Dear Investors and Friends,

 

Humanoid robots endowed with increasingly general AI capability may become one of the largest industries globally at some point in the future. The logic is straightforward. Robotics is the way AI can address physical labour. Human labour is possibly the largest cost line in the global economy; the total addressable market is immense. The global labour market includes over 3.4 billion workers and about USD30 trillion of annual wages, around 30% of global GDP. In one bottom-up broker study, about 40% of workers may be substituted by robots. Of course, that does not mean this will all happen, much less that it will happen soon.

​

We can imagine an eventual installed base, decades from now, of humanoid robots measured in the billions rather than the millions. One forecast is for 8 million humanoids in operation in the US by 2040 and 63 million by 2050, while a China model points to 1.5 million units by 2030, 7.4 million by 2035 and 59 million by 2050. Elon Musk has gone further and argued that humanoids could eventually outnumber humans. It’s clearly a subject about which there is scope to dream.

 

Why humanoid rather than some other robotic form factor? Because the world is already built for humans. Factories, warehouses, kitchens, hospitals, malls, staircases, and tools are all designed around the dimensions, dexterity and movement of humans. Specialized robots, perhaps with wheels or four arms, may remain better at some specific tasks, but humanoids fit the world as it is.

 

We accept that humanoid robotics is very likely to be a huge industry. The harder question is timing. How long will it take? That is the key investment problem.

 

Timing matters for two reasons. First, the longer adoption takes, the harder it is to invest with confidence in a company today as a long-term winner in the distant future. Today’s leaders may be disrupted before the market reaches meaningful scale. Also, there are myriad potential profit pools in the value chain: semiconductors, models, simulation, controls, power systems, sensors, actuators, materials, batteries, assembly and maintenance. The slower the humanoid industry ramps up the harder it is to predict who will win. Second, even if we can correctly identify a future winner, the present value of distant cash flows is much lower. A very large cash flow stream twenty years from now may be worth relatively little today. It’s a timeless lesson that a huge TAM and an attractive investment are not the same thing.

 

Tesla’s long-term incentive plan points to a long, slow ramp. Elon Musk’s controversial compensation plan has a 10-year structure beginning in 2025, with one milestone tied to delivery of 1 million “bots.” That suggests that an ostensibly bullish industry insider defines success as 1 million cumulative units by 2035. That implies sales of only about 200,000 units in 2035 if we assume a linear ramp. To us, rapid mass adoption, which we think is possible but not certain, would be much faster than this.

 

Historical analogies for transformative technology scaling

Some historical comparisons may be helpful. Let’s consider the emergence of railroads and then automobiles in the United States. Each is a case study on the journey from technical feasibility to commercial usefulness and finally mass adoption. Steam railroads were technically feasible in the 1820s. The first commercial railroad in the United States opened in 1826. By 1840 the country had almost 3,000 miles of track, and by 1860 about 30,000 miles. Yet the first transcontinental link was not completed until 1869. It took roughly 40 years from first commercial operation to a truly national network. The growth of railroads did not just require locomotives and track, it also needed financing, rights of way, standardized operating practices, timetables and even standardized time zones.

​

Cars followed a similar but faster pattern. Automobiles were on US roads by the 1890s, and Ford introduced the Model T in 1908. The growth inflection was the moving assembly line in 1913. US motor vehicle registrations rose from less than 500,000 in 1910 to nearly 10 million in 1920 and more than 26 million by 1930. It took about 15 years from a commercially viable mass-market product to truly broad adoption.

 

Railroads and cars each required three sequential transitions. First, the technology had to work. Second, it had to become economically useful. Third, the surrounding system had to reorganise around it. Humanoid robots should follow the same path. We are moving from stage one toward stage two: from “can it work?” to “can it perform valuable tasks?” The key bottlenecks today are hand dexterity and physical AI sophistication. Stage three, diffusion, will likely be slower: factories, maintenance systems, safety processes and financing models all need to adapt. Presumably, humans will also be more vocal about having their labour substituted than horses were in the early 1900s!

 

Early evidence around humanoid volume scaling

Near term, the industry is at the stage of pilots, sample orders, controlled deployments and low-volume industrial use cases. Repetitive, limited tasks in factories and warehouses are an easier problem than the variability of everyday household life.

​

Tesla has publicly pointed to an eventual Optimus cost of roughly $20,000 to $30,000. A recent GGII industry survey argues that average humanoid unit cost in China fell to about $15,000 from $20,000 in 2025—though these robots are not directly comparable to Optimus. Meanwhile the average payback period shortened to around 12 months from 18 months last year and 48 months in 2023. A one-year payback is attractive for businesses. GGII now projects 2026 shipments of about 80,000 units, up sharply from 2025, with a possibility of 100,000 if pricing continues to fall.

​

GGII says factories will account for roughly 72% of 2026 demand. That said, there are challenges. Fault rates in unstructured environments are still said to be around 8% to 12%, versus roughly 2% to 3% for conventional industrial robots. Surveyed companies in China still see about $10,000 as the real mass-adoption price band, though that obviously will vary depending on robots’ capabilities. Today’s robots are still limited in what they can do. Hands remain the key bottleneck to useful work today: robots can dance, but once asked to manipulate real objects reliably, performance remains weak.

 

The global ecosystem: the US leads in the brain, China in much of the body

The humanoid robotics ecosystem today has an obvious US-China split. The US is strongest in the “brain.” Nvidia is central because advanced GPUs, simulation environments and robotics development tools are foundational inputs to embodied AI. Tesla remains the most important listed full-system humanoid player. Even local Chinese industry experts believe that Optimus leads local start-ups in financing capacity, physical AI models, system integration capability, and manufacturing know-how.

​

China’s role is more on the component side. But we’d argue that in public markets at least, the Chinese humanoid value chain is more investable. China’s supply chain covers a wide variety of components: actuators, motors, reducers, screws, sensors, magnets, castings, bearings etc. There is a lot of overlap between China’s EV component suppliers and the nascent humanoid value chain. If cars are evolving into robots with wheels, then robots are in some sense the cars of tomorrow with legs and arms.

 

Our research on the component value chain has led us to focus on actuators. For readers less familiar with the terminology, an actuator is the mechanism that creates movement at a joint. The actuator system comprises screws, reducers, motors, sensors, ball bearings and encoders. A reducer is a gearbox that lowers motor speed and increases torque. An encoder measures position and speed and feeds that information back to the control system. Actuators may account for roughly half of total robot bill-of-materials cost, making them arguably the single most important physical profit pool in the component stack.

​

 

​

​

​

​

​

​

​

​

​

​

​

​

​

 

Variis’ Eko Yin recently travelled to China, and, while in Shenzhen at an industry conference, got a chance to check out some Chinese robotic products.

​

​

​

​

​

 

 

 

 

We recently invested in Zhejiang Sanhua Intelligent Controls

During 1Q26, we invested in Sanhua, which is, in our judgment, an attractive listed way to benefit from the humanoid future we expect. The company holds early leadership in humanoid robotic actuators.

Sanhua is not a speculative humanoid start-up. It is an established global leader in refrigeration control components and automotive thermal management. Our work highlights its strength in both areas, its broad customer set, and its overseas manufacturing footprint. The company already has deep competence in precision electromechanical control, large-scale manufacturing and demanding customer qualification.

 

We expect Sanhua to become a major supplier of robot actuators. Sanhua’s role is more likely to be as an integrated actuator/module assembler, rather than a supplier of every internal subcomponent. Tesla is sourcing actuators for Optimus from Sanhua and its competitor Tuopu.

 

The existing Tesla relationship is significant. Sanhua supplies Tesla with electronic expansion valves and thermal modules for EVs. The Tesla auto business today accounts for about 15% of Sanhua’s total revenue.

 

Sanhua has announced an expansion plan of more than RMB5 billion, including more than RMB3.8 billion specifically for robotic actuators and controllers. Thailand is Sanhua’s offshore production location for robot actuators.

 

Sanhua also fits our investment philosophy closely. We look for businesses with durable competitive advantages, management teams capable of converting those advantages into long-term value, and stakeholder characteristics that protect the company’s licence to operate. Sanhua already has those traits in its existing franchises: valuable industrial intellectual property, high share in key components, deep customer integration, a global manufacturing footprint and a culture built around technical execution and customer service. For Sanhua, the humanoid robotics opportunity extends the reinvestment runway of an already high-quality business into what may become a very large adjacent market.

 

We also like today’s valuation. The current share price embeds core business plus limited robot option value rather than demanding that we fully price the humanoid success case. The stock price today requires the existing refrigeration and auto-thermal businesses to continue compounding, with robotics adding limited option value over time. In our base case, under which the stock would double over five years, even by 2030 robot actuator revenue would still be only about $600 million, or roughly 7% of forecast HVAC plus EV revenue that year. It is a moderate contributor to growth sitting alongside two already large businesses. $600 million of revenue would be about 1,600 sets per week at $7,000 per set. That is only enough for about 83,000 humanoid robots per year and is broadly consistent with a gradual industry ramp rather than an aggressive near-term mass-market scenario.

 

Sanhua trades in both Shenzhen, as an A share, and in Hong Kong, where it listed in 2025. The two shares have an identical underlying claim on the business. With Mainland investors much more bullish, the A share line trades at a 55% premium to the Hong Kong line. Naturally we have invested in the Hong Kong line.

Why suppliers may be the better route than robot makers

 

We think it makes more sense to invest in suppliers than in robot makers. The reason is not that robot makers will not matter. The problem is that there are no obvious listed pure-play winners today. After examining robotics companies in China, including UBtech and Unitree (expected to IPO in 2Q26), we believe it is too early to assess the durability of their business franchises.

 

By contrast, many listed suppliers already have real businesses and transferable capabilities. The listed Chinese supply chain can be thought of in functional buckets. There are actuator and subsystem suppliers such as Sanhua, Tuopu and Hengli. There are motor suppliers such as Inovance, Moons and others. There are gear, reducer and motion-control names such as LeaderDrive, Shuanghuan and Zhongda Leader. There are a range of sensing, bearing and structural-part suppliers. Several of these companies are either already on our Focus List or under active consideration.

 

Sanhua’s own stock market record versus that of its OEM customers in China demonstrates the potentially superior investment returns in the upstream supply chain. In both HVAC and automotive thermal management, Sanhua has historically been a better way to participate in end-market growth than owning many of the final assemblers themselves. Chinese HVAC OEMs and especially auto OEMs have often had to fight through price competition, product cycles, channel pressure and heavy capital intensity. Sanhua, by contrast, has sold critical components into multiple customers, benefited from rising content per unit, and avoided taking single-brand risk. Since its Shenzhen IPO in 2005, Sanhua’s shares have dramatically outperformed many of its OEM customers and are up 70x before considering dividends.

 

Longer-term implications: what else should we be thinking about

With the ongoing rapid development of super-capable LLMs, AI is set to radically change knowledge work. A debate is raging about whether AI will manifest as a massive productivity boost, potentially changing and even expanding demand for knowledge work, or whether a large portion of human knowledge workers will prove redundant. Humanoid robots pose an analogous question for blue-collar workers. Sure, having a robot fold your laundry may sound nice, but if you are a low-income factory or agricultural worker, that robot may not look so friendly.

​

In EM, the relative size of the blue-collar labour pool is much larger than in DM. Almost nine-tenths of global blue-collar workers are in EM, where they represent more than half of the labour pool. In DM, blue-collar workers are less than one-fifth of the total. The potential impact of humanoid robotics in EM on employment, especially in societies that do not have robust social safety nets, could be enormous. It’s important for us that even while we are thrilled by the near infinite possibilities of humanoid robots, we are also keenly aware of disruptive and possibly harmful impacts as this industry scales.

 

Humanoid robotics implications for Variis’ investment strategy

The long-run TAM for humanoid robotics appears enormous. The world’s stock of physical labour is too large a cost for this not to become a major industrial category, assuming the technology becomes sufficiently capable and affordable. Humanoids are likely to be a particularly important form factor because the world is already built around humans. But from an investment perspective, the question is timeline, not TAM.

​

We prefer to invest in the humanoid robotics trend through high-quality listed suppliers with real businesses today and likely participation in the future humanoid value chain, rather than through speculative attempts to identify the definitive robot OEM winner too early.

 

Sanhua is our clearest example of this approach. It is already a strong industrial company. It is close to Tesla. It has relevant manufacturing know-how. It is investing serious capital in robotic actuators and controllers, including capacity preparation in Thailand. And it offers exposure to a critical future industry without requiring us to set aside our discipline around business quality and valuation.

 

It is also worth noting that humanoid robots may eventually become an additional demand driver for the broader AI picks-and-shovels value chain. A scaled installed base of humanoid robots would require massive incremental compute. Just to give a sense: humanoid robots have roughly 10x the memory and logic semiconductor content of an iPhone, and annual humanoid sales volumes could eventually rival today’s smartphone volumes at a billion units per year. In that sense, the humanoid theme may reinforce, rather than sit apart from, the existing AI infrastructure opportunity set. That matters for us because a significant part of the strategy today is invested in AI picks-and-shovels businesses.

 

Conclusion

We launched the Variis strategy on 4 Oct 2023. Variis Partners LLP was set up in November 2022, and we spent nearly a year preparing for launch. After 3.5 years working together to build Variis, we are proud of the foundations of the business and the long-term relationships we are fostering.

​

Finally, we remain deeply grateful for your trust and support. Your partnership enables us to stay focused on what we do best: identifying exceptional businesses and investing with conviction. We look forward to updating you again next quarter.

 

Thank you for your continued interest and support!

​

Leila, Eko, Rufus and Jamie

 

Disclaimer

 

FOR PROFESSIONAL INVESTORS AND ADVISORS ONLY

The contents of this document are communicated by, and are the property of, Variis Partners LLP. The information and opinions contained in this document are subject to updating and verification and may be subject to amendment. No representation, warranty, or undertaking, express or implied, is given as to the accuracy or completeness of the information or opinions contained in this document by Variis Partners LLP or its directors. No liability is accepted by such persons for the accuracy or completeness of any information or opinions. As such, no reliance may be placed for any purpose on the information and opinions contained in this document. The information contained in this document is strictly confidential and is not intended to be advice or an offer or solicitation to invest. The value of investments and any income generated may go down as well as up and is not guaranteed. Past performance is not a reliable indicator of future results.