Thousands of AI Chip Startups Across Asia Battle for Access to TSMC Capacity

For Asia’s thousands of AI chip startups, designing a breakthrough processor is only the beginning. Their first major battle is getting a place in TSMC’s production queue.

Like Nvidia, these companies are fabless businesses. They design chips and rely on foundries to manufacture them. But only a handful of major foundries operate globally, and the most advanced capacity is tightly contested.

“

TSMC remains the leader in the field for companies targeting more advanced foundry processes, and securing an allocation of TSMC capacity is not easy.

Counterpoint Research data showed TSMC held 73% of the foundry market in both the first and second quarters of the year. Samsung Electronics ranked second with about 7%. For startups trying to establish themselves at the cutting edge, building a relationship with a foundry can be as important as the chip design itself. They are working to prove that they deserve a place on the manufacturers’ customer lists.

“TSMC remains the leader in the field for companies targeting more advanced foundry processes, and securing an allocation of TSMC capacity is not easy,” said Chan Yip Pang, an executive director at Vertex Ventures.

“

Very complex.

FuriosaAI, a South Korean company founded in 2017, is among the few Asian startups to secure TSMC capacity. Its senior vice president for products and business, Alex Liu, said the company had to obtain capacity from TSMC, high-bandwidth memory, packaging partners and customer validation at the same time. The process, he said, was “very complex.”

FuriosaAI has raised $246 million and delivered products to customers and partners including Samsung and LG. Its flagship RNGD AI inference chip entered mass production in January using TSMC’s 5-nanometer process. Asus assembled the first batch of 4,000 chips. FuriosaAI plans to produce another 16,000 this year, bringing total shipments to 20,000. Each RNGD chip is estimated to cost $10,000.

“

At the most advanced processes, even a small tape-out and the initial validated production run are extremely expensive.

Agrani Labs, headquartered in Bengaluru, is one of the few startups in India and Southeast Asia attempting to build advanced AI chips on a similar model. Chief executive Dheemanth Nagaraj is in talks to raise more than $100 million. Investors describe his team as among Bengaluru’s leading chip-design specialists, while about 150,000 engineers in the city are estimated to work for U.S. semiconductor companies such as Intel and AMD.

Yet even $100 million and strong chip-design expertise may not be enough to win TSMC’s cooperation. “At the most advanced processes, even a small tape-out and the initial validated production run are extremely expensive,” said Arjun Rao, a partner at Speciale Invest. “Given the massive global expansion of AI computing capacity, securing production capacity is a major obstacle for startups.”

“

Given the massive global expansion of AI computing capacity, securing production capacity is a major obstacle for startups.