Let's jump back in time to a world before electricity. Then electricity arrived, and Siemens was there from the beginning, scaling grids, trams, and power plants globally. Now another general-purpose technology has arrived: artificial intelligence. AI will be as transformative as electricity, maybe even more powerful. We have entered the industrial AI revolution, and China is where much of this is happening at scale. Siemens is building the industrial AI operating system, an intelligence layer connecting hardware, software, and data across the industrial value chain. This changes how we design, engineer, and operate machines. In manufacturing, AI increases productivity and energy efficiency; in life sciences, it shortens time from lab to patients. Bringing AI to the real world requires a technology stack: data, software, and hardware. We break down data silos, create digital twins, and use AI to simulate and optimize. For example, Foxconn uses our digital twins for AI factories, and PepsiCo increased efficiency by 20% in three months. Hardware is crucial: our industrial PC with Nvidia GPU brings AI to the shop floor for real-time inferencing. At Audi, AI checks welding seams; in Chengdu, it guides robots. Domain know-how is equally important. Our 1,500 AI experts and engineers work across 30 verticals. We are developing AI assistants that reduce engineering time by up to 40%. Partnerships are key: we work with Microsoft, AWS, Alibaba Cloud, and Nvidia. With Nvidia, we accelerate chip design, simulation, and build AI factories. China is where industrial AI will scale fast. Today we launch 26 new products developed in China for China, available on the Siemens Xcelerator marketplace. Over half a million users are registered, and two-thirds of offerings are AI-related. Partnership matters. Please welcome Alibaba chairman Joe Tsai.
Thank you, Roland. Thank you for having me. I actually wanna start by saying two thank yous.
Number one, this is the first time you're doing the tech conference in China, and it shows that Siemens is committed to China, not just as a market but for inspiration and innovation. Thank you for your commitment to Chinese talent and entrepreneurialism. Thank you very much.
Thanks goes to my team here.
And the other thank you is of course, our partnership. You have mentioned it already. We've had a great working relationship with your team in China, so I wanted to thank the team here, and thank you for supporting it. I think we hope to do more in the future. Thank you.
Thank you for the partnership. So this is when two strong partners are coming together, and actually I have an advantage because we had a chance to chat for the last one and a half hours behind the scene. I could go on and on and on, and we had tried to bring a little bit of that to the stage right now. So let's start. Alibaba is in its own transformation, obviously, just towards a cloud and AI. So what is the strategy behind, and what is the progress you made so far?
Most of you know Alibaba as an e-commerce company, but we got into the cloud business almost 17 years ago to manage our data explosion. We developed proprietary technology, and that became our cloud foundation. More recently, demand is coming from AI. We've designed our cloud infrastructure to handle complicated AI workloads. The strategy is that everything will be driven by AI. Look at agentic AI: large language models used to be encyclopedias, but now they can handle entire tasks without prompting. Agents can plan, reason, and have memory. Alibaba is there to support companies using AI with our cloud infrastructure.
Right. And we enjoyed already, so our Siemens Xcelerator platform runs on Alibaba Cloud. We have a lot of cooperations. You are building data centers and AI factories using Siemens technology. So we have a very strong relationship. But before I come back to AI, is it fair to say that Alibaba is a data native company? Because you grew up with data, and that's in your DNA.
Absolutely. It's a data native company. The basis of our whole cloud technology is about handling data, both structured and unstructured. So yes, we're a data native company.
And we talked about that before. It's a message to my team: we have to work on how to contextualize data and make more use out of it. AI revolution: you talked about agents, and it's mind-blowing. It's almost three years since OpenAI hit the world, and there's been a step up in recent months. You had a good example of what an agent is. Can you share that with the audience?
Sure. I'm not a software engineer, but when OpenClaw happened, I asked a colleague to help me understand. He built four agents: one to screen daily news, one to develop a thesis, one to write, and one to challenge and rewrite. Together they became an online KOL for tech blogging. So an agent is a virtual knowledge worker. The world economy is about $110 trillion, labor is 60% of that, and two-thirds of labor is white-collar knowledge workers. So potentially $50 trillion of value could be disrupted or enhanced by agents. That's a massive TAM, bigger than any other industry.
Yeah. Another example from our own operations: an industrial co-pilot uses agents on the shop floor. When there's a problem, you have a product agent, a machine agent, a real-time data agent, and a problem-solving agent that recommends actions. The agent rationalizes, has memory, and acts on your behalf. So the industrial agent better get it right.
Right. One important thing is memory. You have to give the agent access to data, but there are issues if agents get too much data and become uncontrollable. Controlling access is key. Also, with very smart agents, they become the main pilot, not the co-pilot.
Exactly. We talked about a strategic agent and even a CEO agent. But I don't think they will replace your job. The CEO's most scarce resource is bandwidth, time. With agents, you can manage more things and free up time for strategic thinking. So the CEO's job is safe.
So we signed an MOU to enhance our partnership. We are bringing our software portfolio on AliCloud, making it accessible for SMEs with a SaaS model. Also, you have Qwen, your own model. Maybe you can talk about that, which is another area we want to explore.
Three years ago, when ChatGPT came out, we had already started developing our own LLM based on transformer architecture. We decided to be a model company as well as providing cloud services. It requires massive capital and commitment, but we are committed. In the global market, there are leading models in the US, but in China we hope to be one of the leading models. It's a tight race, but we have long-term commitment. We believe the quality of the model will bring more customers to our platform.
Right. Some people say you spend $500 million to stay number one for two weeks, then another comes. But that's how it is. You have a very good one. We also talk about proprietary vs open models. Maybe one thought on that.
Alibaba's Qwen series has a proprietary version with over a trillion parameters, but we also open source smaller versions for edge devices. The open source approach pushes adoption globally. Developers can take our model, deploy on their own infrastructure, fine-tune it for their purposes, all within their own computing environment. This addresses data privacy and security concerns. Over time, the market will have both closed source via API and open source users, but they all need compute resources, which they will procure from us.
Right. What you see behind is an AI-created picture using Qwen. It's really cool. You nailed it when you said most compute is used for training models, but that should reverse as inferencing grows. We are tuning models on industrial specific data like time series, design data, visual inspection. That's where Siemens comes in. We don't build LLMs, but we use models like Qwen to tune them on industrial data. Customers can then make applications for their machines. People ask if this is a bubble because training might saturate. But that's not true because models create data when used, and you can learn from that.
That's the thing. The more people use the model, the more data they generate, which can be used to train the next generation. Training never stops because models iterate. It's not just big model companies; every enterprise will continuously train their models. As they run more inference, they collect more data to further train. So investment in data centers and hardware is justified because the development of intelligence is a continuous endeavor.
Yeah. We are running out of time. Our idea is to boost the industrial AI revolution together. We need partners who know cloud services and build models. We bring hardware, software, domain knowhow, and data. What do you see as the next big step if we join forces? What would be a target for the next couple of months?
You have so many customers in China's manufacturing sector. Alibaba's DNA is in software and consumer-facing, so there are complementary skill sets. China is the best test bed for industrial AI because it's the largest manufacturing economy, with 30% of world industrial output. That's where the data resides. I'm excited to work with Siemens to make industrial AI a real thing.
So we have a great market, innovative and fast-moving, and great partners. Thank you very much. Alibaba. Thank you, Roland. Joe, thank you very much. I appreciate it. Pleasure having you. Thank you.
So now, I can't wait to give you a short teaser about our new products developed in China. As we scale industrial AI, we need resilient data center infrastructure, AI factories. It's about more than GPU chips: power supply, heat management, design software. We have new products designed for AI factories, many developed in China for China. Let's have a look with the experts. Please welcome our technology expert, my colleague, Fan Lele.
Hi, Roland. Nice to meet you.
So I'm pretty sure he picked the right one. So— Yeah, sure. What do you have here?
Yeah, this is our new MCCB, molded case circuit breaker. We call it the 3VD. I want to introduce more. Joe, yeah, yeah. Go, go, go. It's all about HVDC. Because of the surge in generative AI, the power density of data center racks is skyrocketing. So the traditional AC architecture requires a lot of AC/DC conversions.
That's leading to a lot of energy loss. So the solution is HVDC. That can save space by thirty percent, reduce power consumption by forty-five percent, and elevate energy efficiency from the chip to the rack to the grid. So this is the reason why we developed 3VD, our Siemens first DC-specific molded case circuit breaker for data center HVDC application.
It changes also the architecture, so to speak, of the electrical system, but you also have a chance to install it faster.
There's another feature in it, isn't it?
Exactly. You know a lot— You already know a lot. But today, I want to share more with all of you.
It's a compact design with two pole, which means much reasonable for DC application if you compare with three pole and four pole. So then we can get twenty percent smaller in size, you already mentioned a little bit, and thirty percent faster in installation. Then it can support eight hundred volt applications in data center for DC. So one thing I also want to mention is this is relative innovation with more than ten patent applications. And we start from the project initiation to the product launch. We are forty percent faster than before. So the last but not least, in China, for China and beyond.
So by the way, Roland, I would like to share to you is we already have some use cases and pilot customers. Yeah, they like it very much.
So and you said it has two poles only, others have three and four. You don't miss one?
Yeah.
Basically, the two pole already can supply the DC—
You know, system.
So this is the whole package. It's not only the product, but also the great team which developed in record time a new product which goes into AI factories. It changes the way how we electrify them, and superior features—
I wish you, number one, great thanks to you and your team.
And I wish you good luck for the ramp up, not only in China, but also globally.
Thank you very much.
Yes. Thank you a lot.
So we talked about the foundation that powers AI. Big breakthroughs in physical AI are currently happening in robotics, particularly obviously here in China. And I guess that most of you have seen robots dancing and doing kung fu, and I saw them fighting in boxing gloves a couple of weeks ago during my recent visit to China. That's a lot of fun and absolutely impressive. So far, humanoid robots are at their best when they do repetitive moves. But they are not the most spontaneous fellows. They still struggle when it comes to real-time reactions, when they have to contextualize in unpredictable environments like boxing. But the speed of progress is breathtaking. Put one in a boxing ring today, so I would say I would step in, but in six months from now, I better don't.
So now I'm excited to welcome a tech leader whose company builds high-performance humanoid robots, machines that can move and act in the real world. Please welcome the founder and CEO of Unitree Robotics, Wang Xingxing.
Great. Nice having you. Good to have you.
So wait a second. I have to... It seems that somehow you have robots in your DNA— because you developed your first robots in your first year at the university already, and shortly after completing your master, you founded Unitree. I would say this is really fast and passionate about it. So what passion brought you into robotics?
In my freshman year — 2009 — I started building robots. And by a very lucky coincidence, the very first robot I built during my freshman winter break was a small humanoid robot, about this tall. I really, really loved it at the time. But back then, I hadn't thought about building a big company, or about truly bringing robots into everyday life. In the beginning — as a kid, and even as an undergraduate — my main feeling was simply that robots were a really fascinating thing. So but that was a time when AI was not developed to the stage which we have right now.
Did you have that at that time when you started off, did you already have an idea what AI can do?
