The fabless model has dominated the chip manufacturing industry for decades. [Source: author]
I used to believe that the core competitive advantage of tech hardware giants like NVIDIA, AMD, and Apple lay in their secret know-how guarded as the crown jewels of their business, locked away in massive vaults, behind ten doors, five moats, and an army of hungry dogs. Surely, I thought, these companies would never let anyone else lay eyes on their most precious intellectual property.
Well… they do protect their secrets, but not in the way I imagined.
The reality is far more surprising as I found reading Chip War by Chris Miller: while these companies fiercely guard their chip designs, they entrust the actual manufacturing, so the physical creation of their silicon masterpieces, to outside specialists. Once again: the largest chip designers don’t manufacture their products. This counterintuitive approach, known as the fabless model, has quietly become the foundation of the modern semiconductor industry. Let’s explore why the world’s leading chip designers don’t build their own chips, and how this unexpected strategy drives innovation at a global scale.
What Does “Fabless” Mean?
Fabless refers to a business model where a semiconductor company focuses on chip design and innovation, while contracting out the production of those chips to third-party manufacturers (foundries - “fabs”). For example, NVIDIA invests its expertise in designing the architecture of a graphics processing unit (GPU) — how the billions of transistors and their interconnections should be layed out — then hands off the physical manufacturing to a partner. It’s similar to how Apple designs its iPhones but relies on specialist manufacturers like Foxconn for assembly. This separation wasn’t always present. In the early decades of microchips (1960s–1970s), most semiconductor firms were vertically integrated, designing and fabricating their own chips.
So why did the industry go fabless? The answer is specialization and cost driven by the Moore’s Law. Building and operating a cutting-edge chip fabrication plant is astronomically expensive — the newest TSMC fab is expected to cost around $49 billion. Moore’s Law, which states that the number of transistors on a chip doubles roughly every two years, drives rapid evolution in manufacturing processes. Each new generation requires new equipment, materials, and techniques — essentially a new fab. Only a handful of companies with enormous customer bases can afford such investments every couple of years.
In the fabless model, chip designers and chip manufacturers play very different roles. Chip designers focus on the brains of the operation: they create the architectural blueprint, develop intellectual property (IP), and build the software and drivers that make chips useful. They also handle system integration, ensuring the chip works seamlessly in real-world devices. On the other side, chip manufacturers take those designs and turn them into physical reality. Their responsibilities include silicon wafer fabrication, cutting-edge process technology research, and high-volume production in ultra-clean environments. They also perform rigorous quality control and testing to ensure every chip meets performance and reliability standards.
By outsourcing fabrication, chip design companies avoid the massive capital costs and can focus on innovation and research. This creates a virtuous cycle: fabless firms invest in design and new architectures, while specialized manufacturers (foundries) push the boundaries of fabrication technology, serving many clients and justifying regular upgrades.
Press enter or click to view image in full size
The division between chip designers and chip manufacturers in the fabless model. [Source: author]
The Rise of TSMC
The fabless revolution relies on solid manufacturers to build the chips. This is where TSMC (Taiwan Semiconductor Manufacturing Company), by far the world’s largest chip manufacturer, comes in. In the 1980s, engineer Morris Chang pioneered the concept of a pure-play foundry: a company that manufactures chips for others but does not design its own. TSMC, founded in 1987, became a “silicon workshop” for fabless companies, providing state-of-the-art manufacturing as a service. Crucially, TSMC does not compete with its clients - it only manufactures chips for others, it doesn’t sell its own branded chips. This neutrality encouraged dozens of companies to trust TSMC with their designs. In contrast a company looking to manufacture their cutting-edge chips would always choose TSMC over, say, Intel (one of a few giants still both designing and manufacturing their chips. Samsung is another one) — due to fear of potential consequences related to handing over chip designs to a competitor.
Get Pawel Rzeszucinski, PhD’s stories in your inbox
Join Medium for free to get updates from this writer.
Subscribe
Over time, TSMC’s foundry model proved incredibly successful. Concentrating global chip fabrication into a few specialist firms creates huge economies of scale. TSMC can invest at a scale no single chip design company could justify, serving thousands of customers worldwide. Apple, Qualcomm, AMD, Nvidia, and many others rely on TSMC for their most advanced chips.
TSMC’s dominant position is undisputed - in the second quarter of 2025, it controlled about 70.2% of the global pure-play foundry market. In comparison the closest competitors are Samsung Foundry with 7.3% and SMIC with 5.1%. See the image with the source below.
Press enter or click to view image in full size
Global Chip Manufacturing Market Share (Q2 2025). [Data source: TrendForce.com]
TSMC is also at the forefront of Moore’s Law, specializing in manufacturing and leading the charge in shrinking transistor sizes generation after generation. When TSMC opened its doors in 1987, chips were built on 3µm processes, which are huge by today’s standards. Fast forward to 2025, and we’re talking 1.4nm, a mind-bending ~2000 times smaller.
Press enter or click to view image in full size
Historical evolution of semiconductor transistor size since TSMC was founded [Source: ResearchGate]
Let’s stop for a second to reflect on the size of transistors today to fully appreciate what humans are capable of building — we just said TSMC is preparing to build a 1.4nm fab. For comparison, a human hair is about 80,000 nm thick. Let that sink in…and let’s go further — a typical virus, like influenza, measures around 100 nm, which means a modern transistor is 50 times smaller than a virus. Let’s go deeper! DNA? Its double helix is roughly 2.5 nm across — so a transistor is nearly the size of a strand of life’s blueprint. We’re now engineering structures at a scale where silicon rivals biology, carving patterns smaller than anything visible under a regular microscope. It’s not just small — it’s atomic-level precision.
Press enter or click to view image in full size
A modern transistor (2 nm) compared to DNA helix (2.5 nm), flu virus (100 nm), and human hair (80000 nm). Note: Images not to scale for clarity [Source: author]
The Counterintuitive That Works Like a Charm
The semiconductor industry has undergone a major shift from vertically integrated companies to the fabless model, where chip designers and manufacturers specialize in different roles. Fabless companies like NVIDIA, AMD, Qualcomm, and Apple focus on architecture, intellectual property, and software, while outsourcing fabrication to specialist foundries such as TSMC, Samsung Foundry, and GlobalFoundries.
This change is driven by cost and complexity. Building a cutting-edge fab costs tens of billions of dollars, and keeping up with Moore’s Law — which demands smaller transistors every few years — requires constant, expensive upgrades. Only foundries with massive scale and customer bases can afford this pace.
TSMC dominates this space, holding about 70% of the global foundry market in 2025, serving top tech giants. Its process evolution — from 3 microns in 1987 to 1.4 nanometers soon — illustrates the relentless march of miniaturization. At the current size transistors are nearly the size of DNA strands, smaller than most viruses, pushing manufacturing into atomic-level challenges.
The fabless model enables designers to innovate while manufacturers tackle fabrication’s physics and economics. Together, they deliver the chips powering smartphones, AI systems, and cloud infrastructure — proof that specialization drives progress in the modern semiconductor value chain.
Hello
ReplyNested test
Nested anonymous user comment