The core bottleneck for large-scale edge AI deployment is shifting from computing power to data transmission efficiency. Toread's subsidiary Shanghai Tongtu recently signed a strategic cooperation agreement with Horizon Robotics to jointly develop large model bandwidth compression technology, directly targeting the data transfer bottleneck in model inference for autonomous driving and robotics. This cooperation moves Tongtu's pre-research technology into the industrial application phase.
Technical Pain Point and Three-Layer Compression Solution
When deploying large models on edge devices, the massive parameters and intermediate state data consume significant bandwidth during chip-level data transfer, creating the so-called 'memory wall.' Tongtu's three-layer compression system targets input data, intermediate state data, and model parameters. For video input, the compression ratio reaches 5:1. For intermediate states like KV Cache, it can achieve an additional 50% compression over Google's TurboQuant algorithm. For models with hundreds of billions of parameters, it delivers 30% compression without retraining.
These technical metrics mean that under the same bandwidth, edge devices can run larger models or significantly reduce latency for existing models. For Horizon Robotics, known for its automotive-grade 'Starry 6P' chip, combining its platform with Tongtu's compression technology will directly improve inference efficiency in cockpit and driving fusion systems.
Performance Validation and Dual-Business Synergy
The confidence in this technology is backed by financial data. Toread reported Q1 2026 revenue of RMB 496 million, up 39.41% year-on-year. Its chip business generated RMB 179 million in revenue, surging 206.92% year-on-year, accounting for 36.09% of total revenue, becoming the company's primary growth engine. The outdoor business grew 8.02% during the same period, forming a pattern of 'high-growth chips, stable outdoor.'
Institutions have responded positively. Huaxi Securities maintains a 'buy' rating, forecasting net profits of RMB 237 million, 369 million, and 454 million for 2025-2027. CICC raised its 2026 revenue forecast by 29% to RMB 1.952 billion and EPS forecast by 60% to RMB 0.15. These projections reflect recognition of Toread's 'outdoor + chip' dual-core strategy since 2021, which has built a full-chain capability of 'sensing-interaction-display' through five chip acquisitions.
Key Scenarios for Industrial Deployment
The Tongtu-Horizon cooperation is not an isolated event. Toread has applied its full-stack compression technology to its second-generation smart exoskeleton and outdoor smart shoes. During the 618 shopping festival, the Crest C3 smart exoskeleton sold over 100 units in a single day on Tmall, weighing only 1.8 kg with a thrust-to-weight ratio of 10:1. The full-chain self-control from chips to end products is taking shape.
Toread Vice President Wang Hongjian predicted at the Zhongguancun Forum that China's edge AI market will grow from approximately RMB 600 billion in 2026 to RMB 2.1 trillion by 2030. Tongtu's compression technology can penetrate eight major scenarios, including personal devices, smart spaces, automotive, industrial internet, embodied robots, and AR/VR. This means that a technology that solves a universal bottleneck for the global AI industry and completes joint development with a leading company in real commercial scenarios is no longer a concept but a real industrial variable.