During the recent Ma Year Spring Festival Gala, comedic actress Cai Ming shared the stage with a bionic humanoid robot, while "Xiao Bu Mi" robots navigated backstage, attracting photos and New Year greetings from various artists. These robots were developed by Beijing Songyan Power Technology Group Co., Ltd. (Songyan Power), representing its two main product categories: bionic humanoid robots and general-purpose bipedal humanoid robots.

Following this sustained brand exposure, Songyan Power's founder, Jiang Zheyuan, plans to further expand the company's scale in 2026. However, challenges remain. Automakers like Tesla and Li Auto are entering the field with capital and manufacturing advantages, squeezing space for startups. Despite this, Jiang maintains a "cautious optimism." He believes automakers' substantial capital allows them to pursue a "super long-term" strategy, whereas the immediate priority for startups is achieving commercial viability and self-sufficiency. Due to differing focuses on niche scenarios and strategic timelines, the two sides are not currently competing directly; real head-on competition might not emerge for another five to ten years.

The visibility gained from the Spring Festival Gala is becoming a catalyst for Songyan Power to accelerate its commercialization efforts. As the sector grows more crowded, attention is focused on whether Songyan Power can carve out a differentiated path to success in 2026.

In a recent dialogue, Jiang Zheyuan shared insights on Songyan Power's plans and the future of the humanoid robot industry.

**Pursuing a Differentiated Path**

For the Ma Year Gala, Songyan Power participated in a more distinctive segment: a comedy sketch. Comedian Cai Ming performed alongside a bionic replica, creating a moment where audiences couldn't distinguish the real from the robotic, avoiding the highly competitive market focused solely on physical movement control and becoming a highlight of the show.

"The biggest punchline of our sketch was the final scene where the real 'Grandma' Cai Ming appeared alongside the bionic version, making the audience realize that the Cai Ming on stage earlier was actually the bionic one. That was the core gag of the entire program," Jiang explained.

From Jiang's perspective, as a hard-tech company based in Beijing, the opportunity to appear on the Spring Festival Gala was not just a commercial opportunity but also stemmed from a sense of inherent mission.

This was not Songyan Power's first breakthrough into the mainstream. Earlier, at the 2025 Beijing Yizhuang Half Marathon & Humanoid Robot Half Marathon, Songyan Power's N2 robot unexpectedly won second place, gaining fame "overnight." However, at that time, the startup lacked the capacity to handle the sudden influx of attention and order inquiries. "There were regrets," Jiang admitted.

Learning from that experience, Songyan Power was better prepared this time, with its production team operating through the Spring Festival holiday. Jiang summarized the strategic focus for 2026 with two keywords: "penetration" and "exploration."

The former refers to deepening engagement in existing markets. Within its current commercial footprint, Songyan Power aims to capture a larger market share through more refined operations. The latter targets undefined "uncharted territories," with K-12 education and the consumer-grade market seen as key sources of growth.

Jiang believes that the several thousand well-funded top universities and key high schools are a fiercely competitive "red ocean" for all robot manufacturers. Beneath the surface, however, lies a vast number of county-level high schools, vocational colleges, and ordinary universities. These "mid-tier" and even "lower-tier" institutions have long been excluded from embodied AI due to high price barriers.

Songyan Power's humanoid robot "Xiao Bu Mi" is priced around ten thousand yuan, a strategy designed to match the purchasing power of these customers. "They might not afford robots costing tens of thousands, but a ten-thousand-yuan robot for students to experience technology is something they can likely afford," Jiang stated.

**Assessing the Demand for "Factory Work"**

With Songyan Power, Yinhe General, Yushu Technology, and Magic Atom appearing at the Gala, the humanoid robot sector is poised for an intense battle for scale in 2026. But this is not just a contest among startups; the battlefield is expanding to include automakers.

Overseas, Tesla fired the first shot. In January 2026, Elon Musk announced the discontinuation of the Model S and Model X—flagship models that established Tesla's reputation—to convert the Fremont, California factory into a production line for the Optimus humanoid robot. Tesla plans an annual production capacity of one million Optimus units, with production expected to begin by the end of 2026.

Musk's advantages extend beyond capacity; his plan involves Optimus directly utilizing core AI technology and sensors from Tesla vehicles. The competition is also heating up domestically. Li Auto CEO Li Xiang recently emphasized in an internal meeting the need to strengthen the company's positioning in embodied AI and enter the humanoid robot arena beyond automobiles.

This collective push by automakers into the market alongside startups stems from a shared anticipation of a market boom. Current expectations for humanoid robots extend beyond exhibitions and education to industrial manufacturing and home services.

UBS Group estimates global demand for industrial applications of humanoid robots will reach 30,000 units in 2026. Analyst Phyllis Wang suggests that as robotic "brains" evolve, their ability to perform repetitive tasks in workshops has significantly improved. Although only a few products can autonomously complete simple tasks currently, growth momentum is初步确立.

According to UBS data, Yushu Technology and Zhiyuan Robot collectively contributed nearly 10,000 units of the approximately 18,600 units shipped globally in 2025, dominating the market.

However, there is a noticeable gap between the optimistic reports from capital markets and the reality on factory floors. While UBS sees industrial scenarios as the next growth frontier, debate continues over whether humanoid robots are truly suitable for "factory work."

Visits to leading domestic textile, cosmetics, and smartphone manufacturers in 2025 revealed highly automated modern factories. On mature production lines, robotic arms handle precise operations, automated equipment completes multiple steps, and a single worker can monitor the entire process. An insider from a Shenzhen smartphone manufacturer explained that machines not only automate the manufacturing process but also enable self-correction and verification, improving efficiency and yield.

Furthermore, some phone manufacturers, leveraging their production line expertise, develop their own automated equipment to further enhance efficiency and reduce costs. In such highly specialized, efficiency-maximizing systems, the role of current-generation humanoid robots is limited.

"Short-term, we won't focus most of our efforts on industrial scenarios," Jiang noted.

**Data Remains the Biggest Challenge**

A more imaginative space lies in the home—scenarios where robots can organize clothes, cook meals, and provide emotional companionship, akin to a保姆. But home environments require robots to possess highly generalized capabilities.

An insider from a Zhejiang-based AI model company pointed out that homes are extremely non-standardized environments. Every floor plan, furniture arrangement, and even family living habit varies greatly, presenting an exponential challenge to a robot's generalization ability.

"The difficulty with models essentially stems from limitations in computing power and algorithms, which are not insurmountable. The real challenge is whether we can collect sufficiently large-scale and diverse data," Jiang admitted. He believes that once data volume breakthroughs occur, achieving embodied intelligence attributes is "not particularly difficult."

However, obtaining data within private home spaces is challenging. While devices like robotic vacuum cleaners have begun collecting some environmental data, the high-dimensional data required by humanoid robots involves deeper privacy boundaries.

Industry insights suggest some humanoid robot companies use synthetic data for training. For instance, some AI model companies specializing in home design possess vast amounts of physically accurate spatial data, creating virtual home environments to train robots. However, whether such data quality can meet robotic needs remains uncertain.

"Especially data related to internal family life is very sensitive," Jiang acknowledged. Under the premise of strict legal compliance and user privacy protection, establishing an effective pipeline for transmitting data from end-user homes to cloud training is a challenge for which "the industry doesn't yet have a particularly good solution."

Perhaps the grand narrative of humanoid robots fundamentally changing the world still requires time. But making them accessible tools for some people first is a crucial step toward that future. The conclusion of the Spring Festival Gala marks the beginning of the continued campaign for humanoid robots in 2026.

**Excerpts from the Dialogue:**

**All-Weather Tech:** Was the bionic humanoid robot a key reason for being selected for the Gala? **Jiang Zheyuan:** Yes, the bionic humanoid robot was indeed a major factor. The interaction between the bionic Cai Ming and the real one on stage, leading the audience to realize a robot was involved, was a significant highlight.

**All-Weather Tech:** What was the rationale for participating in the Gala? Was there internal debate? **Jiang Zheyuan:** Participation allows our technology, products, and capabilities to gain widespread exposure through this highly popular platform, helping us break through and enhance overall brand awareness. There wasn't major internal controversy.

**All-Weather Tech:** How do you balance scale and profit? **Jiang Zheyuan:** For a founder, these require trade-offs early on. This year, our priority is scale and maintaining healthy cash flow. Sufficient cash reserves are the foundation for everything.

**All-Weather Tech:** Any plans you can share for this year? **Jiang Zheyuan:** Two keywords: penetration and exploration. Penetration means increasing product adoption in existing scenarios through optimization. Exploration involves pioneering new areas like consumer-grade and K-12 educational robots. In AI, our goal is to prove our capability in building large models and gain recognition on relevant leaderboards. Technically, we'll participate in marathons and invest in R&D to solidify our position.

**All-Weather Tech:** Why focus on the education sector? **Jiang Zheyuan:** The industry is still nascent, with fragmented, small-scale applications. Targeting education is the most practical and feasible approach currently. While top-tier institutions number only in the thousands, there's a vast potential market in less affluent schools that could afford a ten-thousand-yuan robot for tech education.

**All-Weather Tech:** You previously required personal approval for reimbursements over 500 yuan. Is that still the case? **Jiang Zheyuan:** Yes, I'm still involved in detailed management, and the company maintains a frugal culture.

**All-Weather Tech:** With company growth, will your bandwidth be sufficient? **Jiang Zheyuan:** This is a concern. As headcount grows, managing every detail becomes harder. But I'm pleased we've brought in executives to assist with management. Our employees are passionate, loyal, and enjoy their work. My hope is for everyone to work happily, which actually reduces management difficulty as employees are self-motivated.

**All-Weather Tech:** Is now the time for humanoid robot scaling? **Jiang Zheyuan:** Not exactly. 2025 was initial exploration, with deliveries in the thousands. 2026 marks the start of true large-scale growth; we expect industry-wide sales to exceed ten thousand units this year.

**All-Weather Tech:** What prevents home helper robots currently? **Jiang Zheyuan:** It hinges on data breakthroughs. With sufficient large-scale, diverse data, implementation isn't overly difficult. Progress in robotic 'brains' isn't as fast as imagined; the industry lacks high-quality data and new collection methods. We are preparing solutions for data scarcity.

**All-Weather Tech:** Will automakers like Li Auto entering the robot space impact the industry? **Jiang Zheyuan:** I respect Li Auto. Whether they compete depends on if they target our current scenarios (e.g., companionship, education), which are minuscule compared to their auto business. They can afford a super long-term view with strong cash flow. We must focus on short-term commercialization and self-sufficiency. Significant competition is unlikely for perhaps five to ten years.