🆓 Free article · Anyone can read this article for free

How Many Companies in the World Actually Tape Out Chips?

Conclusion first

About 900-1,100 companies worldwide are estimated to do production tape-outs with a full-mask set. Including MPW and shuttle runs, the figure rises to about 1,250-1,650 companies. Counting only registered design companies brings the total to about 5,700-6,300, but for most of them, whether they have ever taped out cannot be verified individually. This article is the first census to separate those differences in numbers.

The companies directly confirmed by name and evidence in this survey number 640. Tier A, which sells production products, accounts for 511 companies, and Tier B, where a tape-out is confirmed but production has not yet begun, accounts for 129 companies — Tier A 511 / Tier B 129. By country, the survey confirmed China 173, US 147, Taiwan 97, Korea 56 and Japan 33, plus Europe 66, Israel 22, India 15, Canada 10, Southeast Asia 8, Australia 7, and South Africa/Brazil/Mexico/Turkey 6. Even so, 640 is not the final answer. It is a floor that can be proved by name, because companies without verifiable public evidence were not included. The full 640-company census list is published alongside this article.

Why no one knows this number exactly

A tape-out is a contract between a foundry and a customer. Contracts are private. Unless a company announces it itself, outsiders cannot know which company taped out on which process and when. Listed companies leave clues in disclosures and earnings. Unlisted companies become visible only when a product actually starts selling. Companies that close often disappear without a trace.

Answering this question therefore takes more than finding a single roster. Ranges have to be narrowed by layering numbers of different kinds. This article places confirmed facts first and builds estimates on top of them.

Start with confirmed facts

The hardest number is the customer count disclosed directly by TSMC. TSMC said it manufactured 12,682 different products for 534 customers in 2025, using 305 process technologies — TSMC 534 customers / 12,682 products / 305 technologies. In 2024, the customer count was 522. Two cautions apply here. First, 12,682 is not the number of products newly taped out that year. It is the number of product types in production that year, and it includes older products. Second, the 534 customers are not only fabless companies. Companies with their own fabs and system companies are also included.

The second fact is China. According to the China Semiconductor Industry Association (CSIA), China had 3,626 chip design companies in 2024 — CSIA China 3,626. Yet only about 20%, roughly 725 companies, had revenue above 100 million yuan. A large share of the remaining roughly 2,900 companies is very small. Companies that have never taped out, companies that provide design services, and companies that have effectively stopped are likely mixed in. The figure of 3,626 shows the registered scale of design companies. It is not the number of companies that tape out.

The third fact is Korea. Korean fabless companies are known to number about 120 — Korea ~120. The gap between China and Korea is already close to 30 times in company count alone. Company count, however, is not the same as competitiveness. What matters is how many of those companies repeat production tape-outs in practice.

There is also a historical benchmark. The Fabless Semiconductor Association, the predecessor of the Global Semiconductor Alliance, put the worldwide fabless count at about 1,300 companies in 2010. Fifteen years later, registered design companies in China alone are close to three times that figure. The number shows how far the industry's center of gravity has moved.

Method: three tiers

This census divides companies into three tiers.

Tier A is production. The company is selling actual chip products. It has revenue, listed-company disclosures, and products in shipment. Anyone can verify it from product pages and earnings.

Tier B has taped out but is not yet in production. The company has a tape-out announcement, has taken part in an MPW or shuttle run, or has working silicon and a prototype confirmed by a reliable source. Most startups sit in this tier. Some companies move up from this tier to production, and some disappear.

Tier C is registered only. The company is registered as a design company, but whether it has taped out cannot be verified individually. This tier is handled in numbers only, not by name. The moment unverified names enter a roster, a census stops being statistics and becomes speculation.

The boundaries were also made explicit. Companies that sell only IP, with no chip of their own, were excluded. Companies that provide design work only, with no product of their own, were also excluded. Foundries, outsourced assembly and test companies, and equipment companies were excluded. Conversely, companies were included if they tape out a chip of their own, whatever their corporate form. That covers fabless product companies, IDMs with their own design organizations, and the internal silicon organizations of system companies that design chips directly for smartphones and data centers. Internal organizations were counted as separate rows, with the parent company identified.

How the figure of 900-1,100 was derived

The starting point is a fact: TSMC's 534 customers.

To that, companies that do not use TSMC are added. Some use only domestic Chinese foundries. Some analog and power companies use only specialty-process foundries. Some mainly use other large foundries. Selecting only companies that do not overlap with TSMC customers added 280-420 companies. The principle is that a company using two foundries at once still counts as one company. The figure already removes an assumed overlap of 35-50%.

Finally, IDM internal design organizations that do not appear on foundry customer lists are added. These are design organizations that make chips only in their own fabs. Excluding IDMs that are already TSMC customers added a net 60-100 companies.

The sum is 874-1,054 companies. Rounded, that is 900-1,100 companies. That range is the central answer of this article. No single point figure is given because most large foundries do not disclose customer counts. Showing uncertainty as the width of a range is more honest than speaking as if an undisclosed number were known.

Why including MPW gives 1,250-1,650

Tape-outs come in different kinds. A full-mask tape-out creates a new mask set and enters production. The cost is high. An MPW run shares one wafer among the designs of several companies. The cost is far lower, and startups and research organizations use it as the route to check silicon for the first time.

Adding companies that have taped out only through MPW and have not yet reached production to the range above adds 350-600 companies. That breaks down as 250-400 companies from the Chinese long tail and 100-200 companies from startups and university spin-offs outside China. The combined total is 1,224-1,654 companies, rounded to 1,250-1,650 companies.

The difference matters. "Companies that have ever taped out" and "companies that do production tape-outs" are different populations. The former is larger than the latter by several hundred companies. The number that matters for industry policy and investment is usually the latter. Counting only the former mixes in a bubble.

What the 640 named companies tell us

The largest blocs in this roster are, as expected, China, the US and Taiwan. The three together account for 417 companies, 65% of the total. The remaining 35% is spread widely across Korea and Japan, Europe and Israel, India and Canada, Southeast Asia and Australia, and elsewhere. Chip design is concentrated in a small number of hubs, and at the same time it is actually happening in more countries than is often assumed.

China's 173 companies center on listed fabless companies. Compared with China's total of 3,626 registered design companies, companies confirmed by name and evidence are still less than 5%. That does not mean the rest are all fake. It means there is no verifiable public evidence. China's long tail is the largest part of this industry, and the least visible.

The US total of 147 companies includes many internal silicon organizations, in addition to listed fabless companies and IDMs. As cloud companies, electric vehicle companies and networking companies design their own chips, the boundary of a "semiconductor company" itself has widened. The share of internal organizations in the US roster is clearly higher than in other countries.

Taiwan's 97 companies are the group with the highest density of listed design companies. The confirmed ratio is high even against an estimated registered total of 300-350 companies. A listing-centered industry structure makes the roster solid.

Korea's 56 companies correspond to close to half of a fabless population known to be about 120 companies. The figure combines listed companies and unlisted companies that actually ship products. Japan's 33 companies have a high share of IDMs and their internal design organizations. In both countries, the population itself is smaller than in China and the US, and the confirmed ratio within it is relatively high.

Europe's 66 companies present a different picture. Large design organizations with their own fabs and small and mid-sized fabless companies in automotive and sensor fields are both thick. Israel's 22 companies are a dense group of product companies in fields such as automotive, radar and satellite communications. India 15, Canada 10, Southeast Asia 8 and Australia 7 are small in scale, but they show that companies actually producing chips exist in each country. For New Zealand, this survey found no verified product fabless company. Recording an absence as an absence is also part of a census.

How to read these numbers

First, company count and tape-out count are different. One internal silicon organization counts as one company, but in practice it tapes out several chips every year. A small fabless company, by contrast, may tape out only once every few years. The industry's actual level of activity is closer to product counts and mask counts than to company counts. The fact that TSMC produces 12,682 product types in a year shows that scale.

Second, the long tail changes fast. Tier B companies either move up to production or disappear. Registered company counts also change greatly from year to year. The figures in this census should be read as a snapshot as of October 2026.

Third, cross-country comparisons must be read together with the population. Comparing only confirmed company counts makes countries with developed disclosure systems look advantaged. Only when the registered population and the confirmed ratio are seen together can industry transparency be separated from actual scale.

Limitations

This census has three clear limitations.

📩 Following this story?

Get the follow-up analysis — earnings, reports and HBM supply — in your inbox daily at 10:30 KST. Subscribe free →

First, the roster is still a floor. Name rosters for Europe and Israel, India, Canada, Southeast Asia and Australia were completed, but in the unlisted long tails of China and the US many companies could not be included for lack of public evidence. Absence of a name does not mean absence of a company.

Second, most large foundries do not disclose customer counts. That is why the width of the production range reaches close to 200 companies. As foundry disclosure grows, the range will narrow.

Third, even among the 640 companies on the roster, the strength of evidence differs. Companies confirmed by listed-company disclosures are the most solid, and startups confirmed only by a tape-out announcement are the weakest. Each row of the roster records the type of evidence, the source and a confidence level alongside it. Anyone citing the roster should check those columns as well.

Even so, this census matters because it is the first to present "the number of registered design companies" and "the number of companies that actually tape out" separately. Judging an industry from China's single figure of 3,626, or speaking as if a few hundred known companies were the whole world, does not hold up once that separation is made.

Start by defining the word tape-out

Tape-out is an industry term for the moment design data is handed to a foundry. The name comes from the past, when that data was sent on magnetic tape. Data is now sent as files, but the term remains. For a design team, the moment is like submitting an exam answer sheet. After it, the circuit cannot be changed. Changing it means making the masks again and paying the cost and time again from the start.

That is why a tape-out is the heaviest decision in chip design. The finer the process, the more sharply mask costs rise. On a mature process the level is manageable, but on an advanced process, a mask set combined with design staff, verification costs and software licenses can reach an amount on which the fate of a company rests. That is why startups issue press releases for tape-out announcements. A tape-out is a declaration that "our design is going into silicon, not just words."

Put the other way, a design company that does not tape out has not yet taken that exam. Business registration, an office and design staff alone cannot establish the substance of a semiconductor company. Whether it has proved itself in silicon is the dividing line. That is why this census uses tape-out, not registration count, as its standard.

The picture changes by company type

This roster recorded companies in four types.

The first type is the fabless product company. It designs and sells chips under its own brand and leaves manufacturing to a foundry. It is the largest share of the roster. In countries with many listed fabless companies, this layer is thick and verification is easy.

The second type is the IDM internal design organization. It is the design division of a company with its own fab. These organizations sell chips externally and also use them only in their own products. They often do not appear on foundry customer lists, so foundry numbers alone miss this layer. A substantial part of design capability in Japan and Europe is in fact in this form. That is why this layer has a large share in the European roster.

The third type is the system company's internal silicon organization. A smartphone company, a cloud company or an electric vehicle company designs a chip directly for its own product. These organizations do not sell semiconductors externally, but by tape-out scale they are often larger than specialist fabless companies. Excluding this layer would miss a substantial part of the actual design principals for data center and mobile chips. Conversely, when this layer is counted in "the number of semiconductor companies," the parent company must be identified. That avoids the error of counting the same group several times.

The fourth type is the ASIC service company. Among companies that design a customer's chip on its behalf and take responsibility through tape-out, only those with their own product flow and design assets were included. Simple labor-dispatch design services were excluded. This boundary requires judgment because industry structures differ by country, and the roster marks the type separately so readers can distinguish it.

Why the China numbers are the hardest to read

China is the largest country in this census and also the most opaque. The figure of 3,626 registered design companies is larger than in any other country in the world. Yet companies confirmed by public evidence are only a part of it. There are reasons for that. Chinese unlisted companies have no disclosure obligation, and when a product circulates only inside China's domestic supply chain, traces remain scarce in public sources, not only in English but also in Chinese. Lack of confirmation and non-existence are different things.

This article therefore presents two numbers for China at the same time: a registered population of 3,626 and 173 companies confirmed by name. The truth lies somewhere between them, and the tape-out estimate bridged the gap by adding customers dedicated to domestic Chinese foundries as a separate range. The method is not perfect, but it is better than explaining China with either number alone.

Another feature of China's long tail is volatility. Many companies are created and many disappear. A company on this year's roster cannot be guaranteed to be in the same place next year. China's company count should be read as a flow, not as a stock at one point in time.

What Tier B means for startups

The 129 Tier B companies are the most dynamic layer in this census. They have taped out but have not yet proved themselves with production revenue. If a company in this layer moves up to Tier A in the next survey, the industry has a new supplier. If it disappears from the roster, that tape-out did not lead to a product.

For investors and industry policy, Tier B is a watch list. A tape-out itself proves design capability, but it does not prove business success. After silicon comes back, gates remain: customer acquisition, yield, packaging and the software ecosystem. That is why a tape-out must not be mistaken for the finish line. Conversely, it is also unfair to assess a company that has not even taped out on the same line. The tier distinction is a device to mark that middle ground.

How to use this census

This roster and model can be used for three purposes.

First, a reality check on market size. When someone presents a figure such as "there are thousands of design companies worldwide," the first question should be whether that figure uses a registration standard or a tape-out standard. The two figures differ by more than three times.

Second, a comparison of industry structure by country. A high confirmed ratio means a transparent, listing-centered structure. A large population with a low confirmed ratio means a thick long tail. That is a difference in structure, not a judgment that one side is superior.

Third, a starting point for verifying an individual company. Each row of the roster carries the type of evidence and a source. When citing a specific company or reviewing a transaction, the order is to re-check that evidence in the original source. This census is not the end of verification. It is the entrance to verification.

Three questions the numbers raise

The figures in this census are answers and, at the same time, questions.

The first question concerns concentration. Among the 640 companies confirmed by name, 65% are in China, the US and Taiwan. By registered population, China's share is overwhelming. Chip design as an activity looks geographically widespread, but in reality it is strongly concentrated in a small number of hubs. Whether that concentration eases or deepens will depend on the success or failure of each country's industry policy.

The second question concerns boundaries. Including or excluding system companies' internal silicon organizations greatly changes company counts in the US and several other countries. In an era when designing one's own chips becomes a basic capability of large companies, the scope of the semiconductor industry itself becomes a premise of the analysis. This census chose to include internal organizations while naming the parent company. When comparing with statistics that do not include them, that difference must be taken into account.

The third question concerns speed. Registered company counts grow fast, but the number of companies doing production tape-outs does not grow as fast. Advances in design tools and reuse of design assets have lowered entry barriers, while the gate to production remains expensive and narrow. The structure is one of a widening entrance and a narrowing exit. In that structure, the core of industry strategy becomes building the ability to connect a tape-out to production and revenue, rather than competing on company count. Policies to increase the number of companies and policies to make companies strong must be designed differently.

What will change in the next survey

The next survey plans to improve three things. First, it will dig deeper into the unlisted long tails of China and the US, to confirm companies left out this time for lack of public evidence. Second, it will re-check rows with medium or lower confidence against original disclosures, to raise the solidity of the roster. Third, it will record companies that have disappeared and companies that have been acquired separately. An industry's scale becomes accurate only when the record includes companies that have disappeared, not only companies newly created.

A census is not a table made once and finished. It is closer to a ledger that must be updated each time a tape-out occurs. This first survey is the first page of that ledger.

Frequently asked questions

How many companies in the world actually tape out chips?

On a full-mask production basis, the estimate is about 900-1,100 companies. Including MPW and shuttle runs, it is about 1,250-1,650 companies. The companies directly confirmed by name and evidence in this survey number 640.

Do all 3,626 Chinese design companies tape out?

No. The 3,626 figure is the number of design companies on a 2024 registration basis. Only about 20% have revenue above 100 million yuan, and for most of the rest, whether they have taped out cannot be verified individually.

Does TSMC's count of 534 customers equal the worldwide number of tape-out companies?

No. The 534 figure is the customer count of TSMC alone. Companies that do not use TSMC and design organizations that make chips only in their own fabs must be added to reach a worldwide scale. Conversely, when one company uses several foundries, double counting must be avoided.

Why were IP companies and design service companies left out?

Because they do not tape out a chip of their own. IP companies license designs, and design service companies act on behalf of customers' chips. ASIC service companies with their own product flow, however, were included.

How many Korean companies are there?

The Korean companies confirmed by name and evidence number 56. The total Korean fabless population is known to be about 120 companies, so the confirmed companies correspond to close to half of the population.

How many European companies are there?

The European companies confirmed by name and evidence number 66. Germany 8, the United Kingdom 9, Switzerland 8, and France and the Netherlands 7 each form the center. Europe is characterized by large design organizations with their own fabs and small and mid-sized fabless companies in automotive and sensor fields being thick together. Registered European design companies are estimated at 350-450, so 66 should also be read as a verifiable floor.

Is an MPW tape-out a real tape-out?

Yes. An MPW run also uses real masks and a real wafer, so it is a real tape-out. However, several designs share one wafer, so the cost is low, and the purpose is silicon verification, not production. This census therefore does not place companies that have done only MPW in the same tier as production companies, and counts them separately in Tier B or in the MPW-inclusive estimate. Adding them without distinction would make the industry's scale look inflated.

How often should these numbers change?

They should be updated at least once a year. Tape-out announcements, listings, acquisitions and closures happen quickly. The figures in this article are as of October 23, 2026.

Methodology, regional estimates and de-duplication assumptions are set out in the model document published alongside this article, and the full 640-company census list is available as the census data page. Each company on the roster is marked with the type of evidence and a source. This survey used only public sources, and no part of it asserts private contract information between a foundry and a customer as an estimate. Every figure shown as a range states in the article and the model document why it is a range. The most important rule of this census is that facts and estimates were kept separate to the end, so that a reader with the same sources can reach the same range by calculating again.

📩 Missed today's analysis? Get tomorrow's in your inbox

Daily Silicon at 10:30 KST plus earnings and analyst-report breakdowns, first. Samsung Electronics, SK hynix, TSMC and HBM supply, read by practicing engineers. Join 300+ subscribers.

Subscribe free →

Free · No spam · Unsubscribe anytime