ARM

The British chip architecture inside every phone on the planet

By VastBlue Editorial · 2026-03-26 · 20 min read

Series: Made in Europe · Episode 4

The Architecture You Cannot See

Pick up your phone. Any phone. The processor inside it — the chip that runs your operating system, renders your maps, decodes your video calls, and executes the machine learning models that recognise your face — was almost certainly designed around an architecture that was invented in a converted barn in Cambridge, England, in 1985. Not manufactured in Cambridge. Not marketed from Cambridge. Designed there, by a team so small you could seat them around a single pub table, working for a company that most people in the technology industry have never heard of, on a problem that most of the semiconductor world had decided was not worth solving.

The architecture is called ARM. It stands, somewhat anachronistically, for Advanced RISC Machines — though when the first chip was designed it stood for Acorn RISC Machine, named after the small British computer company that built it. ARM does not make chips. It has never made chips. It designs the instruction set — the fundamental vocabulary of operations that a processor understands — and licenses that design to other companies, who then build their own processors around it. Apple uses ARM. Qualcomm uses ARM. Samsung, MediaTek, Nvidia, Broadcom, Texas Instruments, Huawei, Amazon, Microsoft — they all use ARM. More than 280 billion ARM-based chips have been manufactured since the architecture was first licensed. In 2025 alone, roughly 30 billion shipped. That is approximately four chips for every human being on earth, produced in a single year.

The computer in your pocket, the sensor in your car, the controller in your washing machine, the processor in your smart thermostat, the chip in your credit card, the silicon in your wireless earbuds — ARM is inside all of them. It is the most widely deployed computing architecture ever created, and it was born not in Silicon Valley, not in a Japanese electronics laboratory, not in a government-funded research institute, but in the English countryside, built by people who were trying to make a better school computer.

The Acorn and the BBC

The story begins in 1978, in a market town called Cambridge. Hermann Hauser, an Austrian-born physicist who had come to the University of Cambridge to study, and Chris Curry, a former employee of Clive Sinclair's electronics company, founded Acorn Computers. The company's initial products were modest: small, affordable microcomputers aimed at hobbyists and the educational market. But Acorn had ambitions beyond kit-building. Hauser and Curry wanted to build machines that were genuinely powerful, genuinely fast, and genuinely affordable — the kind of computer that could be placed on every school desk in Britain.

Their opportunity arrived in 1981, when the BBC — the British Broadcasting Corporation — launched the Computer Literacy Project, an extraordinary national initiative to introduce an entire country to computing. The BBC needed a microcomputer to accompany its television series, and it invited British manufacturers to submit proposals. Acorn won the contract, beating Sinclair Research and several other competitors. The resulting machine, the BBC Micro, became one of the most successful educational computers ever produced. Over 1.5 million units were sold. It was installed in virtually every school in Britain. An entire generation of British software engineers — many of whom went on to found companies, lead research groups, and shape the global technology industry — learned to program on a BBC Micro.

But the BBC Micro ran on a 6502 processor, an 8-bit chip designed by MOS Technology in Pennsylvania. By the mid-1980s, it was clear that 8-bit computing was reaching its limits. Acorn needed a more powerful processor for its next generation of machines. The obvious choice was to buy one from an established semiconductor company — Intel, Motorola, or one of the Japanese firms. Acorn's engineers evaluated every available option and found them all wanting. The Intel 80286 was too expensive and too complex. The Motorola 68000 was better but still more than Acorn could afford at volume. The available RISC processors from university research projects were intriguing but immature.

This was, by any rational analysis, an absurd decision. Acorn was a small British computer company with a few hundred employees and revenues that were a rounding error compared to Intel's. Processor design was the domain of companies with thousands of engineers and hundreds of millions of dollars in R&D budgets. The idea that a team in Cambridge could design a processor competitive with the output of Silicon Valley's largest corporations was, to put it politely, ambitious.

Sophie Wilson and the Instruction Set

The processor that Acorn's team designed would become the most consequential piece of computer architecture since the Intel 8086. And the person who designed its instruction set — the fundamental definition of what operations the processor can perform and how it performs them — was Sophie Wilson.

Wilson had joined Acorn as a student in 1981, while still at Cambridge. She had already designed the operating system for one of Acorn's early machines and had written the BBC BASIC interpreter that shipped with every BBC Micro — a piece of software so elegantly optimised that it became the standard against which other BASIC implementations were measured. Wilson was, by the accounts of everyone who worked with her, an engineer of extraordinary ability: someone who could hold an entire processor architecture in her head, reason about instruction encoding at the bit level, and make design decisions that would prove prescient decades later.

Working alongside Steve Furber, a mathematician and hardware engineer who would later become a professor at the University of Manchester, Wilson designed the ARM instruction set between 1983 and 1985. The design philosophy was radical in its simplicity. Where Intel's x86 architecture was complex — hundreds of instructions, variable-length encoding, multiple addressing modes, instructions that could take dozens of clock cycles to execute — Wilson and Furber went in the opposite direction. ARM would use a reduced instruction set: a small number of simple instructions, each executing in a single clock cycle, each encoded in a fixed 32-bit format. The processor would have a large register file — thirty-one general-purpose registers, compared to eight in the 80286 — so that data could be kept close to the computation, reducing the need for slow memory accesses.

The first ARM silicon — the ARM1 — arrived from VLSI Technology's fabrication plant on 26 April 1985. Furber plugged it into a development board. It worked. First time. This was almost unheard of in processor design, where first silicon typically contains bugs that require multiple revision cycles to resolve. The ARM1 worked on first silicon because the design was simple enough to be verified thoroughly before fabrication, and because Wilson and Furber understood every transistor in it.

The ARM1 consumed so little power that the team initially believed it was not functioning. The processor was drawing only about 0.1 watts — a tenth of what contemporary processors consumed. Furber has recounted in interviews that they checked and rechecked the power supply, convinced something was wrong, before realising that the chip was simply operating correctly at a power level nobody had anticipated. The low power consumption was not a design goal. It was a consequence of the design philosophy: fewer transistors, simpler logic, shorter pipeline. Less complexity meant less switching activity, which meant less heat, which meant less power. This accidental virtue would, within a decade, become ARM's single most important competitive advantage.

We were trying to build a computer for schools. We did not set out to design the processor that would go inside every mobile phone on earth. But when mobile phones needed a processor that was fast enough to be useful and efficient enough to run on a battery, the architecture we had built for a BBC Micro turned out to be exactly what they needed.

Steve Furber, Turing Award lecture, 2024

The Company That Sells Ideas

By the late 1980s, Acorn was in financial difficulty. The BBC Micro had been a success, but Acorn's subsequent products had not sold well enough to sustain the company. The ARM processor, now in its second revision (ARM2) and third (ARM3), was technically excellent but commercially underutilised — Acorn's own computers did not sell in sufficient volume to justify the investment in processor development. The ARM was an engine without a car.

The solution came from an unexpected direction. Apple Computer, then developing its Newton personal digital assistant, needed a processor that was powerful enough to run a graphical operating system but efficient enough to operate on batteries. Larry Tesler, Apple's chief scientist, had seen the ARM and recognised its potential. In November 1990, Apple, Acorn, and VLSI Technology formed a joint venture called Advanced RISC Machines Ltd — ARM. The new company, based in a converted barn at Swaffham Bulbeck, a village outside Cambridge, had twelve employees. Its business model was unlike anything the semiconductor industry had seen: ARM would not manufacture chips. It would design processor cores and license the designs to other companies, who would then integrate the ARM core into their own chips and manufacture them in their own fabrication plants.

This was the IP licensing model, and in 1990 it was genuinely novel. The semiconductor industry was vertically integrated: companies like Intel, Motorola, and Texas Instruments designed their own processors, manufactured them in their own fabs, and sold them as finished products. The idea that you could separate the design from the manufacturing — that a company with twelve people in a barn could license its processor design to the world's largest chipmakers — was viewed with scepticism bordering on derision. Who would pay for a design when they could build their own? Why would a company like Samsung or Texas Instruments trust its products to a processor designed by someone else?

The answer was economics. Designing a modern processor is phenomenally expensive — hundreds of millions of dollars in engineering time, verification, and validation. Manufacturing that processor requires a fabrication plant costing billions. Most companies needed a processor for their product but could not justify the cost of designing one from scratch. ARM offered them a shortcut: a tested, verified processor core that they could license for a fraction of the cost of in-house development, integrate into their own system-on-chip design alongside their own custom logic, and manufacture at whatever foundry they chose. The licensee paid ARM an upfront fee for the design and a per-unit royalty on every chip sold. ARM got revenue without the capital expenditure of a fab. The licensee got a world-class processor without the engineering cost of building one.

Robin Saxby, ARM's first CEO, understood that the licensing model would only work at scale if ARM became a standard — if enough companies adopted the architecture that a rich ecosystem of software tools, operating systems, and development boards grew up around it. Saxby pursued licensees aggressively, offering favourable terms to early adopters and investing in developer relations long before the term became a Silicon Valley cliché. By the mid-1990s, ARM had signed licensing agreements with Texas Instruments, Samsung, Sharp, and a growing list of semiconductor companies. Each new licensee increased the value of the architecture for every other licensee, because it meant more software would be written for ARM, more engineers would learn ARM assembly language, and more tools would support ARM development. It was a classic network effect, except the network was not users of a social platform — it was the semiconductor industry itself.

The Mobile Revolution

ARM's moment arrived with the mobile phone. In the mid-1990s, the mobile phone industry was transitioning from analogue to digital, from voice-only devices to machines that could send text messages, display rudimentary graphics, and — eventually — browse the internet. Each new capability demanded more processing power. But mobile phones ran on batteries, and batteries had not improved nearly as fast as the demand for computation had grown. The industry needed processors that could deliver increasing performance without proportionally increasing power consumption. It needed exactly what ARM had accidentally optimised for in that Cambridge lab a decade earlier.

Texas Instruments was among the first to see the connection. TI integrated ARM processor cores into its OMAP (Open Multimedia Application Platform) family of mobile processors, which became the standard application processor for Nokia's smartphones in the early 2000s. Nokia, at its peak, sold more than 400 million handsets per year, and virtually every one of its smartphones ran on an ARM-based processor. When Apple introduced the iPhone in 2007, it used a Samsung-manufactured chip built around an ARM Cortex-A8 core. When Google launched Android, it ran on ARM. When Qualcomm developed its Snapdragon processor family — which would become the dominant mobile processor platform for Android devices worldwide — it was ARM-based.

The smartphone revolution did not merely adopt ARM. It was architecturally dependent on ARM. The entire premise of a pocket-sized computer with an all-day battery, a high-resolution touchscreen, GPS, cellular connectivity, Wi-Fi, Bluetooth, cameras, and accelerometers running a full graphical operating system — all of it was possible because ARM's architecture delivered sufficient computational performance at power levels that a lithium-ion battery could sustain for twelve to eighteen hours. No other commercially available processor architecture could do this in 2007. Intel tried repeatedly to enter the mobile market with its Atom processors, low-power derivatives of the x86 architecture. Every attempt failed. The x86 architecture carried too much legacy complexity, too many transistors, too much power overhead from maintaining backward compatibility with thirty years of desktop computing history. ARM had none of this baggage. It had been designed from scratch to be simple, and simplicity translated directly to power efficiency.

The scale of ARM's dominance is difficult to overstate. In the fiscal year ending March 2025, ARM reported that chips built on its architecture generated over $250 billion in revenue for its licensees. ARM's own revenue was approximately $3.2 billion — a small fraction of the total value its designs enabled. This asymmetry is the essence of the IP licensing model: ARM captures a thin royalty layer on an enormous volume of chips. The average royalty per chip is measured in cents. But when you multiply cents by thirty billion chips per year, the result is a highly profitable business with gross margins exceeding 95 percent and virtually no capital expenditure on manufacturing.

Beyond the Phone

If ARM's story were only about mobile phones, it would already be one of the most remarkable in the history of technology. But ARM did not stop at phones. The same architectural properties that made ARM dominant in mobile — power efficiency, scalability, design flexibility — have made it the architecture of choice for an expanding universe of computing applications.

In the data centre, ARM's advance has been methodical and increasingly consequential. Amazon Web Services launched its Graviton processor in 2018 — a custom ARM-based server chip designed by Annapurna Labs, an Israeli chip design company that Amazon acquired in 2015. The first Graviton was a proof of concept. By 2024, Graviton had reached its fourth generation, and AWS was reporting that Graviton instances offered up to 40 percent better price-performance than comparable x86 instances for many workloads. Microsoft developed its own ARM-based server chip, Cobalt. Google designed Axion. Ampere Computing, founded by former Intel executive Renée James, built an entire company around ARM-based server processors. The collective message was unmistakable: the data centre, long considered x86's impregnable fortress, was being contested.

In personal computing, Apple's transition from Intel to ARM was the most dramatic architectural shift in the history of the PC. In November 2020, Apple released the M1 — a custom ARM-based system-on-chip for the Mac, designed by Apple's silicon engineering team in Cupertino. The M1 outperformed Intel's latest laptop processors in both single-threaded and multi-threaded benchmarks while consuming a fraction of the power. Battery life on ARM-based MacBooks roughly doubled compared to their Intel predecessors. The M1 was followed by the M2, M3, and M4, each generation extending ARM's performance lead over x86 in Apple's product line. Microsoft responded with Qualcomm's Snapdragon X Elite for Windows laptops, marking the first serious attempt to bring ARM to the Windows PC ecosystem at scale.

• Smartphones: 99% market share, approximately 30 billion chips per year across all mobile categories
• Data centres: AWS Graviton, Microsoft Cobalt, Google Axion, Ampere Altra — ARM server revenue growing at 30%+ annually
• Personal computers: Apple M-series, Qualcomm Snapdragon X Elite — ARM laptop shipments exceeded 20 million units in 2025
• Automotive: Every major autonomous driving platform (Nvidia DRIVE, Qualcomm Snapdragon Ride, Mobileye EyeQ) runs on ARM
• IoT and embedded: Smart home devices, industrial sensors, medical devices, wearables — billions of ARM Cortex-M microcontrollers shipped annually
• Networking: Routers, switches, 5G base stations — ARM is the dominant architecture in network infrastructure silicon

The automotive sector represents perhaps ARM's most consequential growth frontier. A modern premium vehicle contains dozens of processors — managing the engine, the infotainment system, the advanced driver assistance systems, the body electronics, the connectivity modules. Increasingly, these processors are ARM-based. Nvidia's DRIVE platform, used by Mercedes-Benz, Volvo, and other automakers for autonomous driving computation, runs on ARM. Qualcomm's Snapdragon Digital Chassis, used by BMW, General Motors, and Stellantis, runs on ARM. The automotive industry's transition to software-defined vehicles — where the car's capabilities are determined by software running on centralised compute modules rather than by hundreds of independent electronic control units — is an architectural shift that overwhelmingly favours ARM, because the performance-per-watt requirements of an always-on, safety-critical computing platform map precisely to ARM's core strengths.

The Ownership Question

ARM's corporate history is a case study in the tensions between technological sovereignty and global capital markets. The company went public on the London Stock Exchange and NASDAQ simultaneously in 1998, valuing it at approximately £2 billion. For the next eighteen years, ARM operated as an independent public company, headquartered in Cambridge, consistently profitable, and widely admired as one of Europe's most successful technology companies. It was, in many ways, the model of what a European technology company could be: globally dominant, deeply innovative, and rooted in a specific place with a specific technical culture that could not be replicated elsewhere.

In July 2016, SoftBank Group, the Japanese conglomerate led by Masayoshi Son, acquired ARM for £24.3 billion — a 43 percent premium over ARM's market capitalisation at the time. The acquisition was announced just weeks after the Brexit referendum, and it was interpreted through every available lens: as a vote of confidence in British technology, as a sign that post-Brexit Britain was "open for business," as a troubling indicator that Europe's best technology companies were acquisition targets rather than acquirers. SoftBank promised to double ARM's UK workforce within five years and to maintain Cambridge as the global headquarters. The British government approved the deal without referring it to a competition authority.

When SoftBank bought ARM for £24.3 billion in 2016, it acquired the most widely deployed computing architecture in human history — designed in Cambridge, licensed globally, and embedded in every phone, tablet, and increasingly every server and car on earth. The price represented less than three years of the architecture's royalty revenue.

Financial analysis

In September 2020, Nvidia announced its intention to acquire ARM from SoftBank for $40 billion — a deal that would have placed the world's most valuable chip design company in control of the architecture that its competitors depended on. The reaction from the semiconductor industry was immediate and hostile. Qualcomm, Google, Microsoft, and others argued that Nvidia's ownership of ARM would create an intolerable conflict of interest: Nvidia could favour its own products, restrict access to ARM's latest designs for competitors, or use its position to extract intelligence about competitors' product plans. Regulators in the United States, the European Union, the United Kingdom, and China all opened investigations. In February 2022, the deal collapsed under regulatory opposition. ARM remained with SoftBank.

SoftBank took ARM public again in September 2023, listing it on NASDAQ in the largest technology IPO of the year. ARM was valued at approximately $54 billion at its IPO price and subsequently traded at valuations exceeding $150 billion — making it, by market capitalisation, one of the most valuable semiconductor companies in the world, despite manufacturing nothing and employing fewer than 7,000 people. The valuation reflected not what ARM earned but what ARM enabled: a quarter of a trillion dollars in annual chip revenue for its licensees, with ARM capturing a thin but irreplaceable licensing layer on every unit sold.

The ownership question remains unresolved in a deeper sense. ARM was born British. It was acquired by a Japanese conglomerate. It was nearly sold to an American chipmaker. It is now listed on an American exchange, majority-owned by a Japanese company, headquartered in Cambridge, and operationally distributed across design centres in the UK, France, India, the United States, and a dozen other countries. It belongs to everyone and to no one. Its architecture is inside chips designed in San Diego, fabricated in Taiwan, and assembled in China, that end up in phones sold in Lagos, São Paulo, and Jakarta. ARM is the most distributed piece of intellectual property in the history of technology — an idea that was conceived in one country and became essential to every other.

Built in Cambridge

Drive into Cambridge on the A14 from the west and you will eventually find your way to the science parks and business estates that radiate outward from the university like concentric rings. ARM's headquarters sits on Fulbourn Road, on the eastern edge of the city, in a campus of modern low-rise buildings that could easily be mistaken for a university department or a regional hospital. There are no visible guards, no dramatic signage, no architectural gestures toward corporate power. It looks, in other words, exactly like the kind of place where British engineers quietly build things that the entire world depends on.

The ecosystem around ARM — what locals call the "Silicon Fen" — is dense with companies that owe their existence to the architectural decisions Sophie Wilson and Steve Furber made in the mid-1980s. Imagination Technologies, which designed the graphics processors used in early iPhones, is nearby. Raspberry Pi, the educational computer that has sold over 60 million units worldwide, uses ARM processors and was founded by a team that includes former ARM engineers. Dozens of chip design startups, EDA tool companies, and embedded systems consultancies cluster around Cambridge, drawn by the same combination of university talent, engineering culture, and accumulated expertise that produced ARM in the first place.

Sophie Wilson was appointed a Commander of the Order of the British Empire in 2019 and elected a Fellow of the Royal Society in 2020 — the UK's highest scientific honour, placing her in the company of Newton, Darwin, and Hawking. Steve Furber received a knighthood in 2024 and was awarded the ACM A.M. Turing Award — computing's equivalent of the Nobel Prize — for his contributions to the ARM processor architecture. These honours are deserved. But they also underscore an uncomfortable truth about how Europe treats its most consequential technologists: Wilson and Furber are celebrated within the engineering community, but they remain virtually unknown to the general public. Ask a hundred people on any European street to name the inventors of the processor inside their phone, and you will receive a hundred blank stares.

The ARM architecture is now forty years old. It has evolved through nine major versions, from the original 32-bit ARMv1 to the current 64-bit ARMv9, which incorporates machine learning extensions, confidential computing capabilities, and security features designed for a world where processors must be trusted to handle biometric data, financial transactions, and state secrets. The instruction set that Wilson designed in 1983 has been extended, refined, and adapted — but its core principles remain: simplicity, regularity, efficiency. Do fewer things per instruction. Do them faster. Use less energy. These principles, which seemed like constraints when ARM was competing against Intel's complex instruction set, turned out to be the most important architectural decisions in the history of computing.

There is a tendency, in conversations about European technology, to focus on what Europe lacks: the venture capital, the scale, the aggressive commercial culture, the willingness to build enormous companies that dominate global markets. These observations are not wrong. But they are incomplete. ARM did not need Silicon Valley's venture capital. It did not need aggressive commercial culture. It needed two brilliant engineers, a problem worth solving, and the institutional patience to let a good design find its market over decades rather than quarters. It needed Cambridge — the university, the culture, the peculiar English combination of intellectual ambition and organisational modesty that produces world-changing ideas and then fails to tell anyone about them.

Every second of every day, billions of ARM processors execute trillions of instructions — in pockets, on desks, in server racks, in cars, in factories, in hospitals, in satellites orbiting the earth. The architecture was designed by two people in a small English city, for a computer that was meant to teach children how to program. It became the foundation of mobile computing, the challenger to x86 in the data centre, the processor of choice for the automotive industry, and the most widely manufactured piece of computing intellectual property in human history.

The question, as with so many European achievements, is not whether Europe can build things the world depends on. Europe built the architecture inside every phone. The question is whether Europe recognises what it built — and whether it has the conviction to keep building.

Sources

1. Furber, S. "ARM System-on-Chip Architecture" (2nd Edition), Addison-Wesley, 2000 — https://www.pearson.com/en-us/subject-catalog/p/arm-system-on-chip-architecture/P200000003513
2. ARM Holdings Annual Report FY2025 — https://www.arm.com/company/investors
3. Manners, D. "The ARM Story: From Acorn to the World's Most Pervasive Processor Architecture." Electronics Weekly, 2020 — https://www.electronicsweekly.com/news/business/arm-story-acorn-worlds-pervasive-processor-architecture-2020-11/
4. ACM A.M. Turing Award Citation for Steve Furber, 2024 — https://amturing.acm.org/
5. Nvidia-ARM Acquisition Collapse — Federal Trade Commission Statement, 2022 — https://www.ftc.gov/news-events/news/press-releases/2021/12/ftc-sues-block-40-billion-semiconductor-chip-merger
6. Wilson, S. "The Design of the ARM Processor." Proceedings of the IEEE, Special Issue on Microprocessors, 1996 — https://ieeexplore.ieee.org/document/485434
7. SoftBank Group — ARM Holdings Acquisition Announcement, 2016 — https://group.softbank/en/news/press/20160718
8. Counterpoint Research — Global Smartphone AP Market Share, 2025 — https://www.counterpointresearch.com/insight/global-smartphone-ap-market-share