For the past couple of years, much of the attention has been on large models generating text and drawing images. However, this year the AI world is undergoing a massive, quiet shift. AI is no longer confined to the screens of computers and phones; it is stepping out of the digital space to touch the real physical world.
The recently viral Cybercab robotaxi from Tesla Inc (NASDAQ: TSLA), which has just entered trial operation, is a vivid signal of this transformation. This car, with no steering wheel and no pedals, is essentially a complete physical AI system. Beyond autonomous driving, NVIDIA Corp (NASDAQ: NVDA) is also pouring significant resources into the same direction, building world models and full end-side toolchains for robots, pushing AI computing power from expensive cloud servers down to physical terminals.
Right now, the market's focus is mostly on the flashy end products: vehicles, robot bodies, and general-purpose large models. Yet the perception layer that provides the "visual senses" behind these products is often overlooked. Hai Qing Zhi Yuan (HKG: 01392), listed in Hong Kong, is a key player deeply entrenched in this often-underestimated field.
The market's label of "Physical AI First Stock" for Hai Qing Zhi Yuan is not just a hollow slogan. The company has not opted for eye-catching robot chassis or chased the widely discussed general large models. Instead, it focuses on the most fundamental multi-spectral perception. Just as the Cybercab must first rely on sensors to see roads, pedestrians, and obstacles before its models can make judgments and execute driving commands, Hai Qing Zhi Yuan captures what the human eye cannot see: infrared thermal radiation from overheating equipment and ultraviolet arcs from electrical faults. It translates hidden risks in industrial environments into data that AI can parse, essentially giving machines eyes that surpass human vision.
Looking across the board, from autonomous vehicles and humanoid robots to inspection equipment in factories, the starting point of every physical AI story is perception. Machines must first "see" the real world before they can understand their environment and execute actions. If the perception link falls short, even the most powerful large model becomes like a wise but blind master, full of thoughts yet unable to see what is happening around it.
As the industry wave rolls forward, the market's yardstick for evaluation is also changing. In the past two or three years, the market chased computing power and AI hardware concepts, with many names gaining popularity on stories alone. But this year, the tide has turned. Capital is no longer satisfied with intangible conceptual narratives. Investors are asking the most fundamental questions: Can this AI technology actually be implemented? Can it genuinely generate revenue? The entire AI industry is shifting from "telling stories on the investment side" to "delivering results on the output side."
Coinciding with the new round of adjustments to the Stock Connect scheme taking effect, Hai Qing Zhi Yuan has reached its own liquidity milestone. However, the market has not followed the script of "rising upon inclusion" that some had predicted. The batch of newly included stocks collectively opened higher but then fell, with short-term funds taking profits as the good news materialized, causing some market volatility. But true smart money is never swayed by a few days of price action. When short-term sentiment creates price fluctuations, that is exactly when we should calm down and carefully assess the true quality of the company.
Looking at the company's just-released interim report, this multi-spectral perception technology has long left the laboratory and translated into tangible business results. In the first half of fiscal 2026, the company achieved total revenue of RMB 410 million, up 84.5% year-on-year. After excluding one-off listing expenses, adjusted net profit was RMB 27.6 million, up 21.9% year-on-year, with revenue and profit both growing simultaneously. The most impressive figure was revenue from multi-spectral AI large model services, which surged to RMB 320 million, a 382.5% year-on-year jump, pushing its revenue share from 29.8% to 78%.
Once a company primarily selling hardware modules, high-value AI services have now become the absolute main growth engine. Of course, behind this report card lie the company's proactive choices. To concentrate resources on high-value businesses, the company deliberately scaled back some low-margin hardware orders, leading to a decline in module-related revenue. The increased hardware proportion in large projects also temporarily weighed on overall gross margins. However, as the share of high-margin large model services continues to rise and the product mix keeps improving, there is a clear path for gross margin repair.
What is even more commendable is that while growing profits, the company did not embellish its financials by cutting R&D. R&D expenses in the first half reached RMB 35.119 million, a 125.3% year-on-year increase. Balancing commercial orders with continued deep investment in the technology foundation is precisely where the quality of this profitability comes from.
Across the entire industry, the market space for multi-spectral AI is rapidly expanding. According to Frost & Sullivan data, the global market size is expected to grow from RMB 85 billion in 2025 to RMB 305.4 billion by 2030, with China being the fastest-growing core market. Based on 2025 revenue, Hai Qing Zhi Yuan already ranks first in China's multi-spectral AI large model services market with a 23% market share.
Beyond technology, recognition from official bodies and the industry is also accumulating. The Zhiyuan Origin large model has completed generative AI filing with the Cyberspace Administration of China and has been shortlisted for Shenzhen's model service institutions pool, securing a key compliance pass for conducting business with government and enterprise clients. It has also made Head豹 Research Institute's TOP10 list for physical AI application practices, standing alongside leading tech companies, and has been recognized as a Guangdong Province manufacturing single champion enterprise.
The technology is no longer limited to IDC data centers; it is heading to more diverse application scenarios. In August, Hai Qing Zhi Yuan partnered with Yunshen Chu Technology to integrate multi-spectral perception modules into quadruped inspection robots. This gives robot dogs "super eyes" for detecting arcs and overheating hazards, enabling them to operate in high-voltage substations, cable ducts, and energy storage facilities, upgrading traditional periodic manual inspections to continuous proactive risk warnings. On September 3, Academician Li Erping of the Chinese Academy of Engineering led an expert team to visit Hai Qing Zhi Yuan, conducting on-site research of the company's multi-spectral AI technology, product systems, and application scenarios. They discussed future industrial cooperation around multi-spectral perception, robot control, and complete machine manufacturing, representing top industry experts' endorsement of this technological path.
Looking ahead, the company's future growth path is clear. IDC security operations and maintenance form a solid foundation validated by its financials: existing customers are expanding from pilot data centers to full campus deployments, directly raising average order values. As numerous intelligent computing centers across the country come online, physical security is shifting from an optional feature to an industry necessity, with business likely to further penetrate the entire computing chain, from cabinets to battery rooms to busways. Meanwhile, if the "multi-spectral + embodied intelligence" model for power inspection passes customer validation, it could open a second growth curve. Continuous iteration of end-side models is liberating large models from expensive central servers to terminal hardware, significantly lowering deployment barriers and potentially expanding the customer base from large state-owned enterprises to a vast number of SMEs. The overseas market layout offers even longer-term imagination.
The activation of Stock Connect provides Hai Qing Zhi Yuan with a doorway to southbound capital liquidity, allowing more mainland investors to see this hard-tech company rooted in the perception layer. At moments when good news is realized, there will always be short-term funds choosing to cash out, and market fluctuations are the norm. The rise and fall of a few candlesticks reflect only the game of sentiment and chips, not the pace of industry progress. From Tesla's robotaxis to the endless stream of robotics projects, the era of physical AI is slowly unfolding. The market loves to debate the reasoning capabilities of large models and the execution actions of robots, yet often overlooks the most foundational link: if machines cannot see the real physical world, everything else is impossible. Perception, this often-underestimated link, is precisely the most basic and unavoidable gateway to physical AI. The value repricing of the perception layer may have only just begun.