South Korean startup DeepX plans to spend more than $100 million developing a next-generation AI chip built on Samsung Electronics' 2-nanometer process, Bloomberg reported — a bet that a small challenger can carve out a niche in a market dominated by far larger incumbents.

The chip, designated the DX-M2, is intended for on-device and edge AI inference rather than data-center training, per reports. The more than $100 million covers performance verification and production engineering across the program, not merely a single tape-out. First silicon is expected in March 2027, fabricated on Samsung's most advanced 2nm node.

The deal matters for both sides. For DeepX, Samsung's leading-edge process is a route to power-efficient silicon that can compete on performance per watt. For Samsung, an external customer for its 2nm node is a valuable endorsement as it races TSMC for advanced foundry business.

It is also a sign of where AI chip competition is widening. Beyond Nvidia, hyperscaler-designed accelerators and Chinese alternatives, startups are pitching specialized inference silicon for phones, robots, cars and other devices where latency, cost and power — not raw training throughput — decide the winner. DeepX's challenge will be proving out that silicon on schedule, in a market where software ecosystems, not just transistors, define success.