Arm enters CPU manufacturing: the biggest bottleneck in chip design is verification, not design
The real time sink in a chip cycle is verification, debug and documentation, not RTL; 80%-90% of Arm engineers already use AI daily, and Arm is breaking its 98.5%-margin licensing model to build CPUs itself.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Arm is building CPUs itself because customers are pushing it to
Arm used to sell only IP licenses, with 98.5% gross margins, never touching inventory or wafers. One step forward, its compute subsystem already had customers grabbing pre-assembled CPU, GPU and memory subsystems as modules. The next step came because customers who wanted a general-purpose CPU for the large-model era couldn't find a supplier and went straight to Arm to push. Before announcing, Arm asked almost every license customer, and opposition was minimal: a bigger software ecosystem benefits every customer, so Nvidia, Amazon, Microsoft and Google — all companies making Arm server chips — publicly supported it.
— Rene HaasBanning AI tools is sending engineers back to the library
The total chip design cycle is 24-36 months; the most time-consuming parts are not RTL generation or architecture mapping but verification, validation, debug and documentation. AI is especially suited to these steps, and about 80%-90% of Arm engineers already use AI daily. Turning the tools off would be like telling people in the 1990s they could only use the internet two hours a day and had to go back downstairs to the library for the rest. AI is currently weakest at RTL generation and physical design, because models rely on public corpora while the industry's most critical details are private data — exactly the gap companies like Arm can fill.
— Rene HaasFrom idea to layout is five years away, not two or three
The host asked: once AI eats verification, the most time-consuming step, can 24-36 months shrink to 6-12 months? Rene Haas gave a measured answer: 2-3 years is unrealistic; five years or more, for designs with not-too-complex constraints, going from idea all the way to GDSII (the layout file handed to the fab) is possible. If you type into a tool "make a chip 10% faster, 20% cheaper and 30% more power-efficient than some reference," don't expect one-click completion. His judgment is that in 5-10 years the whole industry's design methodology will change noticeably.
— Rene HaasAI supply and demand are still far apart; the bottleneck is the construction site, not the wafer
Discussing the AI bubble, Rene Haas said you have to separate stock-level valuation froth from the industry chain itself; he stated plainly that supply is still "far short" of demand. Transformer inference and training eat both compute and memory, and demand won't stop. The real hard constraint will move to data center construction: many projects are running late and need more workers, and in various places resistance to new data center builds is emerging. Over the next 3-5 years the constraint may not be wafer capacity or memory but the physical construction step staying tight.
— Rene HaasSoftBank doing cloud is about being a customer for chip companies
For asset-heavy, capex-intensive chip/AI founders: capital availability is itself a gate, and strategic partnerships need to reach early into the supply chain, PE and banks. Behind Arm is SoftBank, and not just as a shareholder; SoftBank has just announced it will build a "neo cloud"-style cloud service, which lets young chip companies avoid only knocking on Microsoft/Google's door for a design win and instead form a direct customer relationship with the Arm/SoftBank ecosystem — one more path to deployment.
— Rene HaasHumanoids won't unify everything; specialized robots will coexist long-term
Rene Haas thinks robot 1.0 is purpose-built: mechanics and software are both optimized for a single action, and changing a production line means almost starting over. A new generation of robots that can redefine tasks through training and have a general-purpose mechanical body will first eat repetitive labor in construction, infrastructure and security. Form factor won't go in a single direction: factories, warehouses and tools are all designed to human dimensions, so humanoids have an advantage; another set of tasks is better suited to specialized forms. Cost must come down first, and delivery and warehousing will be the earliest automated scenarios.
— Rene HaasRestricting chips to China is an unwinnable infinite game
On US manufacturing he is direct: the US should build more fabs at home, needed for both national security and supply-chain diversification. On export controls, he talks about an "infinite game" — restricting chips to China cannot produce a victory with an end state, and may instead push key technology out of the US. He also rejects the idea that data centers are "big warehouses with no jobs": energy, liquid cooling and the surrounding ecosystem are all jobs that can be created in the US.
— Rene HaasIn a token world, every path goes through a CPU
If you use the token factory metaphor for accelerators, Rene Haas thinks the real system question is: who orchestrates and arbitrates where those tokens go, and delivers them to users. He believes every path goes through a CPU: beyond training, there's inference scheduling, system orchestration, context and tool calls. At the edge, on devices that can't fit a 50W GPU, Arm's power advantage is even clearer. The CPU hasn't disappeared; it's forming sustained demand alongside accelerators.
— Rene HaasIn their own words · checked verbatim
There's no computing problem that's ever been invented that doesn't utilized and can't utilize the microprocessor. It is the heart of everything. All roads lead through it around it past it. Something has to do the orchestration, arbitration decision around where those tokens go. That's what CPU is doing.
Rene Haas0:00
So now we're in that soup ourselves from the standpoint of.We're also having to figure out how to buy substrates and buy wafers and buy memory, et cetera, et cetera.
Rene Haas2:37
but I think in five to 10 years... We're going to see some amazing differences relative to how chips are designed.
Rene Haas10:35
I call BS on that because if you think about whether it's around energy.Liquid cooling.All of the things that make the data center better.That's all those are all jobs that can be created and done here.
Rene Haas28:20
Figures
| Gross margin of Arm's traditional licensing business | 98.5% | 6:02 |
| Share of Arm engineers using AI tools daily | 80%-90% | 8:07 |
| Total chip design cycle | 24-36 months | 8:07 |
Glossary
- IP licensing
- A business model where a chip company sells design blueprints rather than physical chips.
- compute subsystem
- A semi-finished design where Arm pre-assembles components like CPU and memory, leaving customers less room to modify.
- RTL
- A type of code describing a chip's logic level, between architecture and physical layout.
- GDSII
- The final physical layout file a chip design hands to the fab for production.
How to listen
AI infra investors, chip founders, cloud decision-makers, and hard-tech people who want to understand why Arm is moving from IP to selling CPUs.
If you already know Arm's licensing ecosystem, skip 0:49-2:37; first-time listeners can jump straight into the chip-building story at 2:37.