Semifive said on Monday it signed a contract with a U.S. AI fabless company to develop a next-generation AI inference accelerator. The deal is worth about 70.3 billion won ($52 million). It secured its first spec handoff project in the North American market.

Spec handoff is a model in which Semifive takes full responsibility for the process, from detailed design to packaging, testing, mass production and software development, if a client presents only core performance targets and specifications. Under the existing turnkey model, clients completed chip design drawings themselves and then handed over work mainly focused on outsourced manufacturing. Under spec handoff, Semifive leads the process from the point specifications are handed over through mass production of the finished product.

The contract is Semifive's largest-ever single deal. It exceeds 40 percent of last year's annual order intake of 168.4 billion won. Compared with new orders of 118.9 billion won in the first half of this year, it is equivalent to about 60 percent, the company said. Semifive plans to use the order as a foothold to expand its presence in North America and strengthen its technological competitiveness in the global ASIC market.

Semifive said it applied technology that reduces the memory power burden by using LPDDR6, a next-generation low-power DRAM. It also applied PCIe Gen5, a high-speed interface, to minimize bottlenecks in large-scale data transfer. The company stressed it plans to implement a design that secures both computing throughput and operational stability, leveraging experience from big-die projects of up to 800 square millimetres.

It plans to go through tape-out in the first half of 2027 and begin mass production from 2028 for global hyperscalers and cloud service providers (CSPs).

Semifive CEO Jo Myung-hyun (조명현) said demand is expanding in earnest for custom semiconductors optimised for customers' services and workloads, alongside rapid growth in the AI inference market. He said the company will successfully commercialise the next-generation inference accelerator and establish itself as an indispensable partner for developing core custom semiconductors in the global hyperscale infrastructure market.