In the first half of 2026, AgiBot ranked second on the global humanoid robot revenue list with a revenue of CNY 610 million. Official data shows that its cumulative production of humanoid robots has exceeded 20,000 units. While the outside world focuses on its shipment pace and industrial scenario deployment, it is easy to overlook another more hidden front: thermal management.
In October 2026, Wenqing Zhikong received a strategic investment from AgiBot; almost at the same time, the market discovered that AgiBot also invested in another thermal management startup, Haina Zhileng. Looking further back, AgiBot's self-developed PowerFlow joint motor has long integrated liquid cooling circulation into its products. Thermal management supply chain enterprises such as Tongxing Technology, Dongyangguan, LINGYI iTECH, and Yonggui Electric have also been bound to AgiBot through joint venture platforms or strategic partnerships.
From AgiBot's moves, let us discuss what thermal management really means for humanoid robots, and why it is becoming the most underestimated threshold on the eve of mass production.
01. AgiBot's Liquid Cooling Chess Game: Simultaneous Moves in Capital, Products, and Ecosystem
The most noteworthy aspect of AgiBot's foray into liquid cooling is that it has not placed all its bets on a single technology route or a single supplier.
First is capital positioning, investing in two companies with a 'one stable, one fast' strategy.
Wenqing Zhikong follows a whole-machine system-level liquid cooling route. It transplants the systematic thermal control approach of NEVs to robots: the liquid cooling plates for joints, controllers, and batteries converge into a unified pipeline, transported by micro-pumps to heat dissipation terminals, supporting zoned temperature control. This solution addresses the basic temperature control during 8 hours of continuous robot operation, keeping the whole-machine temperature within a stable range.
Haina Zhileng was founded in February 2026, completing its angel round and angel+ round of financing in May and August respectively, with AgiBot's investment subsequently entering the public eye. It targets another dimension: chip-level instantaneous high heat. When the robot jumps or carries heavy loads, it bursts with huge power consumption within seconds, causing instantaneous heat accumulation. Haina Zhileng's solution uses high-heat-capacity modules to quickly absorb heat during sudden load increases and slowly release it when the load drops, while developing 3D conformal cold plates adapted for circular and irregularly shaped joints, allowing the coolant to exchange heat along the curved surfaces. Simply put, Haina Zhileng manages instantaneous thermal shocks under extreme conditions, forming a 'one stable, one fast' complementarity with Wenqing Zhikong.
Second is product integration, with liquid cooling directly entering the joints.
On AgiBot's self-developed PowerFlow joint motor, the liquid cooling circulation heat dissipation system has been implemented as part of the product: the continuous operating temperature is reduced by 20%, the peak torque density reaches 350N·m/kg, the peak torque exceeds 350N·m, and the weight is only 1.6kg. The dual-arm load capacity reaches 40kg, which is sufficient to perform high-intensity tasks such as bolt tightening on automotive final assembly lines and 3C precision assembly. This means that liquid cooling in the AgiBot system first delivers performance metrics through the integrated heat dissipation solution for high-power joints.
Finally, ecosystem synergy, bringing mature thermal management capacity into the supply chain radius.
Tongxing Technology is bound to AgiBot through Shanghai Jushen Huizhi, focusing on micro liquid cooling and biomimetic sweating cooling. The former targets narrow, high-heat-flow spaces such as joints and computing power modules, while the latter replicates the sweat gland mechanism, suitable for extremely small spaces like dexterous hands. Dongyangguan and AgiBot established a joint venture to produce cold plates, heat dissipation modules, quick connectors, and other cold plate liquid cooling components at the Shaoguan base; the long-term plan of LINGYI iTECH's joint venture includes heat dissipation solutions, and it already possesses CDU, liquid cooling plate, and liquid cooling module capabilities; Yonggui Electric has signed a deep strategic partnership with AgiBot, covering the connectors and wiring harnesses in the liquid cooling pipeline.
AgiBot's venture capital team judges that 2026-2027 is a critical introduction period for joint R&D between liquid cooling suppliers and whole-machine manufacturers. In the second half of 2027, liquid cooling solutions are expected to be first implemented in robots for industrial scenarios, followed by penetration into motion intelligence, interactive intelligence, and home scenarios. This pace aligns with AgiBot's own mass production ramp-up: as of September 2026, cumulative deliveries have exceeded 20,000 units, and Elf G2 has completed 6 consecutive days of 10-hour daily in-line mass production verification on Longqi Technology's production line.
Hangjiashuo's judgment: The brilliance of AgiBot's layout lies in locking in the two technology routes of 'steady state' and 'transient state', using self-developed joints to deliver short-term performance, and then using ecosystem cooperation to solve manufacturing and supply chain issues. Thermal management has not been entirely internalized, but key nodes have been occupied in advance.
02. Thermal Management is Not an Accessory, but a Prerequisite for Mass Production
If we lay out the robot's energy ledger, we will find that the position of thermal management is far more forward than an 'accessory'.
Industry data shows that the energy conversion efficiency of humanoid robots is only 5%-10%, with about 90% of the energy ultimately converted into waste heat. Taking Tesla's Optimus as an example, a single 100-meter movement consumes a total energy of 8.28 million joules, with effective mechanical work being only 2,350-4,700 joules. The total waste heat exceeds 40,000 joules, and the local heat flow density of the joints exceeds 100W/cm², which is more than 100 times the average level of the human body surface. This heat is concentrated in three places: joint actuators contribute 50%-60% of the whole-machine heat, with 20-50 actuators packed with motor windings, drivers, and reducers; the computing power unit accounts for about 20%, with the instantaneous power consumption of GPU/NPU reaching 50-300W; the power supply system contributes another 12%-15%.
More trouble is that these heat sources are stuffed into a closed cavity of human-like size, without air ducts like server cabinets, nor heat dissipation space like EV chassis. The consequence is: once the chip temperature exceeds the threshold, it triggers frequency reduction. The industry's '10℃ rule' means that for every 10℃ increase in the operating temperature of electronic components, the lifespan is shortened by about 50%; once the joint temperature breaks through the 80-100℃ range, torque accuracy drops, materials age faster, and the system can only forcibly reduce power for self-protection, resulting in slower and stuttering movements, or even direct shutdown in severe cases.
Why liquid cooling?
Actually, air cooling has not been without effort, but two structural constraints of humanoid robots have pushed air cooling to its physical ceiling: the gap in the joint cavity of dexterous hands may be less than 2mm, making it impossible to fit traditional centrifugal fans; the whole-machine power consumption is also climbing rapidly, with the heat dissipation power of a single unit rising from 120W in 2023 to 650W in 2025, while the power limit of air cooling is about 800W — humanoid joint servos have generally broken through 1200W. The efficiency gap of liquid cooling is of an order of magnitude, with heat dissipation efficiency reaching 10-50 times that of air cooling, capable of reducing the core temperature from 100℃ to below 60℃.
More critically, thermal management for humanoid robots cannot be retrofitted later. Servers purchase heat dissipation accessories after the main body is built, but humanoid robots must simultaneously plan pipelines, cold plates, and pump valves during the main body design phase — it changes from an 'accessory' to an inseparable part of the whole-machine structure. This also explains why AgiBot needs to simultaneously integrate joint liquid cooling, invest in system-level and chip-level thermal management companies, and attract supply chain enterprises to establish joint venture platforms in the early stage of product definition. Because thermal management is not a lesson to be made up after the robot is built, but a prerequisite that must be considered during the design phase.
03. Thermal Management is Becoming a Collective Consensus Among Whole-Machine Manufacturers
Broadening the horizon, thermal management becoming a prerequisite for mass production is already a collective consensus among whole-machine manufacturers.
The lesson from Tesla's Optimus is the most direct. In the durability batch test in October 2025, the robot commonly experienced high-temperature alarms for joint motors during prolonged walking, and insufficient heat dissipation design redundancy was identified as the core defect for suspending mass production. The Gen 2 version shuts down due to overheating after 8 hours of continuous operation, with the mean time between failures staying at 8 hours. Musk later reduced the power consumption to below 100W, targeting over 8 hours of continuous operation, which is essentially bypassing the heat dissipation bottleneck by reducing heat generation.
Inovance, on the other hand, provides a direct industrial-grade answer for liquid cooling. Its wheeled humanoid InoBot W1 directly integrates the liquid cooling circuit into the high-computing-power edge-side chip Thor, enabling long-term full-load operation in environments above 40℃, stably outputting 2070 TOPS of computing power. The whole machine reaches the IP54 protection rating, and with dual-battery hot-swapping, it supports 7x24 hours of continuous production. Here, liquid cooling is not a performance enhancement item, but a guarantee of availability in industrial scenarios.
Yushu's choice also illustrates the point. Its prospectus mentions heat dissipation management multiple times, with solutions covering the native integration of 'air cooling + phase change micro water cooling', embedding heat dissipation functions into the joint and body structure, rather than retrofitting. At the 2026 humanoid robot half-marathon, Rongyao 'Lightning', Noetix Robotics, Yushu, and Tiangong almost collectively shifted to liquid cooling or a liquid cooling + air cooling combination. Rongyao 'Lightning', equipped with Huake Lengxin's high-speed levitation micro-pump, won the championship with a time of 50 minutes and 26 seconds. That micro-pump is only 30mm in size, weighs less than 100g, yet can withstand 500G impact and exceed 20,000 RPM.
Shenwan Hongyuan's calculation provides a quantitative anchor: by 2030, the global active heat dissipation market size for humanoid and quadruped robots will be about CNY 18 billion, of which the penetration rate of liquid cooling solutions will continue to leap with the increase of whole-machine power consumption. Thermal management is no longer an 'accessory' market, but a structural incremental track that scales synchronously with whole-machine shipments.
04. Hardware is the Entry Ticket, while Algorithms and Data are the Endgame
Returning to AgiBot's layout, there is another judgment worth highlighting separately: hardware is easily leveled by the supply chain, and the true barrier lies in the 'thermal-control coupling algorithm' and multi-condition heat generation database.
For the same robot, jogging, carrying heavy loads, and continuous operation in a 40℃ workshop have completely different heat generation distributions. The heat dissipation system cannot operate independently; it must be deeply linked with motion control — what the robot's next move is, the thermal management system needs to predict in advance and dynamically allocate heat dissipation power.
AgiBot sets the liquid cooling implementation pace to start with industrial scenarios in the second half of 2027. The logic is that industrial scenarios will not pay for 'running fast', but only for continuous work without failures. Jingling G2's 6 consecutive days of 10-hour daily in-line verification on Longqi Technology's production line is closer to the real test site for thermal management than any track record.
The competition for humanoid robots is shifting from 'whose demo is more explosive' to 'who can continuously run a full shift on a real production line'. Thermal management is not the most conspicuous link on this path, but it is likely the first to keep most players out of the door of mass production. Liquid cooling is the entrance, and cold plates, pump valves, and pipelines are the entry tickets; but what truly determines who can go to the end is the thermal data and temperature control algorithms accumulated watt by watt in real working conditions.
Only when a robot quietly finishes tightening the 1000th bolt in the factory with the joint temperature always stable below the threshold, is the value of thermal management truly seen.
This article is compiled based on publicly available network information, for reference only, and does not constitute investment advice.