Fabless manufacturing

Based on Wikipedia: Fabless manufacturing

The Revolution That Made Your Smartphone Possible

In 1969, a small group of engineers did something that seemed crazy at the time: they started a company to design computer chips without owning a single machine to make them. This was like opening a restaurant without a kitchen, or launching an airline without planes. The semiconductor industry thought they were nuts.

They were actually pioneers of one of the most important business model innovations in technology history.

Today, the device you're using to read or listen to this almost certainly contains chips designed by companies that have never manufactured a single transistor. Apple, Nvidia, Qualcomm, AMD—these household names don't make their own chips. They design them, then hand off the blueprints to someone else to build. This approach is called fabless manufacturing, and it's quietly become the dominant model in the semiconductor industry.

Why Making Chips Used to Mean Making Chips

To understand why fabless manufacturing was revolutionary, you need to understand what "fab" means. A fab—short for fabrication facility—is the factory where semiconductor chips are physically manufactured. These aren't ordinary factories.

A modern semiconductor fab costs somewhere between ten and twenty billion dollars to build. That's not a typo. A single factory can cost more than the annual gross domestic product of some small countries. The equipment inside requires air thousands of times cleaner than a hospital operating room. Workers wear head-to-toe "bunny suits" because a single speck of dust can ruin thousands of chips. The manufacturing process involves hundreds of steps, some using light with wavelengths shorter than anything found in nature.

Before the 1980s, if you wanted to be in the semiconductor business, you had to own one of these factories. Companies like Intel, Texas Instruments, and Motorola were what the industry calls "integrated device manufacturers," or IDMs for short. They did everything in-house: designed the chips, manufactured them, assembled them, and tested them. This vertical integration made sense when the technology was simpler and the capital requirements were merely enormous rather than astronomical.

But this model created a massive barrier to entry. Want to start a chip company with a clever new design? First, raise several billion dollars for a factory.

The Spy Satellite Connection

The unlikely origin of fabless manufacturing involves Cold War espionage satellites and a defense contractor's reluctance to push the boundaries of physics.

In the late 1960s, Control Data Corporation needed custom chips for the Computer 469, a one-pound aerospace computer that would eventually fly on classified "Spy in the Sky" satellites. They hired General Instrument Microelectronics—known as GIM—to develop the processors. The chips worked beautifully, running at five megahertz, which was cutting-edge speed for the era.

Then Control Data wanted the next generation.

GIM looked at the specifications and essentially said: this is too hard, we're not doing it. The technical challenges were beyond what they wanted to tackle. But the engineers who had actually built the first chips knew it was possible. Control Data encouraged them to leave GIM and start their own company.

In 1969, LSI Computer Systems was born. The "LSI" stood for Large Scale Integration, referring to the density of circuits they could pack onto a chip. But here's the crucial part: they had no factory. They were a design company that would hire existing fabs to manufacture their chips.

This required solving problems nobody had solved before. How do you ensure quality when you don't control the manufacturing? How do you get a factory certified to make chips for satellites when it's not your factory? LSI Computer Systems had to develop entire new processes for monitoring production, inspecting wafers, and managing assembly at facilities they didn't own.

The chips they made—designated LSI0101 through LSI0105—were marvels of engineering. They operated at five megahertz while consuming minimal power, packaged in compact forty-pin metal cases. These weren't just chips; they were proof that design and manufacturing could be separated.

The Foundry Revolution

For about two decades after LSI Computer Systems, fabless companies existed but remained relatively rare. They relied on excess capacity at integrated device manufacturers—essentially renting time on other companies' production lines. This worked, but it had problems. The IDMs were also competitors. Why would Intel help a potential rival make better chips?

The real transformation came in 1987, when a former Texas Instruments executive named Morris Chang founded Taiwan Semiconductor Manufacturing Corporation, better known as TSMC. Chang's insight was simple but transformative: create a company that only manufactures chips, never designs them.

This business model is called a "pure-play foundry." TSMC would be the kitchen that cooked whatever recipes its customers brought. It would never compete with them by serving its own dishes. This neutrality was crucial. Suddenly, chip designers could outsource manufacturing without worrying that their foundry partner might steal their ideas.

TSMC became the cornerstone of the fabless model. Dozens, then hundreds, then thousands of chip design companies emerged, freed from the burden of building their own factories. Innovation exploded. If you had a good idea for a chip, you could focus entirely on the design while TSMC handled the manufacturing.

The Economics of Specialization

Why does separating design from manufacturing work so well?

Consider the economics. Building a leading-edge fab costs around fifteen billion dollars and takes three to five years. The equipment becomes obsolete within a few generations. The process technology—the specific recipes for etching circuits onto silicon—requires massive ongoing research and development investment. Only a handful of companies in the world can afford to stay at the cutting edge.

Now imagine you're a startup with a brilliant idea for a new type of artificial intelligence chip. Under the old model, your business plan would need to include "Step One: Raise fifteen billion dollars." Under the fabless model, you can design your chip, send the blueprints to TSMC or another foundry, and receive finished chips a few months later. Your capital requirements drop by orders of magnitude.

This works because foundries achieve economies of scale that individual chip designers never could. TSMC manufactures chips for Apple, Nvidia, AMD, Qualcomm, and hundreds of other companies. Their factories run at high utilization rates, spreading the enormous fixed costs across massive production volumes. They can invest in research and development that no single customer could justify.

Meanwhile, fabless companies can pour all their resources into design innovation. They don't need manufacturing engineers or billions in capital equipment. They need clever architects and sophisticated software tools. This specialization makes both sides more efficient.

The Rise of an Industry

In 1994, the fabless model had grown large enough to warrant its own trade organization. Jodi Shelton, along with executives from about half a dozen fabless companies, founded the Fabless Semiconductor Association. Their goal was to promote and legitimize this business model globally. By 2007, the organization had grown so much—and become so international—that it renamed itself the Global Semiconductor Alliance.

The transition reflected a fundamental shift in the industry. Fabless wasn't a niche anymore. It was becoming the default.

Major integrated device manufacturers started converting to the fabless model entirely. Conexant, which made chips for communications equipment, shut down its fabs. Semtech followed. Even LSI Logic—despite having "Logic" in its name and a long history of manufacturing—eventually went fabless.

Today, some of the most valuable technology companies in the world design chips without manufacturing them. Apple designs the processors in iPhones, iPads, and Macs, but TSMC makes them. Nvidia designs the graphics processors powering artificial intelligence systems, but TSMC makes those too. AMD designs processors that compete with Intel, and—you guessed it—TSMC manufactures them.

The Hidden Geography of Chips

The fabless model has created an unusual geography of semiconductor production. Design happens primarily in the United States, where companies like Nvidia, Apple, and Qualcomm are headquartered. Manufacturing happens predominantly in Taiwan, South Korea, and increasingly mainland China.

This concentration has become a geopolitical concern. Taiwan produces over ninety percent of the world's most advanced chips. TSMC alone manufactures chips for nearly every major technology company. If something disrupted Taiwanese chip production—a natural disaster, a pandemic, or a military conflict—the effects would ripple through every industry that uses electronics. Which is to say, every industry.

The United States and European Union have both launched massive initiatives to bring more chip manufacturing back to their territories. Intel is building new fabs in Arizona and Ohio. TSMC is constructing plants in Arizona and Japan. Samsung is expanding in Texas. Billions of dollars in government subsidies are flowing into semiconductor manufacturing.

But catching up is extraordinarily difficult. Manufacturing expertise accumulates over decades. The supply chains for specialized chemicals, gases, and equipment have concentrated in Asia. Workforce training takes years. The fabless model created tremendous efficiency, but it also created dependencies that are now causing strategic anxiety.

The Giants of Fabless

The largest fabless chip companies have become some of the most valuable corporations on Earth. Nvidia, which designs graphics processors that turned out to be perfect for artificial intelligence, briefly became the world's most valuable company in 2024. Its market capitalization exceeded three trillion dollars.

Broadcom, which makes chips for networking and communications, Qualcomm, which dominates mobile phone processors, AMD, which competes with Intel in personal computers and servers, and MediaTek, a Taiwanese company that supplies chips for budget smartphones—these are the giants of fabless. Each does billions of dollars in annual revenue while owning zero manufacturing capacity.

The model has enabled specialization that would have been impossible in the integrated era. There are now fabless companies focused on nothing but Bluetooth chips. Others specialize in power management, or wireless charging, or ultra-low-power processors for smartwatches. The barrier to entry dropped from "build a fifteen billion dollar factory" to "hire some smart engineers and rent time at a foundry."

The Opposite Model

Not everyone went fabless. Intel famously insisted for decades that designing and manufacturing chips in-house gave it crucial advantages. When you control the factory, the reasoning went, you can optimize your designs for your specific manufacturing process. You can iterate faster. You can keep your secrets better.

For a long time, this worked. Intel's manufacturing was so far ahead that competitors couldn't catch up. But maintaining that lead required ever-increasing investment. Eventually, TSMC's scale advantages became overwhelming. By the early 2020s, Apple and AMD chips manufactured at TSMC were outperforming Intel's in-house designs. Intel announced it would start using TSMC for some of its chips—an acknowledgment that the pure integrated model was no longer sustainable.

There's also a hybrid model. Companies like Samsung manufacture chips for themselves and for external customers. They compete with TSMC as foundries while also competing with fabless companies as designers. This creates complicated dynamics where Samsung might manufacture chips for a company it's also competing against.

An Alternative Approach: ARM's Model

ARM Holdings took the fabless concept even further. They don't just avoid manufacturing—they avoid making complete chip designs at all. Instead, ARM designs the fundamental architecture of processors and licenses it to other companies.

Think of it like this: Intel designs and manufactures complete chips. Nvidia designs complete chips and has them manufactured by TSMC. ARM designs the blueprints for the core of a chip and sells those blueprints to companies like Apple, Qualcomm, and Samsung, who then add their own customizations and have the final designs manufactured at foundries.

Nearly every smartphone processor in the world uses ARM's architecture. Apple's M-series chips in MacBooks use ARM. Amazon's server chips use ARM. This ultra-fabless model—licensing intellectual property rather than selling physical chips—has made ARM's architecture arguably the most important in computing, even though ARM itself never manufactures or even finishes designing a single chip.

What Fabless Manufacturing Means for Innovation

The fabless revolution fundamentally changed how innovation happens in semiconductors. When the barrier to entry was a multi-billion-dollar factory, only large corporations could play. Now, a startup with a clever idea and a few million dollars in venture capital can design a chip, have TSMC manufacture it, and potentially disrupt an entire market.

This democratization accelerated the pace of innovation dramatically. New types of chips emerged for specific applications that would never have justified their own factories: chips optimized for machine learning, chips for cryptocurrency mining, chips for autonomous vehicles, chips for smart home devices. The explosion of specialized silicon that powers modern technology would have been impossible under the old integrated model.

The tradeoff is dependency. The fabless model works only if foundries exist and have capacity. When demand for chips surged during the pandemic—as everyone suddenly needed laptops and data centers for remote work—the foundries couldn't keep up. Car manufacturers couldn't get the chips they needed. Consumer electronics faced shortages. The efficiency of concentration became the vulnerability of dependency.

Still, there's no going back. The economics of semiconductor manufacturing have made the fabless model not just attractive but necessary for most companies. Even Intel, the last great holdout, is adapting to a world where the best chips might not come from the company that manufactured them.

Those engineers in 1969, designing spy satellite chips without a factory to their name, started something bigger than they could have imagined. They proved that in the semiconductor industry, ideas could be separated from the silicon that embodied them. That insight—that design and manufacturing are different businesses—reshaped an entire industry and enabled the computing revolution we now take for granted.