Naver D2SF invests in on-device AI startup TraceWeave
TraceWeave's Phase-Titan architecture keeps memory use constant as input grows and runs AI agents on a smartphone CPU.
Naver's corporate venture capital arm Naver D2SF announced on October 21, 2026, a new investment in TraceWeave, an on-device AI startup, according to aitimes.com. The size of the investment was not disclosed.
TraceWeave has developed its own AI model architecture, called Phase-Titan, which targets the memory bottleneck of transformer-based models. According to the company, Phase-Titan keeps memory usage constant as the volume of processed information grows, while the KV-cache of transformer models expands with it.
The architecture is designed to run tool calls, interpretation of results and decisions on the next action on-device, without a cloud connection. TraceWeave has demonstrated an AI agent carrying out multi-step tasks using only a smartphone CPU.
The startup has also built its own AI model rather than relying on existing open-source models. It is currently verifying the performance of an in-vehicle agent together with a global automotive electronics software company.
Earlier funding and automotive work
TraceWeave was founded last year by researchers from ETH Zurich, Google and KAIST. The company raised 3 billion won in a seed round at a valuation of 30 billion won, and previously ran a proof of concept with the autonomous driving affiliate of a global automaker group.
Quick answers
How much did Naver D2SF invest in TraceWeave?
The amount was not disclosed in the announcement.
What is Phase-Titan?
It is TraceWeave's AI model architecture, designed to keep memory usage constant as processed information grows, unlike transformer models whose KV-cache expands.