The deployment of large models on edge devices has long been constrained by a physical bottleneck: data movement. During model inference, vast amounts of parameters and intermediate state data are transferred between chip memory and computing units, creating a 'memory wall' that throttles compute power. The recent strategic partnership between Toread's (300005.SZ) Shanghai Tongtu Semiconductor and Horizon Robotics targets precisely this pain point—using full-stack data compression to systematically reduce the load at three levels: input, intermediate state, and model parameters, aiming to make large models run faster and more efficiently on automotive-grade chips.
Technical Depth: How Three-Tier Compression Dismantles the Memory Wall
Tongtu's compression system is not a single algorithm but a full-chain three-tier architecture. The first tier targets high-volume input data such as video, compressing it to one-fifth of its original size, directly reducing input bandwidth pressure on edge devices. The second tier addresses intermediate state data (e.g., KV Cache) generated during inference, achieving an additional 50% compression beyond Google's similar achievements, effectively alleviating the memory wall bottleneck. The third tier handles models with hundreds of billions or even trillions of parameters, achieving 30% parameter compression without retraining, offering plug-and-play deployment.
This technical approach aligns closely with Google's recently released TurboQuant memory compression algorithm. Wang Hongjian, Vice President of Toread Group and CEO of Shanghai Tongtu Semiconductor, stated at the 2026 Zhongguancun Forum Annual Conference that both parties have confirmed that 'using compression to liberate computing power' is a deterministic industry direction. For traditional industries like textiles and apparel, the significance of this breakthrough is that future edge AI devices can process data locally without relying on high-bandwidth cloud connections, enabling real-time processing for smart wearables and real-time quality inspection in smart factories.
Performance Validation: Chip Business Enters High-Growth Channel
The confidence in the technology is ultimately validated by financial data. In the first quarter of 2026, Toread reported revenue of RMB 496 million, up 39.41% year-over-year, and net profit attributable to shareholders of RMB 56.45 million, up 14.44%. The chip business stood out: revenue reached RMB 179 million, surging 206.92% YoY, accounting for 36.09% of total revenue. The chip business has become the company's primary growth driver, while the outdoor business grew 8.02% YoY, forming a dual-engine growth pattern.
Institutional investors have also responded positively. On June 17, Toread received 11 institutional investor surveys. Huaxi Securities maintained a 'buy' rating, forecasting net profit attributable to shareholders for 2025-2027 at RMB 237 million, RMB 369 million, and RMB 454 million respectively. CICC raised its 2026 revenue forecast by 29% to RMB 1.952 billion and EPS forecast by 60% to RMB 0.15. These figures indicate growing capital market recognition of Toread's transformation from an outdoor brand to a 'hard tech' enterprise.
Industry Impact: Full-Chain Self-Sufficiency from Chips to End Products
Since establishing its 'outdoor + chip' dual-core strategy in 2021, Toread has acquired five chip-related companies, building a full-chain chip product matrix covering 'perception-interaction-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 sports exoskeleton and outdoor smart shoes. During the 618 shopping festival, the company's self-developed intelligent sports exoskeleton Crest C3 began pre-sales, achieving daily sales of over 100 units on Tmall, with a total weight of only 1.8 kg and a thrust-to-weight ratio of 10:1.
The full-chain self-sufficiency from underlying chips to end products is accelerating. For industries like textiles and apparel that rely on IoT and smart manufacturing, this means future smart equipment (e.g., smart sewing machines, automated cutting systems) may integrate stronger edge AI capabilities, enabling real-time decision-making without cloud dependency, reducing network latency and data security risks.
Market Outlook: Edge AI as a Trillion-Yuan Blue Ocean
Wang Hongjian predicted at the 2026 Zhongguancun Forum Annual Conference that China's edge AI market size could grow from RMB 600 billion in 2026 to RMB 2.1 trillion by 2030. With its full-stack compression capabilities, Tongtu can penetrate eight major scenarios, including personal devices, smart spaces, smart vehicles, industrial internet, embodied robots, and AR/VR wearables.
For textile industry professionals, this trend implies that over the next three to five years, the deployment cost of smart factories, smart warehouses, and smart stores will drop significantly due to the proliferation of edge AI chips. The Tongtu-Horizon collaboration essentially paves the way for edge AI—when large models no longer depend on high-bandwidth clouds, textile companies can perform tasks such as fabric defect detection, inventory forecasting, and personalized customization locally, improving efficiency while ensuring data privacy.