The large-scale deployment of edge AI has long been stuck on a deceptively simple yet stubborn issue: enough computing power, but data cannot be moved fast enough. When large models have hundreds of billions of parameters, the data transfer between chips and memory during inference consumes far more bandwidth than the chip's computing bottleneck. In April, Toread's subsidiary Shanghai Tongtu signed a strategic agreement with Horizon Robotics to jointly develop large model bandwidth compression technology. This is not an isolated lab project but a key move that cracks the cost barrier of edge AI industrialization.
Technological Breakthrough: Three-Layer Compression Reshapes the Economics of AI Computing
Tongtu's technical system essentially builds a 'data highway' between computing power and storage. Its full-stack compression solution covers three levels: at the input data level, it compresses large-capacity data like video to one-fifth of its original size, meaning edge devices no longer need expensive high-bandwidth interfaces; at the intermediate data level, it achieves an additional 50% compression over Google's TurboQuant algorithm, directly alleviating the memory wall bottleneck during inference; at the model parameter level, it achieves 30% compression for billion-parameter models, with plug-and-play functionality and no need for retraining.
What does the combined effect of these three compression layers mean for buyers and factories? Taking a smart exoskeleton device as an example, it originally required a high-cost automotive-grade chip and dedicated memory modules. Now, it may only need a mid-range chip and standard memory to run a model of the same scale. The total hardware BOM cost for edge AI devices could drop by 30% to 50%. This is not incremental optimization but a step-change in the cost structure.
Industry Transmission: The Linkage Effect from Chips to Textile Smart Equipment
Since establishing its 'outdoor + chip' dual-business strategy in 2021, Toread has acquired five chip-related companies, building a chip product matrix covering perception, interaction, and display. Tongtu's compression technology is not only an independent growth driver for the chip business but also directly empowers its smart equipment product line. During the 618 shopping festival, Toread's self-developed smart exoskeleton Crest C3 launched for pre-sale, achieving daily sales of over 100 units on Tmall. The device weighs only 1.8 kg with a thrust-to-weight ratio of 10:1.
Behind this product's success is Tongtu's compression technology's substantial improvement in edge inference efficiency. For the textile industry, a similar logic is transmitting to scenarios like smart textile machinery, smart warehousing, and industrial quality inspection. A traditional fabric inspection machine, if deploying an AI quality inspection model, previously required a GPU server or high-end edge computing box. Now, with model compression technology, an industrial camera paired with a mid-range ARM chip can run a defect detection model. This makes it possible for small and medium-sized textile enterprises to access AI quality inspection systems at a lower threshold.
Performance Validation: Chip Business Becomes the First Growth Curve
The confidence in technological deployment is ultimately validated by financial data. In the first quarter of 2026, Toread's chip business revenue reached 179 million yuan, a year-on-year increase of 206.92%, accounting for 36.09% of total revenue. During the same period, the outdoor business grew by 8.02%, forming a dual-wheel drive pattern of 'high chip growth, stable outdoor growth'. Institutional signals are also positive: Huaxi Securities maintains a 'buy' rating, and CICC raised its 2026 revenue forecast to 1.952 billion yuan.
The rapid growth of the chip business means that Toread's identity shift from an 'outdoor brand' to a 'hard-tech platform' is being recognized by the capital market. For textile industry professionals, this signal is worth noting: when a traditional consumer brand achieves a second curve leap through chip technology, can other traditional industries replicate a similar path?