With the witness of 2,500 audiences from over 30 countries and regions, &#34;ZhiHuiJun&#34; Peng Zhihui completed his second public speech in 2026, the last one being on March 30 during the ceremony for the 10,000th robot mass production line of ZhiYuan.
More than 2,500 audiences filled the entire venue, and a display board at the entrance listed a partner list covering six dimensions. I used AI to count, and there are roughly around 170 companies, with at least 30% coming from overseas.
This actually releases a very important signal: ZhiYuan's next story will focus on overseas markets.
&#34;The real watershed of embodied intelligence is not just about AI models entering the physical world, but rather our products beginning to enter real workflows. Our logic must shift from selling robots in the first two years to delivering results.&#34;
This genius from Huawei stood before global partners to narrate ZhiYuan's new logic. Together with founder Deng Taihua, they spent nearly three hours painting a blueprint for what they call the &#34;Year of Deployment&#34; for embodied intelligence, along with a hidden goal of transforming from 1 billion in productivity over 3 years to 100 billion over 8 years.
<h2>The Era of &#34;Selling Robots&#34; is Over</h2>
When Peng Zhihui took the stage, Deng Taihua had just finished his speech. Their styles of expression were different, but both mentioned one point: embodied intelligence is transitioning from &#34;development mode&#34; to &#34;deployment mode.&#34;
Peng Zhihui's entry point was a statement by Jensen Huang at GTC: &#34;Tokens are the currency of the AI era.&#34; He extended this logic to the physical world: as long as a robot operates in a real environment, it is constantly perceiving, reasoning, deciding, and controlling, with continuous token flow behind it.
&#34;The task space of embodied intelligence is the sum of the digital world and the entire physical world. Whoever can first complete this closed loop will have the opportunity to define the next generation of productivity platforms.&#34;
He does not believe this is just a grand narrative.
Tesla has halted the production lines of Model S and Model X to make space for Optimus robots; NVIDIA stated, &#34;Physical AI has arrived&#34;; even Boston Dynamics, known for its technical prowess, has begun to emphasize &#34;stepping out of the lab into reality.&#34; &#34;The capital market is no longer willing to pay for demos.&#34;
In Peng Zhihui's view, 2026 is a breakthrough window for three reasons: large models enable robots to understand and comprehend the world, and an open-source ecosystem has formed; the robots themselves have crossed the reliability threshold, achieving stable operation 7×24 hours; and the data flywheel is beginning to accelerate—more deployed robots lead to more data collection, stronger model capabilities, and a positive feedback loop is starting to form.
<h2>One Body, Three Intelligences</h2>
The core framework of Peng Zhihui's speech is &#34;One Body, Three Intelligences,&#34; a term that Deng Taihua repeatedly used in the first half.
&#34;One Body&#34; refers to the robot itself—different forms of physical carriers adapt to different scene boundaries, serving as the foundation for all upper-level intelligence. &#34;Three Intelligences&#34; are respectively motion intelligence (basic execution capability), operational intelligence (creating labor productivity value), and interactive intelligence (integrating robots into human workflows and service flows)
He placed this framework within an evolution coordinate from L1 to L5, analogous to the maturity grading of autonomous driving, from rigid pre-programmed entities to compliant control, to high-mobility global adaptability, ultimately reaching a &#34;superhuman&#34; state.
The PPT emphasizes a concept that Zhiyuan may be the only company with four layers of full-stack capabilities and has completed mass production verification on a scale of ten thousand units.
Peng Zhihui emphasized that the significance of this framework lies not in technical benchmarking, but in its definition of a new way of organizing productivity: &#34;Embodied intelligent robots will become new laborers, models will become new infrastructure, open platforms will become new entrepreneurial soil, and ecosystems will become new growth engines.&#34;
Driving the &#34;three intelligences&#34; are four core models developed by Peng Zhihui's team: the BFM Behavioral Foundation Model, the GCFM Generative Operation Control Model, the GO-2 Operational Large Model, and the WITA-Omni End-to-End Interaction Large Model.
Taking the BFM Behavioral Foundation Model as an example, this is the foundation of motion intelligence, similar to the pre-training of large language models: training a unified promptable behavioral foundation model using over 100 million frames and 700 hours of human motion capture data, enabling robots to adapt to new tasks with zero or few samples.
Emphasizing model capabilities aligns with the details of Deng Taihua's earlier speech, where he stressed that Zhiyuan is an embodied model company, with 70% of its R&amp;D personnel, 73% of whom are engaged in the development of the cerebellum and cerebrum.
According to data disclosed on-site, in December 2024, Zhiyuan's scale was still only over 500 people, and by April 2026, the number increased to 1,570.
<h2>Data Flywheel</h2>
Behind all of Peng Zhihui's technical narratives lies an underlying thread: data.
He likened Zhiyuan's entire layout to a multi-curve problem that mutually constrains each other: immature algorithms cannot iterate without a mature entity, models cannot reflect data value without sufficient data, and the ecosystem cannot turn the flywheel without openness. &#34;Once these four things are established simultaneously, capabilities will be rapidly amplified.&#34;
What he referred to as the &#34;deployment state data flywheel&#34; is a key link in the entire technical strategy.
Every moment robots work in factories, shopping malls, and restaurants generates high-quality training data; the more robots deployed, the faster the flywheel turns, the quicker the model iterates, and the stronger the robot's capabilities, which will attract more scene deployments.
To this end, Zhiyuan has specifically incubated the Mifeng Data Company, aiming to achieve 10 million hours of effective data this year, of which 2 million hours will be high-quality real machine data, and has established multiple data generation bases globally. Peng Zhihui stated that Mifeng's data will not only serve Zhiyuan internally but will also be opened to the entire industry in the future: &#34;Creating value for innovation across the entire industry.&#34;
The day before, Zhiyuan partner and Mifeng's chairman and CEO, Yao Maoqing, also detailed the newly launched data trading platform and data collection devices during an on-site exchange, revealing that they would consider using crowdsourcing to collect data.
An industry practitioner said that this is a boon for Zhiyuan, which has a large amount of data, &#34;Zhiyuan has so much data that needs to be traded.&#34;
However, the closed loop of the data platform also needs to address existing issues, including the problem of standardization of data. &#34;If it is real machine data, it is still not universal.&#34;
On-site, I asked Yao Maoqing what the first problem to solve would be for a one-stop data platform to succeed, and his answer was the value closed loop. &#34;Those who purchase data will initially ask one question: How is your data quality? And how do you prove that your data quality is relatively good?&#34;
<h2>&#34;358 Blueprint&#34; and 100 Billion Target</h2>
If Peng Zhihui represents the confidence in the technical route, then Deng Taihua's &#34;358 Blueprint&#34; released in the first half translates this confidence into business language.
According to the planning of Zhiyuan's management, the so-called X curve will be achieved within three years—allowing robots to move like humans, achieving productivity entry, and Zhiyuan will also complete 1 billion in revenue (already achieved). By the end of 2027, within five years, operational intelligence will be implemented, with the number of deployments increasing from thousands this year to tens of thousands next year, and Zhiyuan's revenue will surpass 10 billion.
By 2030, on the eighth anniversary of Zhiyuan's establishment, the deployment data flywheel and algorithm innovation will jointly promote &#34;intelligent emergence,&#34; entering the third growth curve.
&#34;If that day really comes, how much revenue Zhiyuan has will no longer matter. What matters is that we have welcomed a historical moment of productivity leap.&#34; Deng Taihua said when announcing this plan.
Whether it is the 358 Blueprint or specific numerical targets like 100 billion, they are all supported by a series of technological foundations released today and cannot be separated from the ecological and partner promotion.
The day before the speeches of the two individuals, an insider expressed great joy about this year's performance, saying, &#34;Our performance this year is very good, but I cannot tell you the specific numbers.&#34;
This can also be seen as a sign of the flywheel starting to turn, but how fast it can turn will be revealed when the data is announced at the end of the year.
Risk Warning and Disclaimer
The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this is at your own risk

机器人50

300024

00370

In a public speech, Huawei's Peng Zhihui presented the future blueprint of Zhiyuan, emphasizing that embodied intelligence will shift from the development stage to the deployment stage. He mentioned that robots will enter real workflows, aiming for a productivity growth from 1 billion in 3 years to 100 billion in 8 years. Peng Zhihui believes that whoever can establish a closed loop between the digital and physical worlds will be able to define the next generation productivity platform. He also pointed out that the capital market's interest in demonstration projects is waning, and 2026 will be a key year for breakthroughs

华尔街见闻资讯

SIASUN

HK ROBOTICS

Huawei

深圳海清智元科技股份有限公司

Tesla

NVIDIA

NVIDIA Corporation

Huawei's genius youth wants robots to go on the production line to "screw bolts"