How the fabless model changed the semiconductor industry

A clothing brand in Bengaluru can design a shirt, have it stitched by a contract unit in Tiruppur and sell it under its own label. Nobody thinks the brand is less real because it owns no sewing machines. For most of its history the chip industry worked the other way round. A company that wanted to sell chips had to own the plant that made them.

That changed in 1987, when TSMC opened in Taiwan with government backing. It would make chips only to other companies' designs and sell nothing under its own name. Firms that design chips without owning a plant came to be called fabless, after the fab, the trade's word for a chip factory.

TrendForce, a Taiwanese research firm, counts $359.4 billion of revenue at the ten largest fabless firms in 2025, up 44% in a year. Nvidia alone took 57% of it.

01What do fabless, foundry, IDM and OSAT mean?

Four kinds of company now share the work of turning a drawing into a chip you can solder onto a circuit board. Each sits at a different step, and each spends its money on different things.

Two smaller trades grew up around them. Synopsys and Cadence sell the software engineers use to draw and check a chip. Arm licenses processor designs to chip companies and is paid a royalty on the chips that use them. None of these companies runs a chip plant.

The lines blur at the edges. Samsung and Intel run foundry businesses beside their own chips, and Apple designs the chips in its phones and laptops and buys the manufacturing from TSMC. Our semiconductor industry analysis maps the whole chain and the size of each part.

03How do fabless companies earn their money?

A fabless company sells chips, and its biggest cost is buying them. It pays the foundry for each finished wafer, then pays for packaging and testing, and books all of that as the cost of what it sold. What is left, the gross margin, pays for the engineers and software behind the next chip.

That margin is wide when a design is hard to copy. Nvidia's gross margin was 71.1% in the year to January 2026, so of every $100 of sales about $29 went on making the product, memory and packaging included. AMD's was 50% in 2025, and MediaTek's 47.5%.

Two of the five earn money in other ways too. Qualcomm's total revenue was $44.3 billion, because it also licenses its mobile patents. Broadcom's was $63.9 billion, of which about $27 billion came from infrastructure software.

This year has been sharper still. Nvidia reported $96.2 billion of revenue in the single quarter to July 2026, up 106% on a year earlier. What is driving that is set out in our piece on the AI-driven chip boom.

05What does this mean for an Indian chip-design startup?

For a founder, the fabless model is the cheapest way into chips. A startup's money goes on engineers and design software, and India has the engineers. The IT minister told the Rajya Sabha in March 2026 that India hosts nearly 20% of the world's chip design workforce.

Most of them work in the Indian centres of Qualcomm, Nvidia, AMD, MediaTek and others, on chips those companies own. Why so little of that work becomes Indian-owned product is the subject of our piece on India's chances as a chip hub.

How much of the design bill does the government pay?

Under the Design Linked Incentive scheme, run by MeitY through C-DAC, an eligible startup can claim back up to 50% of its design spend, capped at ₹15 crore. A second incentive pays 6% falling to 4% of the chip's net sales over five years, capped at ₹30 crore.

Both are paid after the money is spent, against milestones, so the founder still funds the gross cost first. PIB reported in January 2026 that the scheme backed 24 design projects. Between them they had completed 16 tape-outs, the step where a finished design goes to the foundry, and had six chips fabricated.

Reading those rules against a real budget is part of our grants and schemes work.

Why is foundry access the hard part?

An Indian startup is a tiny customer, and leading-edge capacity is booked by companies buying billions of dollars of wafers. Most start on older, cheaper processes and share a wafer with other designs in a multi-project run. Each design pays a slice of the mask and wafer cost, and gets back a small batch of prototypes.

Volume production is a separate negotiation, with minimum orders and a queue. The IT minister now expects the Dholera fab to begin commercial operations in mid-2028, on mature processes. Until then an Indian fabless company will have its chips made abroad.

Who buys the first chips?

A car or telecom company can take a year or more to approve a new chip, and the startup pays its engineers the whole time. A few Indian teams are past the early stages. Mindgrove Technologies raised an $8 million Series A for its microcontroller chips, and Netrasemi raised ₹107 crore in 2025 in a round led by Zoho.

07Frequently asked questions

What is the fabless model in semiconductors?

A company designs chips and sells them under its own name, and pays a separate foundry to manufacture them. It spends on engineers, design software and licensed designs, and owns no chip plant.

What is the difference between fabless, foundry and IDM?

A fabless firm designs and sells chips without making them. A foundry makes chips to other firms' designs and sells none of its own. An IDM, such as Intel or Texas Instruments, does both.

How do fabless companies make money?

They sell chips for more than the foundry, packaging and test bills. The gross margin left over, 71.1% at Nvidia in its last full year, pays for research and the next design.

Is Nvidia a fabless company?

Yes. Nvidia designs its chips and relies mainly on TSMC to make them. It reported $215.9 billion of revenue in the year to January 2026 without owning a chip plant.

Which fabless semiconductor companies are in India?

Most chip design in India happens in the local centres of foreign fabless firms such as Qualcomm, Nvidia and MediaTek. Indian-owned fabless startups include Mindgrove Technologies, Netrasemi and InCore Semiconductors. Saankhya Labs, another, was bought by Tejas Networks in 2022.

Figures as at October 2026; company figures are from each company's own results.