The large-scale deployment of on-device AI has long been constrained by a physical bottleneck: during inference, the movement of massive parameters and intermediate data consumes far more bandwidth than the computing power itself. In Q1 2026, Toread's chip business posted a 206.92% year-on-year revenue surge, driven by a systematic technological response to this bottleneck. The strategic agreement between Shanghai Tongtu and Horizon Robotics has officially pushed full-stack LLM compression technology into the industrial deployment phase.
Technical Path: Three-Layer Compression Targets the Memory Wall
Tongtu's compression system covers the entire process from input to output. The first layer targets high-volume input data such as video, compressing it to one-fifth of its original size, directly reducing the input bandwidth pressure on end-side devices. The second layer tackles intermediate data generated during LLM inference, achieving an additional 50% compression on top of Google's TurboQuant algorithm, effectively alleviating the memory wall bottleneck. The third layer compresses model parameters for billion- and trillion-parameter models by 30%, without requiring retraining—plug-and-play.
This technical path aligns closely with Google's 2026 TurboQuant memory compression algorithm in core direction, jointly confirming that 'using compression to free compute power' has become a deterministic industry direction. Tongtu CEO Wang Hongjian stated at the Zhongguancun Forum that the on-device AI market is leaping from 600 billion yuan toward 2.1 trillion yuan by 2030, and compression technology is the critical infrastructure to unlock this path.
Industrial Deployment: Deep Integration with Horizon's Automotive-Grade Chips
Horizon Robotics, a leader in automotive-grade AI chips for mass production, reported 2025 revenue of 3.758 billion yuan, up 57.7% year-on-year. In 2026, it launched the 'Starry 6P' cockpit-driving integration chip and the 'Kakaxia' vehicle operating system. This collaboration integrates Tongtu's compression technology directly into Horizon's intelligent driving computing platform, optimizing inference bandwidth and memory usage on automotive-grade chips.
This means the 'bandwidth wall' for on-device AI is moving from lab solutions to production-grade products. For traditional industries like textiles and apparel, the low-latency, low-power characteristics of on-device AI will directly accelerate the deployment of smart quality inspection, flexible production scheduling, and AR-assisted design.
Performance Validation: Chip Business Becomes the Primary Growth Engine
In Q1 2026, Toread reported operating revenue of 496 million yuan, up 39.41% year-on-year, and net profit attributable to shareholders of 56.4457 million yuan, up 14.44%. Chip business revenue reached 179 million yuan, a 206.92% year-on-year increase, accounting for 36.09% of total revenue. The outdoor business grew 8.02% over the same period, forming a dual-wheel structure of 'high chip growth, stable outdoor growth'.
Institutional signals are equally positive. Huaxi Securities maintains a 'buy' rating, forecasting net profits of 237 million, 369 million, and 454 million yuan for 2025-2027. CICC raised its 2026 revenue forecast by 29% to 1.952 billion yuan and its EPS forecast by 60% to 0.15 yuan. Behind these numbers is the market's clear valuation premium on the commercialization prospects of the 'on-device AI compression' technical route.
Strategic Depth: Dual-Business Synergy and Full-Chain Self-Reliance
Since establishing the 'outdoor + chip' dual-business strategy in 2021, Toread has acquired five chip-related companies, building a chip product matrix covering the entire chain from 'perception to interaction to display'. Tongtu's full-stack compression technology serves both as an independent chip business growth pole and as core technical support for Toread's second-generation intelligent exoskeleton and outdoor smart shoes.
During the 618 shopping festival, the company's self-developed intelligent exoskeleton Crest C3 began pre-sales, achieving single-day sales of over 100 units on Tmall. The whole unit weighs only 1.8 kg with a power-to-weight ratio of 10:1. Full-chain self-reliance from underlying chips to end products is accelerating.