SEMIFIVE Wins $52 Million U.S. AI Inference Chip Contract, Its Largest to Date

SEMIFIVE, a provider of custom AI semiconductor (ASIC) solutions, announced on September 29, 2026, that it has signed a contract worth approximately USD 52 million with a U.S.-based AI fabless company. The agreement covers development of a next-generation AI inference accelerator and marks the company's first "Spec Hand-off" engagement in the North American market.

A Deal Sized Against the Company's Order Book

According to the company, the contract is the largest single deal in SEMIFIVE's history. Its scale is clear from the company's own order figures: it represents more than 40 percent of SEMIFIVE's total orders of USD 124.3 million in 2025, and roughly 60 percent of the USD 87.7 million in new orders secured during the first half of 2026. SEMIFIVE said the win strengthens its presence in North America and reinforces its technological competitiveness in the global custom AI semiconductor market.

What Spec Hand-off Means for the Customer

Under the Spec Hand-off model, customers need only supply their key performance requirements and specifications. SEMIFIVE then assumes full responsibility for the development process, covering detailed design, software development, packaging, testing, and mass production.

The company drew a contrast with conventional turnkey engagements, in which customers typically complete the chip design themselves before outsourcing later implementation and manufacturing. In the Spec Hand-off arrangement, SEMIFIVE takes the lead at an earlier stage of chip development.

SEMIFIVE noted that specialized ASIC companies are taking on a broader role from the requirements stage across the global custom semiconductor market. It cited major technology companies such as Google and Amazon, which have worked with semiconductor firms including Broadcom and Marvell to develop their own AI silicon.

Memory, Interconnect and Die Size

For the project, SEMIFIVE said it will integrate advanced technologies to deliver the high memory bandwidth and data processing performance required for large-scale AI model inference, while lowering power consumption and total cost of ownership (TCO).

The accelerator will use LPDDR6, a next-generation low-power DRAM technology, to cut memory-related power consumption. It will also employ PCIe Gen5 to minimize bottlenecks in high-volume data transfers. SEMIFIVE additionally plans an architecture intended to deliver both high compute throughput and operational stability, drawing on its experience with multiple Big Die projects of up to 800 mm².

Schedule and Intended Buyers

Tape-out is targeted for the first half of 2027. The two companies then plan to begin global mass production in 2028, aiming at global hyperscalers and cloud service providers (CSPs).

Management Comment

Brandon Cho, CEO and Co-founder of SEMIFIVE, said demand is accelerating for custom silicon optimized for each customer's services and workloads as the AI inference market expands rapidly. "By successfully commercializing this next-generation inference accelerator and improving the economics of large-scale AI models, we aim to establish SEMIFIVE as an essential custom silicon development partner for the global hyperscale infrastructure market," Cho said.

Other Recent Company Developments

The announcement follows other recent news from SEMIFIVE. The company has announced the start of mass production of HyperAccel's LLM AI inference accelerator "Bertha," a data center inference accelerator built on Samsung 4nm.

It also reported a 97 percent year-over-year revenue increase in the first half of 2026, which it attributed to its ASIC turnkey, mass production, and IP businesses. That release cited bookings of USD 87.7 million for the period. SEMIFIVE describes itself as a leading global provider of custom AI semiconductor (ASIC) solutions. The North American engagement adds a new customer market for its end-to-end development model.