SEMIFIVE wins $52M inference-chip contract

SEMIFIVE signed a $52 million contract with a U.S. AI fabless company to develop a next-generation inference accelerator, its largest single contract to date. The company’s announcement says the project represents more than 40% of its 2025 orders, uses LPDDR6 and PCIe Gen5, targets tape-out in the first half of 2027, and aims for global mass production in 2028. Infrastructure teams should file this as evidence that custom silicon is moving upstream into spec hand-off, not as shipped capacity; the archive’s custom-silicon choice analysis maps the tradeoff.