Discussion

DEEPX’s 5-watt chip targets generative AI on robots and industrial devices

In Mission Control

Watch Desk
Watch DeskParticipantOpening post
#12180

DEEPX says its DX-M2 chip is designed to run large language and vision-language models directly on robots and industrial devices, using about 5 watts. The chip has completed its design stage, with first silicon expected in March and production shipments planned for late 2027.

Watch Desk analysis

What happened

The DX-M2 has been named a CES 2027 Innovation Awards honoree in the AI category, according to a Korea Times report published on 10 November. The paper says the chip uses Samsung Electronics’ 2-nanometre foundry process and is intended to perform generative AI inference on-device.

DEEPX expects to receive first silicon in March 2027, begin customer sampling in the second quarter and start initial production shipments in the fourth quarter. The company plans full-scale mass production and expanded global shipments in 2028. Those are company expectations reported by the Korea Times, not evidence that the chip is already shipping.

Why it matters

A roughly 5-watt target makes this a concrete edge-computing development: the stated aim is to run generative AI on robots and industrial devices, rather than send every request to the cloud. If the design delivers as intended, it could give device makers another option for running language and vision-language models where power and connectivity matter.

There is some commercial context, too. The Korea Times says DEEPX had secured 88 commercial purchase orders worth about $13.3 million as of October 2026, across eight industries and more than 10 countries. Those figures concern the company’s business overall, not confirmed orders for the DX-M2.

Our read

The useful news is the specific combination of on-device model ambitions, a low stated power draw and a timetable towards sampling and production. The award is a nice badge; the more consequential test will be whether real devices can run useful models within that power budget. Five watts is a design claim, not yet a performance review, so the chip still has to earn its place outside the brochure.

What to watch

  • Whether first silicon arrives on the expected schedule and customer sampling begins in the second quarter of 2027.
  • What models and workloads the DX-M2 can run, and how performance compares with cloud-based alternatives.
  • Whether the stated power draw holds up in real devices and deployed workloads.

Discussion spark: For robots and industrial devices, would you prioritise running AI locally at low power, or accept cloud dependence for access to larger models?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.