum, and tech developments into a 5-minute morning briefing. Free, daily, no spam.

  • 8:00 AM Morning digest ready to listen
  • 1:00 PM Afternoon edition catches what you missed
  • 8:00 PM Daily roundup lands in your inbox

We respect your inbox. Unsubscribe anytime. Privacy Policy

© 2026 MEM Digital Ltd t/a Marbl Codes
About Sources Podcast Audio Privacy Cookies Terms Thou Art That
RSS Feed