Key points
- DeepX ties its chips to Intel's AI software
- What DeepX's CEO heard from AMD after meeting Lisa Su
- DeepX's Samsung 2nm chip and IPO timeline
Korean AI chip startup DeepX is connecting its chips to Intel's AI software and has started talking with AMD about working together, chief executive Kim Lok-won said. Kim said the Intel work aims to let users plug in DeepX chips and start running them without manually setting up the software. He expects initial results early next year.
DeepX designs neural processing units, or NPUs, that run AI models with low power consumption. Its target devices include smart glasses, self-driving cars, and drones.
Kim spoke with Yonhap News Agency (연합뉴스) on Oct. 15 at DeepX's office in Pangyo, south of Seoul. The interview was published on Oct. 18 Korea time.
DeepX chips are being linked to Intel's OpenVINO
OpenVINO is Intel's software for running AI models. DeepX is adding its NPU to OpenVINO as one of the processors the software can choose, and it is optimizing models and developer tools for it, Kim said.
DeepX chips already run in industrial PCs that use Intel processors. But users had to install and connect the software themselves, and the work with Intel is meant to remove that step. Kim said DeepX will announce the partnership formally after the work produces results.
Talks with AMD are at an early stage
Kim met AMD chief executive Lisa Su on Oct. 7 at a roundtable in Seoul with Korean AI chip companies. "I confirmed that AMD is willing to work with outside companies and build an open ecosystem in physical AI," he said.
"After the meeting, we also got an email from their technical contact saying they want to follow up," Kim said. At the Oct. 15 interview, Kim said he planned to meet the head of AMD's Korea office the following week to discuss details. For now, the two companies are still confirming the direction and continuing working-level talks, he said.
Samsung 2nm chip planned; CEO wants an IPO within three years
Kim said DeepX plans to tape out a chip on Samsung Electronics' 2-nanometer process this month. A tape-out is the step where a finished chip design is sent to the factory to be made. Kim expects the first chips around March or April next year. The company's first-generation chip, the DX-M1, is already in mass production on Samsung's 5-nanometer process.
From August 2025 to July 2026, DeepX received 77 purchase orders from more than 10 countries, Kim said. About 300 companies around the world are testing its chips, including more than 100 in Korea. DeepX has contracts with 27 chip distributors and plans offices in Japan and Europe. Kim said its chips cost less than $100.
DeepX has finished software work with Amazon Web Services so AI models trained in the cloud can be sent to devices that use its chips. It also plans to release software for running robots later this year, built to work with Nvidia's Isaac robot development platform. Physical AI means AI that controls robots, cars, and machines in the real world. Korea has tripled its AI budget for 2026.
Kim said DeepX has not set a date for an initial public offering. "Becoming a leader in the global industry is our most important goal, and an IPO is a tool and a milestone for that," he said. "As CEO, I would like to do it within three years at the latest." Another Korean AI chip startup, Rebellions, is leaning toward a Kosdaq listing.
Quotes are translated from Korean.
