At the 2026 Yizhuang Half Marathon, Glory’s robot “Lightning” won with a time of 50 minutes and 26 seconds, significantly breaking last year’s record of 2 hours and 40 minutes, becoming the first humanoid robot to complete a real-world race course using a liquid-cooling system. This breakthrough was enabled by leveraging smartphone supply chain capabilities: suppliers such as Lingyi Zhizao and LanSi Technology provided structural components, while Huake Lengxin supplied high-speed liquid-cooling pumps, integrated with smartphone-grade thermal management, battery BMS, SoC computing power, and mass-production processes. Glory avoided the capital-intensive path of in-house motor and gearbox development, instead achieving rapid deployment through system integration. Simultaneously launched, “Yuanqi Zai” is positioned for in-store services and has achieved low-cost, scalable validation through 5,000 owned retail channels, highlighting its commercialization strategy anchored in retail scenarios. This model relies on four structural advantages—owned distribution channels, brand budget allocation, cost amortization logic, and frequent user feedback—creating a clear distinction from financing-driven startups like Unitree and Songyan.
Author and source: Wang Zhiyuan
Lightning is a show, but the real answer is the energetic kid.
At the south gate of Nanhaizi Park in Yizhuang, Beijing, the 2026 Yizhuang Half Marathon kicked off with over 100 robot teams and more than 300 robots—nearly five times as many as last year.
At 8:18 AM, the first robot to cross the finish line was "Lightning," from the Glory "Jueying Chiku Team," using remote control mode, with a net time of 48 minutes and 19 seconds—but it was not the champion.
The champion is another "Lightning," the autonomous navigation version, with a time of 50 minutes and 26 seconds, from the honored "Great Sage Team."
01
What does 50 minutes and 26 seconds mean?
I checked—the men’s half-marathon world record for humans is 57 minutes and 20 seconds, set just this past March 8 by a Ugandan athlete. “Lightning” is nearly seven minutes faster than that.
The second and third place winners are also called "Lightning," and the top three are all from Glory; Glory later said that the top six are all from its team.
Last year’s champion, Tian Gong Ultra, fell into the grass after system glitches following the finish line, and Yuyu’s veteran H1 also collapsed after completing the race; last year, it took 2 hours, 40 minutes, and 42 seconds to finish, but this year, the newcomer brand “Lightning” completed it in just 50 minutes and 26 seconds—cutting the record by two-thirds in its debut year.
This robot did two things right: one explicit, one implicit.
Explicitly, it's a leg.
Lightning is 169 cm tall with an effective leg length of 0.95 meters—this gives the impression of longer legs, and there's an engineering detail behind it. Honor's official explanation is that the 0.95-meter measurement is achieved by shifting the pivot point upward.
What does it mean?
Lift the starting point of the thigh higher. By raising the hip joint’s pivot point inside the body, the portion of the leg that can swing forward becomes longer, resulting in a greater stride length and faster speed at the same step frequency.
Engineer Yao from Glory said in a post-match interview that this concept was inspired by the body structure of long-distance runners.
Compared to Tian Gong Ultra, which stands at 180 cm—11 cm taller than "Lightning"—the official specifications have not disclosed Lightning’s effective leg length. The veteran H1 sent by Unitree is relatively shorter. While Lightning isn’t the tallest, it has the largest leg-to-body ratio.
Paired with a proprietary integrated joint module delivering a peak torque of 400 N·m, the "Lightning" achieves a peak 100-meter speed of 6 m/s. The remote-controlled version reaches a sustained running speed of over 7 m/s.
Legs only solve the problem of running fast; running fast means generating more heat. The harder part that follows is where the real gap emerges, because invisible heat is an issue the entire industry has yet to solve.
I reviewed last year’s half-marathon coverage—the scene was quite vivid, with many robots overheating and shutting down halfway through; staff had to run alongside them, spraying coolant on their joints.
When a humanoid robot runs at high speed, the temperature of the leg joint motors can reach 120°C. Beyond this temperature, motor torque decreases, positioning accuracy declines, there is a risk of permanent magnet demagnetization, and material aging accelerates.
To protect the motor, the system typically reduces power automatically. If the robot slows down, loses strength, or stops altogether, this is usually the reason.
The "lightning" method is liquid cooling.
Proudly, the liquid cooling pipes extend deep into the motor like capillaries, with a high-power liquid pump delivering a heat exchange flow rate of over 4 liters per minute. Each of the four drive motors in the lower limbs is equipped with its own independent liquid cooling loop.
There’s a detail worth noting: the core liquid cooling pump in the "Lightning" system comes from a supplier called Huake Lengxin. Its HD01 magnetic levitation pump operates at over 20,000 RPM, delivers a flow rate of 6 liters per minute, has a volume of approximately 30 mm, and weighs less than 100 grams.
Therefore, Honor's "self-developed liquid cooling system" is more accurately described as a collaborative development between Honor, which designed the solution, and Huazhong University of Science and Technology's ColdCore, which provided the core components.
To be honest, liquid cooling is one of two paths within the humanoid robotics industry.
Unitree uses air cooling, with a patented solution featuring a multi-stage air duct system and dual-fan coordination: one fan covers the thigh motor and drive board, while the other is dedicated solely to the calf motor. Unitree itself has stated that air cooling eliminates the risk of liquid leakage and is better suited for lightweight designs.
Zhìyuán and Dáqià also use liquid cooling, just like Honor, but they started much earlier—Zhìyuán’s PowerFlow joint module has long integrated a unified liquid cooling loop, and Dáqià’s Expedition A1 also uses liquid cooling.
Tesla's Optimus is still under evaluation; Japanese research teams have even developed "motor sweat glands" that dissipate heat by allowing moisture to seep and evaporate through porous structures.
So, Glory wasn't the first to develop liquid cooling, but it was the first to successfully implement liquid cooling in a real-world setting—the difference may sound minor, but it's the gap between mass production and a prototype.
02
A robot team that was just formed a year ago—where did all these things come from? The answer lies in the supplier list of “Lightning.”
In addition to Huake Lengxin, Lingyi Intelligent Manufacturing supplied 159 core metal components covering the hip, leg, foot, and torso.
Lansheng Technology supplied 132 models, covering the head, arms, hips, and legs. AAC Technologies not only manufactured the key structural components of the "Lightning" unit but also handled the head and legs of the "Yuanqi Zai."
I specifically looked into the backgrounds of these companies: Lingyi has been manufacturing iPhone structural components for over a decade, Lansheng initially supplied glass covers for Apple, and AAC is a leading enterprise in mobile acoustic modules.
Glory didn’t hire a specialized robot supplier; instead, they took their existing supply chain list, selected dozens of partners, and built this robot. It’s quite an abstract approach.
Based on my understanding, there are at least five capabilities that smartphone manufacturers can directly transfer to robotics.
First is thermal management: smartphone manufacturers have spent over a decade mastering heat dissipation by packing SoCs into spaces as small as one square centimeter; the processes, materials, and simulation capabilities for liquid cooling solutions are already well-established.
Second, the battery and BMS—charge and discharge curves, temperature protection, and range management—are areas where smartphone manufacturers’ teams may be the most knowledgeable on Earth.
Third is the SoC and visual computing power: smartphone SoCs are already designed to handle multiple cameras, SLAM, and night scene noise reduction simultaneously—capabilities that are precisely what robots need for autonomous navigation.
Fourth, supply chain bargaining power: companies like Lingyi, LanSi, and AAC will immediately respond to a call from Honor for collaboration; a startup might call ten times and still not get a response.
Fifth is mass production processes—controlling yield, production rhythm, and costs at a scale of tens of thousands of phones per day. Companies like Unitree, Songyan, and Tiangong are still gradually figuring this out.
Together, these five capabilities form a very strong hand. In my opinion, whenever you see a smartphone manufacturer announce it’s entering robotics, you can use this checklist to evaluate them: if they meet three or more criteria, they have a systemic advantage; if they only meet one or two, they’re just following the trend.
03
However, this list has its limitations—the true foundation of a robotics company lies in what’s not on the list, such as motors, gearboxes, force control algorithms, and simulation training.
Did you take care of all these key things over the past year?
I reviewed the timeline of Honor's robotics business: In April 2025, Honor established a new industry incubation department, which includes secondary divisions such as the Embodied Intelligence Lab and the Biomimetic Body Research Lab.
By May, at the Honor 400 series launch event, they publicly announced their entry into the robotics field for the first time. Until their championship win in April 2026, it was exactly 12 months altogether.
What does it mean to build a robotics company in one year? I compared a few companies to get a better understanding.
Unitree was founded in 2016 and spent seven years exclusively developing quadruped robots until releasing its first humanoid robot, H1, in 2023, during which time it built deep expertise in motors, gearboxes, and motion control technologies.
The Beijing Humanoid Robot Innovation Center behind Tiangong was established in November 2023, and the first version of Tiangong was released in just six months.
But keep in mind that this is an organization jointly established by UBTECH, Xiaomi, Beijing Electric Machinery, and Yizhuang Robotics, effectively leveraging the existing technological expertise from all four companies.
Songyan Power was founded in September 2023; with an average team age under 30, it secured nine funding rounds in just two and a half years, performed on the Spring Festival Gala, and won second place in a half-marathon—achievements widely recognized in the industry as "extreme speed."
In the industry, companies developing humanoid robots typically require two to three years to build up their technological foundation—Glory achieved this in just one year.
How exactly this year's time was spent is not fully clarified in Glory's publicly disclosed information.
What they mentioned—integrated joint modules, high-dynamic motion control algorithms, and multi-sensor fusion technologies—are all achievements; what they didn’t mention—motor models, gearbox suppliers, simulation training duration, and reinforcement learning frameworks—are the details behind them.
Is this information typically disclosed in the industry? I'm not sure.
Zhiyuan believes the most plausible explanation is this: The core motion system of Lightning is more like Glory defining the design, then partnering with suppliers in the industry to provide key components and jointly developing them together. Huake Lengxin’s liquid cooling pump is an example—the motor and reducer likely followed the same approach.
This is not a criticism. Given the one-year time constraint, this was the only viable approach.
Yu Shu, Tian Gong, and Zhi Yuan have taken a capital-intensive approach, developing motors from scratch internally—this kind of foundation takes a startup a decade to build up.
Honor adopts an integrated approach, selecting mature core components from the industry and using its smartphone manufacturers' system integration capabilities to assemble them into a robot.
Neither path is better or worse—they are completely different in character. One is steadily building a robotics company, while the other applies a smartphone company’s approach to solving robotics problems.
04
I wonder, if Honor doesn’t want to be a robotics company, what is its goal?
The answer is with another bot named "Genki-zai." It stands 136.9 cm tall, is silver in color, and designed to be especially friendly—it even won the Best Gait Award at this half-marathon, with a focus on human-robot interaction.
Glory itself also mentioned that "Yuanqi Zai" will be used in Glory's smart retail stores to help drive traffic and assist with product recommendations.
In addition to these two humanoid robots, Glory is also developing a quadruped robot dog named "An Zai" and a dexterous hand named "Lu Ban." When viewed together, this product lineup makes it clear that "Lightning" is the flagship showcase, while "Yuan Qi Zai" is the one being shipped.
Therefore, Glory is likely establishing a brand for "Yuanqi Zai."
By using the title of "Half-Marathon Champion" to deeply embed the five characters "Glory Robot" into consumers' minds, and then bringing "Yuanqi Zai" into its stores to work, this is its entire strategy; but this strategy alone isn't enough—the most intriguing part is that this path can only be walked by Glory.
I've reviewed it, and there are at least four structural conditions.
The first is our own offline channels. I looked into it, and Honor currently has about 5,000 experience stores and specialty stores in China. Placing one "Yuanqi Zai" in each store would immediately generate 5,000 orders—this is ready-made demand.
Yu Shu, Song Yan, and Zhi Yuan don’t have such direct channels—they either have to go to factories, research institutes, or trade shows, and for every new order, they have to renegotiate from scratch, which is especially difficult.
Second, Glory’s robots don’t rely on selling them for profit; startups selling robots must price each unit above cost, which is a fundamental principle of business.
The value is different—placing a "Yuanqi Zai" in a Honor store is like treating it as a fixed asset of the store, similar to display cabinets or store lighting, where the cost can be gradually amortized over time.
Its value lies in helping the store sell slightly more phones each day—this type of accounting is completely different.
The third point is that the brand budget can be directly repurposed; Honor’s annual brand marketing budget for its smartphone business runs into hundreds of millions, so allocating a portion of it to robotics development would impose no financial strain on the overall group.
Startups can't do this because every dollar they spend comes from funding, and they must account for how they use it in the market.
Lastly, there’s user feedback: when a robot is placed in a store, hundreds of customers come in and out each day, greeting it, asking questions, and taking photos—these interactions provide the most authentic data.
It’s as if “Yuanqi Zai” conducts user testing every day, and this testing scenario is more贴近真实的消费场景 than any enterprise tour offered by a single B2B client. This high-density feedback is simply unattainable by Unitree and Songyan.
Together, these four conditions create an absurdly wide moat that startups simply cannot cross.
It’s also worth looking at the approaches of other companies.
Unitree is seeking an IPO exit; it sold over 5,500 humanoid robots in 2025 and disclosed its IPO filing materials in March 2026, aiming to raise 4.2 billion yuan with a valuation of 12.7 billion yuan.
Songyan Power raised nine rounds of funding within a year, accumulating over 1.5 billion yuan, primarily through financing and emotional value. It previously launched the “Xiao Bu Mi” priced at 9,998 yuan, selling out its initial batch of 500 units within two days. Tiangong, on the other hand, leveraged state-owned platform support and completed its first round of market-based financing of 700 million yuan in early 2026, following a model combining a parent platform with an open-source community.
Three companies, three paths—all good paths—but none of them are the path to glory. The key difference is simple: they all need to bring in money from outside, but Honor doesn’t. It has stores.
Finally, a small detail for you to ponder:
The robot companies appearing on the 2026 Year of the Horse Spring Festival Gala are Unitree, Songyan, Zhiyuan, and Galaxy General—Tiangong and Honor are not included.
The logic is simple: the Spring Festival Gala is for the market, to gain exposure; the marathon is for the industry, to showcase engineering capabilities; and the stores are for Glory itself, to practice real, tangible execution.
Every company has found its own stage, but these stages are nowhere near each other. Don’t you think this approach is quite interesting?