DEEPX has placed smart glasses and 6G AI Radio Access Network (RAN) infrastructure on its product roadmap, targeting both with low-power neural processing unit (NPU) silicon, Digitimes reported.

What is DEEPX positioning for smart glasses?

The Korean AI accelerator developer positions the new device as a low-power NPU for head-mounted displays. Thermal envelope and battery capacity constrain on-device inference in this form factor: smart glasses need sustained AI compute for computer vision, speech recognition, and real-time translation, typically without active cooling or large batteries.

The wearable NPU segment remains underserved. Most existing designs rely on smartphone-class chips run at throttled frequencies, which compromises inference throughput. Smartphone-class NPUs from Apple, Qualcomm, and MediaTek target multi-watt envelopes with active cooling paths, not the sub-500-milliwatt sustained workloads smart glasses demand.

DEEPX's low-power positioning targets this gap. The company has not announced which smart glass OEMs it works with, the process node of the NPU, its target power budget in milliwatts, or a launch window for reference designs. Pricing remains undisclosed.

How does this connect to 6G AI RAN?

Beyond wearables, DEEPX evaluates the 6G AI RAN opportunity. AI RAN denotes the integration of artificial intelligence workloads into base station and radio unit processing, an area several chipset vendors pursue as mobile network operators plan 6G rollouts later in the decade.

AI RAN silicon targets 10-50 watt envelopes with different precision and latency requirements than wearable NPUs. Inference at the radio unit can reduce fronthaul load and enable dynamic spectrum management across massive MIMO arrays. The 3GPP standardization track for 6G is expected to define Release 21 studies around 2025-2026, with AI/ML for the air interface listed as a core work item.

DEEPX has not disclosed which baseband vendors, RAN equipment makers, or mobile network operators it engages for AI RAN trials, nor whether its NPU will deploy alongside or in place of existing digital signal processors in radio units.

What are the technical trade-offs?

A low-power NPU for smart glasses must balance three variables: peak TOPS, sustained TOPS under thermal throttling, and memory bandwidth. Smart glasses typically offer 2-8 GB of LPDDR memory and constrained PCB area.

An AI RAN NPU must instead prioritize determinism and throughput-per-watt under varied radio conditions, with deployment in outdoor base station cabinets.

DEEPX has not disclosed whether the smart glasses and AI RAN products will share a common NPU IP, memory subsystem, or compiler toolchain. The company also has not confirmed whether the silicon will come from TSMC, Samsung Foundry, or another node provider.

What does the move signal for the broader NPU market?

DEEPX's dual-track announcement comes as edge AI silicon fragments by use case. Always-on PC NPUs from Intel, AMD, and Qualcomm occupy the 1-5 watt range. Wearable NPUs must fit below 500 milliwatts for sustained workloads. Automotive NPUs target 30-100 watts with functional safety certifications.

Each segment demands distinct silicon, software stacks, and power management approaches. The roadmap positions DEEPX across two of the fastest-growing edge AI segments, though specifics on silicon, software stack, and customer engagements remain undisclosed. The company has not announced tape-out timing, foundry partner, or first-customer sampling dates for either program.