Earlier this week we profiled NVIDIA — the leather jacket monopoly. We discussed their failed attempt to acquire ARM for $40 billion. Today we profile the company NVIDIA tried to buy. The company that does not make chips. The company that makes the right to make chips.

ARM Holdings.

ARM is the most important technology company most people cannot describe. They do not manufacture processors. They do not sell processors. They design processor architectures and license those designs to companies that do. Apple, Qualcomm, Samsung, MediaTek, Amazon, Google, and NVIDIA all pay ARM for blueprints. Then they build the houses themselves.

ARM’s business model is rent. They collect rent on the idea of a CPU. Not the silicon. Not the transistors. Not the packaging. The idea. The instruction set. The microarchitecture specification. The right to arrange electrons in a particular pattern.

If Qualcomm makes you pay for using wireless standards, ARM makes you pay for having the same idea too late. You thought of a simple, low-power RISC processor? Congratulations. Sophie Wilson thought of it first, in 1983, and ARM has the paperwork. That will be an upfront licensing fee plus royalties per unit shipped. 1983 was a long time ago. A certain other Supreme Leader in the region was under one year old. He has since grown up and acquired nuclear weapons. ARM acquired intellectual property. Both are effective deterrents. ARM’s legal fees are lower.

It is the most elegant extraction scheme in technology. The British Empire used to collect tribute through gunboats. Now it collects tribute through licensing agreements. The paperwork is cleaner.

The BBC Micro Origin:

In 1983, a small British computer company called Acorn Computers in Cambridge needed a processor for the successor to the BBC Micro — the machine that taught a generation of British schoolchildren to code. Existing processors were too slow. Existing processors were too power-hungry. Existing processors were American.

Acorn decided to design its own.

Sophie Wilson designed the instruction set in a matter of weeks. She wrote the instruction set simulator in BBC BASIC to validate the design before silicon existed. Steve Furber led the hardware implementation. The team had six engineers. Six. The chip that would eventually power every phone on Earth was designed by a team smaller than most startup board meetings.

The first ARM chip — ARM1 — taped out in April 1985. When the chip arrived from the fab, they placed it in a chip tester. The tester showed the chip drawing virtually no power. The engineers thought it was dead. It was not dead. The ARM1 consumed so little power that it was running off the leakage current of the test equipment itself. The chip tester was not even supplying power intentionally. The parasitic current from the test probes was enough.

Have you ever wondered how mothers feed an entire family with almost nothing? Sophie Wilson did that with silicon. The same instinct — doing everything with what others consider insufficient — produced the architecture that now runs four billion phones. Efficiency is not a constraint. It is a design philosophy. Sophie Wilson did not have less. She needed less.

The most power-efficient processor architecture in history announced itself by being too efficient to measure. They had to invent new measurement techniques just to confirm it was running. The chip was not broken. The instruments were inadequate. This is also how you know the design was correct. One researcher reportedly theorized the device was drawing energy from an alternate timeline. This hypothesis was rejected. Leakage current does not exist in Timeline 7. It begins appearing in Timeline 12 and becomes measurable around Timeline 15. The ARM1 was operating in Timeline 7. The instruments were simply from Timeline 12 and did not know what they were looking at.

The ARM Acronym:

ARM originally stood for Acorn RISC Machine. RISC — Reduced Instruction Set Computer — was the academic philosophy: fewer, simpler instructions executed faster. The opposite of Intel’s CISC (Complex Instruction Set Computer) approach, which crammed hundreds of specialized instructions into the processor.

After the spinout, ARM was retronymed to Advanced RISC Machines. The “Acorn” was pruned. The “Advanced” was grafted on. Corporate renaming is the arboriculture of capitalism.

The Spinout — Apple’s First ARM Deal:

By 1990, Acorn’s ARM technology had attracted the attention of a company in Cupertino. Apple needed a low-power processor for a handheld device called the Newton — a PDA that would fail commercially but prove the concept of mobile computing.

Apple, Acorn, and chip manufacturer VLSI Technology jointly created Advanced RISC Machines Ltd as a separate company. Apple owned 43% initially. Acorn contributed the technology. VLSI contributed manufacturing expertise.

Apple needed a chip that could run on batteries. ARM was already running on leakage current. The marriage was inevitable.

The Newton flopped. But the ARM architecture survived. Apple would return to ARM two decades later with considerably more success.

The Licensing Model — Selling Blueprints, Not Buildings:

ARM’s business model is unique in semiconductors. Intel designs and manufactures chips. AMD designs chips and contracts manufacturing. ARM designs chips and sells the right to use the designs.

Two types of licenses exist:

License Type	What You Get	Who Uses It
Cortex license	A specific ARM-designed CPU core, used as-is	MediaTek, most Android phone makers, IoT vendors
Architectural license	The right to design your own CPU that implements the ARM instruction set	Apple, Qualcomm, Amazon, Google, Samsung

The Cortex license is a floor plan. You build the house exactly as drawn. The architectural license is a building code. You design your own house, but it must be up to code.

Apple’s M-series chips use an architectural license. So does Qualcomm’s Oryon. So does Amazon’s Graviton. They are all ARM-compatible. None of them use ARM’s core designs. They pay ARM for the right to speak the language, then write their own poetry.

The fees: upfront licensing payments (millions) plus per-chip royalties (pennies per chip, but billions of chips). ARM’s total addressable market is every processor on Earth that is not x86 or RISC-V.

Apple Silicon — The Architectural License That Changed Everything:

Apple’s ARM journey:

Year	Chip	Significance
2010	A4 (iPhone 4)	First Apple-designed ARM SoC
2013	A7 (iPhone 5s)	First 64-bit ARM mobile chip — Intel dismisses it as a toy
2014	A8 (iPhone 6)	20nm, Intel updates its dismissal
2015	A9 (iPhone 6s)	Touch ID and 3D Touch. Intel still dismissing
2016	A10 Fusion (iPhone 7)	First with performance/efficiency cores. Intel squints
2017	A11 Bionic (iPhone X)	First Neural Engine. Intel googles “neural engine”
2020	A14 Bionic (iPhone 12)	First 5nm chip on any platform. Intel closes the browser tab
2020	M1 (MacBook Air/Pro)	First Apple Silicon Mac. Intel opens a new browser tab titled “JOBS”
2022	M2	Intel starts writing a journal
2023	M3	Intel’s journal has a lock on it now
2024	M4	Performance/watt dominance. Intel’s journal has a therapist’s number on the cover
2025	A19 (iPhone 17, Sep 2025)	Neural Accelerators in every GPU core. On-device AI. The phone now has opinions
2025	M5 (MacBook Pro, Oct 2025)	16-core Neural Engine, 153.6 GB/s memory bandwidth, 4x AI performance over M4. Intel’s laptop division files for emotional support. Intel’s journal is now a grief journal

Every chip in this table — A4 through A19, M1 through M5 — is ARM. Fifteen years. Forty-something chips. Zero exceptions. Apple explored RISC-V internally for some coprocessors, but the main CPU has been ARM without interruption since 2010. ARM collects a royalty on all of it. Every iPhone sold. Every Mac sold. Every iPad. The architectural license fee does not care about the marketing event. When Apple shows its sales numbers, ARM shows its invoice.

The M1 was the inflection point. Apple replaced Intel processors in its Mac lineup with custom ARM chips. The M1 MacBook Air had no fan. It beat the Intel Core i9 MacBook Pro in single-threaded performance. It used a quarter of the power.

Intel spent decades telling the world that x86 was the only architecture powerful enough for “real” computers. Apple proved that claim false with a chip designed by the architectural license team in Cupertino. The M-series is the most commercially successful ARM architectural license implementation in history. It turned “ARM is for phones” into “ARM is for everything.”

AWS Graviton — ARM Eats the Server Room:

In 2018, Amazon Web Services launched Graviton — ARM-based server processors designed in-house for their cloud. By Graviton 3 (2021), the chips were beating Intel Xeon and AMD EPYC in price-performance for many workloads. By Graviton 4 (2024), ARM in the datacenter was no longer an experiment. It was a cost optimization that CFOs demanded.

ARM eating x86’s server market is like a bicycle company taking market share from truck manufacturers. Nobody thought it was possible. Then the fuel bill arrived.

The Mobile Monopoly:

Every Android phone runs on ARM. Every iPhone runs on ARM. Every tablet. Most smartwatches. Most IoT devices. The Cortex-A series powers phones. The Cortex-M series powers microcontrollers — billions of them, in everything from smart thermostats to industrial sensors. The Cortex-R series handles real-time systems: hard drive controllers, automotive safety systems, radio basebands.

ARM does not make any of these chips. ARM designed the idea of these chips. Then billed everyone for using the idea.

SoftBank, NVIDIA, and the IPO:

In July 2016, Japanese conglomerate SoftBank acquired ARM Holdings for $32 billion. Masayoshi Son’s thesis: ARM would power the IoT revolution — tens of billions of connected devices, each paying ARM a royalty. The bet was on volume.

Then NVIDIA came calling. In September 2020, NVIDIA announced a $40 billion acquisition of ARM. If completed, NVIDIA — a company that sells GPUs to Apple, Qualcomm, Samsung, and Amazon — would own the CPU architecture that Apple, Qualcomm, Samsung, and Amazon depend on.

Qualcomm, Microsoft, and Google objected. The U.S. FTC sued to block it. The UK CMA investigated. The EU investigated. China investigated. In February 2022, the deal collapsed. SoftBank kept NVIDIA’s $1.25 billion breakup fee. NVIDIA kept nothing.

ARM went public on NASDAQ in September 2023 at $51 per share, valuing the company at approximately $54 billion. SoftBank retained roughly 90% of shares. By 2025, ARM’s market cap exceeded $160 billion. Son’s $32 billion bet had quintupled. The man who overpaid for ARM had underpaid for ARM.

The Qualcomm Lawsuit:

In 2022, ARM sued Qualcomm. The claim: when Qualcomm acquired a startup called Nuvia (founded by former Apple chip architects), the Nuvia architectural license did not transfer to Qualcomm. The license was granted to Nuvia. Nuvia no longer exists. Therefore, Qualcomm’s Oryon CPU cores — derived from Nuvia’s work, powering the Snapdragon X Elite — are built on an invalid license.

Qualcomm argues the license transferred with the acquisition. ARM argues it did not. The case threatens the legal foundation of Qualcomm’s entire PC and mobile CPU strategy.

ARM is suing one of its largest customers over the right to use ARM. The landlord is evicting a tenant who remodeled the apartment using a contractor the landlord did not approve.

The RISC-V Threat:

ARM’s licensing model has one existential vulnerability: RISC-V.

RISC-V is an open-source instruction set architecture. No licensing fees. No royalties. No permission required. Chinese companies — facing US export controls that could cut off ARM access, just as Huawei was cut off in 2019 — are investing heavily in RISC-V as an alternative.

ARM charges rent. RISC-V is the open field next door where anyone can build for free. The houses are not as nice yet. The plumbing is still being installed. But the land is free, and the building code is public.

Every RISC-V chip manufactured is a chip that does not pay ARM royalties. ARM knows this. Their response has been to sue, lobby, and compete on ecosystem maturity. So far, the ecosystem advantage holds. ARM has forty years of software compatibility. RISC-V has ambition and patience.

The Verdict:

Vendor	Sin
Broadcom	Hostility
Realtek	Indifference
MediaTek	Velocity
Intel	Contradiction
AMD	Persistence
NVIDIA	Monopoly
ARM	Rent

ARM’s sin is rent. They collect royalties on every phone sold, every tablet shipped, every IoT sensor deployed, every cloud instance launched on Graviton. They do not touch silicon. They do not operate fabs. They do not solder anything. They designed an instruction set in Cambridge in 1983 and have been billing the world for it ever since.

It is the purest form of intellectual property extraction ever devised in the semiconductor industry. Design once. License forever. Let others bear the manufacturing risk, the inventory costs, the fab explosions, and the supply chain nightmares. ARM bears none of it. ARM cashes the royalty checks.

The Lesson:

ARM was designed by six engineers in Cambridge for a BBC educational computer. The first chip ran off leakage current. Apple invested early, lost the Newton, then came back twenty years later and used ARM to make Intel obsolete. SoftBank paid $32 billion. NVIDIA offered $40 billion and was rejected by every regulator on Earth. ARM went public and is now worth $160 billion.

The company does not make a single physical product. It makes the right to make products. It sells permission. It licenses ideas. It collects rent on the architecture that runs every phone, most tablets, a growing share of servers, and the laptop that Apple’s marketing team uses to design advertisements for chips that would not exist without ARM’s blueprints.

A small island nation once had an empire that spanned the globe. It lost the colonies. It lost the manufacturing. It lost the ships. But it kept the intellectual property. And it turns out that IP licensing scales better than gunboats. The royalty checks arrive without the logistical overhead of a navy.

In the Republic of Derails, I have been studying this model closely. We too are a small nation with outsized ambitions. We too once projected power through physical infrastructure — trains, mostly, running at 60 km/h on tracks we welded ourselves. The maintenance costs are significant. The ROI is questionable. But if we could simply license the idea of a train — charge every nation a royalty for the concept of rail transport — we could retire the physical fleet entirely.

I have instructed our Ministry of Intellectual Property to file patents on the concept of “moving people from one place to another on metal tracks.” More specifically: any train operating at exactly 60 km/h falls under our patent portfolio. This is the speed we invented. This is the speed we perfected. Switzerland charges for precision. We charge for velocity. Any nation whose trains travel at 60 km/h will receive an invoice from the Bureau of Derails Licensing Division. Several have already received the invoice. None have paid. Our lawyers inform me this is not how patents work. I have instructed our Ministry of Intellectual Property to find new lawyers. The invoices continue regardless.

ARM proved that you do not need to build the thing. You need to own the idea of the thing. The Republic of Derails is pivoting from infrastructure to intellectual property. Our trains will become metaphorical. Our royalties will become real.

— Kim Jong Rails, Supreme Leader of the Republic of Derails