NAVER D2SF has put roughly KRW 3 billion into TraceWeave, a startup that built Phase‑TITAN – a new AI architecture that keeps memory use constant even as it processes long text. The company says its 1‑billion‑parameter model needs only 0.04 GB for a 1‑million‑token context, compared with tens of gigabytes for a standard Transformer. TraceWeave demonstrated the model running on a smartphone CPU without any cloud calls and completed a paid pilot with an autonomous‑driving unit of a global car maker.
Why it matters
Device makers could add sophisticated AI assistants to phones, drones or glasses without needing cloud connectivity, lowering latency and protecting user data.
