When large model parameters exceed the trillion level, memory and bandwidth on edge devices become the bottleneck. The industry has finally seen a key breakthrough: Shanghai Tongtu, a subsidiary of Toread, has signed a strategic agreement with Horizon Robotics to integrate full-stack large model compression technology into automotive-grade chips for the first time. This collaboration signals a systematic solution to the 'memory wall' problem in edge AI deployment.
Technical Breakthrough: Three-Layer Compression Tackles Industry Pain Points
The core challenge for running large models on edge devices is not computing power but data movement. During model inference, massive parameters and intermediate data are transferred between memory and computing units, occupying significant bandwidth and causing idle compute resources. Tongtu's full-stack compression technology addresses this with a three-layer system:
- Input layer: compresses large data like video to 1/5 of original size, reducing front-end bandwidth pressure.
- Intermediate layer: achieves an additional 50% compression on top of Google's TurboQuant algorithm, effectively alleviating the memory wall.
- Parameter layer: compresses trillion-parameter models by 30% without retraining, enabling plug-and-play deployment.
The technology has been verified in real-world tests for inference performance, bandwidth usage, and business outcomes. Notably, Tongtu's technical path aligns closely with Google's TurboQuant algorithm, confirming that 'compression liberates computing power' is an industry certainty.
Industrial Deployment: From Automotive Chips to Eight Scenarios
The key to this collaboration is the deep integration between Tongtu and Horizon Robotics. Horizon is a leading player in automotive-grade AI chips, with 2025 revenue of 3.758 billion yuan, up 57.7% YoY. In 2026, it released China's first cockpit-driving integrated intelligent vehicle chip 'Starry 6P' and an operating system. The partnership means large model compression technology will first land in intelligent driving scenarios, solving bandwidth and memory issues on automotive chips.
Tongtu's vice president Wang Hongjian stated at the 2026 Zhongguancun Forum that the edge AI market will form a massive blue ocean, with China's market size expected to grow from 600 billion yuan in 2026 to 2.1 trillion yuan by 2030. Tongtu's compression technology can be applied to eight scenarios including personal devices, smart spaces, smart vehicles, industrial internet, embodied robots, and AR/VR wearables.
Performance Validation: Chip Business as the First Growth Curve
Technical strategy is backed by financial results. In Q1 2026, Toread reported revenue of 496 million yuan, up 39.41% YoY; net profit attributable to shareholders was 56.45 million yuan, up 14.44%. The chip business posted revenue of 179 million yuan, a 206.92% YoY increase, accounting for 36.09% of total revenue. Chips have become the primary growth driver, while the outdoor business grew 8.02%, forming a 'high-growth chips, stable outdoor' pattern.
Institutional analysts are optimistic. Huaxi Securities maintained a 'buy' rating, forecasting 2025-2027 net profits of 237/369/454 million yuan. CICC raised its 2026 revenue forecast by 29% to 1.952 billion yuan and EPS by 60% to 0.15 yuan.
Dual-Business Synergy: Full-Chain Self-Sufficiency from Chips to Smart Equipment
Since establishing the 'outdoor + chip' dual-main business strategy in 2021, Toread has acquired five chip companies, building a product matrix covering 'sensing-interaction-display'. Tongtu's compression technology not only drives chip business growth but also provides core support for Toread's smart equipment, such as the second-generation intelligent exoskeleton and outdoor smart shoes. During the 618 shopping festival, the Crest C3 intelligent exoskeleton achieved daily sales of over 100 units on Tmall, weighing only 1.8 kg with a thrust-to-weight ratio of 10:1.
Full-chain self-sufficiency from underlying chips to end products is accelerating. When a technology solves a core global AI bottleneck, enters joint development with an industry leader in real commercial scenarios, and is backed by strong financial data, it is no longer a 'concept' but a tangible industry variable.