侯晓迪以Bot Auto重返自动驾驶 削减成本迎战融资格局剧变

侯晓迪以Bot Auto重返自动驾驶 削减成本迎战融资格局剧变 硬科技创业者必须放弃估值幻想并像经营传统餐厅一样管理现金流，通过直接交付最终商业服务来渡过技术炒作周期过后的资本寒冬。 发言人通过对比真正的风险投资与随波逐流的共识资本，剖析了初创生态的运作机制，指出后者的注意力通常很难维持五年以上。硬科技创始人面临着严重的时间错位，因为深层核心技术与商业化产品的研发周期往往超过典型的炒作周期，这使他们跌入残酷的“后炒作”现实，直面投资者对绝对落地完备性而非潜力的严苛要求。为了度过这一阶段，创始人必须抛弃估值泡沫，像经营传统连锁餐厅那样思考，严格聚焦于运营盈利并直接交付完整服务。通过将这一生存法则应用于新创企业Bot Auto，发言人借助去除安全员并大幅压缩运营成本，目前已成功实现微盈利。 硅谷资本正由风险资本转向追逐确定性的“共识资本”；两年前市场共识在LLM，如今转向机器人与脑机接口，自动驾驶已从风口跌落。 (0:00) Gartner炒作周期显示共识资本的注意力通常仅有1到5年，而硬科技需要在短暂周期内同时完成底层研发与产品化，极易遭遇周期错配。 (2:15) 处于热潮期的Agentic AI即便只有50%成功率也被追捧，而进入“后热潮”阶段的自动驾驶即使以4100万美元高效率实现无安全员L4卡车落地，仍面临投资人的冷眼与苛求。 (4:59) 后热潮时代硬科技的生存之道在于回归常态商业逻辑，像经营连锁餐馆一样严控运营成本与获客成本，通过端到端提供服务而非售卖技术配方实现造血盈利。 (9:59)

Hou Xiaodi contends that hard-tech ventures inevitably outlast short-lived market hype cycles, requiring founders to transition from speculative, consensus-driven startups into disciplined enterprises modeled after the economics of conventional businesses. Consensus Capital and Diverging Founder Models (0:00) Drawing on a decade in the startup ecosystem, Hou distinguishes between contrarian venture capital and "consensus capital," which clusters around established industry trends to secure low-risk, structured exits from Series A to Series C rounds. Driven by fixed fund lifespans and the institutional need for certainty, consensus capital shifts rapidly across prevailing narratives—moving from large language models two years ago to robotics and brain-computer interfaces today, while leaving autonomous driving out of favor. This capital environment divides entrepreneurs into two operating models. "Startup" founders run time-driven entities focused on selling before a consensus cycle expires, a path Hou observes frequently compromises original technical visions in favor of quarterly board metrics. Conversely, "enterprise" founders operate as results-driven builders, committing to endure across cycles to see foundational technology through to completion and establish lasting institutions. The Hard-Tech Timeline Mismatch and Post-Hype Skepticism (4:59) Hard-tech companies face a severe structural friction within the Gartner Hype Cycle: developing core scientific technology while simultaneously engineering commercial products requires extended timelines, yet consensus attention spans typically last only one to five years. As a result, investor enthusiasm routinely evaporates before deep technologies mature, stranding founders in what Hou terms the "after-hype" environment. Investor scrutiny shifts dramatically between these phases. While active hype cycles tolerate low technical reliability—such as investors funding agentic AI displaying 50% success rates on the speculative assumption that performance will rise to 70% and 90%—after-hype sectors like autonomous driving face extreme skepticism. Hou notes that in this post-hype stage, investors refuse leaps of faith and disregard capital efficiency milestones; for example, demonstrating driver-out Level 4 autonomous trucking on a $41 million budget was dismissed by investors demanding fully completed, de-risked products before committing funds. Operational Discipline and Direct Service Delivery (9:59) To survive after hype recedes, Hou argues founders must abandon Silicon Valley exceptionalism and adopt the operational rigor of conventional chain restaurants, which represent 90% of broader economic activity. Rather than relying on proprietary technology recipes, valuation expansion, or buzz, survival depends on a straightforward unit formula: product value minus operating costs and customer acquisition costs. Hou illustrates this requirement by pointing to his earlier venture, which spent $20 per mile to operate autonomous trucks while earning only $2 per mile—a disparity tolerated only during market exuberance. To reach sustainability, companies must achieve operational profitability; his current venture, Bot Auto, focuses on driver removal and cost control to run a modestly profitable operation. Furthermore, Hou asserts that hard-tech founders cannot wait for customer adoption, but must absorb customer acquisition by directly operating the end service and delivering finished commercial value rather than unintegrated technology components. Hou concludes that long-term deep-tech survival depends on operational self-sufficiency rather than reliance on transient investment waves. However, his framework leaves unaddressed how early-stage hard-tech ventures can fund the capital-intensive research phase required to reach driverless operation before achieving operational cash flow.

自动驾驶创业者侯晓迪指出，硬科技企业面临核心技术与产品化开发周期漫长、与共识资本仅能维持1至5年的短期炒作周期严重错配的生存危机；在热潮退去的现实中，企业必须抛弃对资本信仰的依赖，转向传统连锁餐厅式的商业逻辑，通过压缩运营与获客成本、直接交付最终服务来实现商业化自造血。 共识资本机制与技术热潮的周期错配 (0:00) 侯晓迪将科技投资区分为两类：真正的“风险资本”押注于极少数人相信并能颠覆世界的非共识设想，而广泛存在的“共识资本”则追逐已确立的市场趋势以追求确定性。共识资本促使投资者倾向于在A轮进入、C轮退出套利，即便投资失败也无须承担职业声誉风险。例如两年前共识资本集体追逐大语言模型，如今转向机器人与脑机接口（BCI），而自动驾驶则明显失宠。 在共识资本主导的环境下，追求套现退出的“初创型”创始人往往受时间驱动，核心目标是在共识周期结束前将公司出售，但这常导致原有长远愿景被收购方以季度和年度财报指标所取代；而旨在打造长青机构的“企业型”创始人则必须面对高德纳技术成熟度曲线（Gartner Hype Cycle）揭示的现实——技术走向成熟往往历时漫长，但资本共识的热度通常仅能维持1至5年。 硬科技研发双重困境与后热潮冷遇 (4:59) 学术背景出身的硬科技创业者面临底层技术突破与商业化产品构建的双重负重。侯晓迪表示，算法本身相对廉价，真正产生现金流的是产品；然而，要在短短几年的单一共识周期内同时挤入深科技研发与商业产品化极为艰难，热潮往往在技术完全成熟前便已消退。 当行业跌入“后热潮”（after-hype）阶段，资本的评估标准会发生剧烈转变。在目前处于风口的人工智能体（Agentic AI）领域，50%的成功率仍会被视作重大里程碑，投资者乐观预期其随后将提升至70%乃至90%；但在步入后热潮的自动驾驶领域，投资者拒绝任何远期信念飞跃，要求“完全做出来之后再谈”。侯晓迪指出，其团队以4100万美元资金便实现无安全员L4级自动驾驶卡车的成果，在后热潮语境下仍被投资者以见过“资金效率更高公司”为由冷遇，类似跑出9.80秒的百米成绩却被质疑未打破世界纪录。 连锁餐饮逻辑：运营盈利与全栈交付 (9:59) 侯晓迪反思指出，硅谷创业者往往脱离了占据全球经济活动90%的正常商业世界。在经历了前两次创业的顺风与近期两年的后热潮磨砺后，他提出硬科技企业必须采用低姿态的“连锁餐厅思维”：企业价值等于产品价值减去运营成本与获客成本（CAC），而非依靠独家秘方、估值膨胀或社交媒体炒作。 在狂热周期中，侯晓迪的前公司曾出现每英里运营成本达20美元、而营收仅有2美元的严重倒挂，这种亏损模式仅在热潮期被容忍。而在后热潮时代，企业必须依靠实际运营盈利。其创办的Bot Auto通过拿掉安全员并严格控制运营成本，已实现微薄盈利。侯晓迪强调，面对高昂的获客成本，硬科技公司不能只提供技术“配方”或被动等待客户采用，而必须亲自下场运营，将最终成型的端到端服务直接交付给客户。 侯晓迪总结认为，硬科技公司穿越周期的关键，在于彻底摒弃热潮时期的虚高估值幻觉，转向以运营毛利为核心的传统商业常识，并通过全栈式交付最终服务打破落地瓶颈；但这种模式要求硬科技团队兼具前沿技术攻坚与重资产精细化运营的双重能力，如何在追求微利的严苛运营中持续保持技术领先，仍是后热潮时代企业的长期挑战。

侯晓迪: 风口站对了，猪都能飞起来 [0:00] 侯晓迪: venture capital and consensus capital [2:15] 侯晓迪: to the very end of it [2:15] 侯晓迪: plateau of productivity [4:59] 侯晓迪: after-hype [4:59] 侯晓迪: we wanted to talk to you after you finish it [4:59] 侯晓迪: more capital efficient companies [9:59] 侯晓迪: Restaurant Value = Food Value - Operating Cost - Advertising Cost [9:59] 侯晓迪: tiny bit of money [9:59] 侯晓迪: We are here providing the food. We're not here providing the recipe

[0:00] Hello, everyone. Glad to be here this year. Oh, okay. Okay. Wanted to talk about, not about autonomous driving today, wanted to talk about entrepreneurship in general. You know, every time when I visit the Bay Area, I feel this is a promised land. A lot of miracles happen in here, and more drama, of course. So there will be two types of capitals, in fact, after 10 years in doing all kinds of startups, I found that it's actually not only about VC. There are two types of capitals. One is called venture capital, and the other one is called consensus capital. And venture capital bet on ideas. Ideas where it's big enough to change the world. Ideas where other people don't believe in. So that's why you call them venture capital. But many other capitals are here to bet for consensus. So consensus capital is the type of capital where, okay, two years ago everyone thinking about large language model that the thing to bet So you bet on it Even if your bet is wrong there no problem because everyone is feeling the same And more often you can get in an A round and get out in C round, you can make money. So that's the consensus capital. And people may feel that, well, everyone's talking about VC, VC, and VC. So VC might be the majority of this capitalism. But in fact, what I feel is that the consensus capital is actually more typical. That's a matter of fact. I'm not really trying to say a consensus capital is evil or anything. It's actually they have their reasons. Because every fund, they have a certain limit of years to return to their investors so that they have to bet on something that has a certain amount of certainty. And also, the biggest fund like Andreessen Horowitz actually is making great money for their investors. But after all, they are betting money on the consensus. And look at the consensus today. we have a robotics and we have a BCI, right? And we used to have autonomous driving, but not anymore. But those are the consensus we talk about. So every now and then in the Silicon Valley, I hear people, even in the restaurant, I hear people talking about some buzzwords. And those are the consensus And those what attracts people getting in attracts investment getting in and making miracles happen But the choice for the founders is not always true that we get into something that is a consensus because, hey, if I'm getting into that consensus and my peers are also getting into the same consensus, everyone is competing on the same topic, that happened too often in this promised land in here in the Silicon Valley. And I would say there are two types of founders. One, we call them the founders who does startups. The founder who does startup is a founder that has a certain time. Before the consensus is running out, they got to make their way out. That is a typical startup. So I have several friends who are just repeatedly starting small businesses, getting acquired by big companies, and restart again. So that's a lifestyle. So I call them startups. The other one is enterprise. Because if you really sell your company to a big company, you sell your startup to a big company, there is no way for you to continue the journey. Someone else is taking over, and you have a board that's managing you, and they are not driven by the vision. They driven by the next quarterly report or annual report of the financials So the other one if you trying to do something you wanted to go to the very end of it I call them enterprises So that results So there is a very different mentality between the time-driven and results-driven play. And of course, all of that has been influenced by the consensus. And we all know this. This is a very typical Gartner cycle of hype. Every year, there is a new thing. And I try to specifically choose for this meeting, This is the AI-related Gartner cycle. Not my opinion, but Gartner Institute's opinion about it. People talk about this. I think this is definitely not the first time you saw it. But what I see from this cycle is that there are two things. One, in the general trend, things take a lot of years to mature to the right end, to finally plateau of productivity on the right bottom. But there is a hype cycle. And in fact, hyper-cycle, in other words, is actually the attention span of the consensus. A lot of people focusing in one thing. Of course, this is a phenomenon. But this phenomenon, this party, only lasts for a certain amount of time. Normally not more than five years. Sometimes even one year. Like, how many of you started thinking about starting building a large language model company? [4:59] As of today, not a good business, right? Maybe tomorrow OpenAI will have another feature that kills you. I talk about this to investors back in 2023, but people don't listen. But, you know, this year maybe we can talk about robotics. But, you know, two years in the future, some people would tell you, oh, of course, that's a very natural issue for robotics. Robotics is not a good business in 2025, and the reason is A, B, and C. Two years later, people will tell you that. That becomes a new consensus, of course. So consensus has a very relatively short time span. But in order to achieve something, achieving an enterprise takes much longer time. So this is what I call a mismatch, especially if we're thinking about building hard techs. A lot of people like me, we're coming from academia, and we get into the startup world, or enterprise world, whatever. We build things. Building hard tech itself is already very, very torturing, and it takes a long time. But even worse, we also have to build the product around it. Because building technology, you have great technology, but it doesn't really make money out of it. You need to find the customer. You need to polish your product instead of the core technology Those algorithms are relatively cheap The product is what makes money But building product plus building tech how can we squeeze all of that within one consensus cycle And very often, the consensus cycle is gone before we even finish building the hard tech. That's a sad reality, and that's why many of the companies are suffering. Think about this grid technologies. In my opinion, most of the grid technology, they must go through the cycles. They can't really just stay in one cycle and build everything. If you're building that within one cycle, I'm telling you, you're building a small business. You're not building an enterprise. All of the things, virtual reality. How many of you are talking about virtual reality today? Back in 2016, I have one of my best friends in the Silicon Valley decided to go home, go back to China, and build his own VR company. He's still fighting, but his life is not as great as some of the other companies because it's very hard to raise money for the sake of virtual reality. Similar things with autonomous driving. I'm still working on autonomous driving. I'm so jealous of myself. You know, seven years ago, it very easily become a unicorn and easily go to public, but, you know the chance is no longer there today Genomics robotics quantum computing even if we take a very generic sense artificial intelligence Artificial intelligence has been through multiple hapes Back in 1960s, Dartmouth, right? People are talking about computer vision to be solved in one summer camp. That's not happening. And decades of ups and downs. And Jeff Hinton used to be the celebrity and used to be the heterodox research scientist and then become the celebrity again. And this is the third time he becomes celebrity. But he has this insistence. Only with those type of insistence, great technology will be born. And what happened after the hype? I call them after hype. So I've been living in the after hype world for about two years. Even though I'm still fighting, I still feel the pain. First of all, in the after-hive world, no one takes the pain. You can make a promise saying that, you know, the robotics can do everything, their generalization power is great enough, or you can say, Agent TKI can actually solve all of the problems for you. All of these things are promised, because you have not yet provided hard evidence in doing that. By the way I just learned that for Agent TKI one of the metric interestingly is how How many hours can you run the agent without failing And in fact the metric has a caveat If you have 50 of the success rate we count it in as a duration of the running. Like 50%, come on. That's not a product at all. But people are willing to bet in, because people can always mesmerize themselves and saying, yeah, today, the technology is not perfect, but it's 50%. Maybe next year it gets to 70%, and two years later, 90%. Who knows? People have this kind of optimism about technology. But that thing never happened in the afterworld. Afterworld is that I've been talking to people, hey, we're building the autonomous driving trucks, driving without a driver. I've done it before. So trust me, we can do it. Nah, we wanted to talk to you after you finish it. Of course, we'll finish it, and there will be more challenges and questions. That's one. The second is that in the after-hype world, No one cares about the relative progress. I'm always joking that I've been presenting to other people that we have delivered the driver out, level four driver out the truck without a human with only 41 million of spending. That's a wild number. Even to myself, I feel very proud of the team. But investors say, well, you know, I've seen more capital efficient companies out there. [9:59] This is like a very, I would say, humiliating to me, as if I were telling people, I've run a 100-meter sprint with 9.80 seconds. And some people say, oh, reminds me about the world record. So this is where I felt that no one really cares about the relative progress. And the worst is that for us, the founders, we're not prepared for all of it. I wanted to do a tally here. How many of you have been working in the startup world for more than five years? Raise your hand, please. I don't see many. And to me, I've been working in this world for about 10 years, a little bit more than 10 years. But I have did my account. My first startup was at the Borne of Deep Learning, where I certainly am the beneficiary of the hype. And my second startup was in autonomous driving. I lived probably six years in the hype, which I enjoyed the hype. And then the third startup, a little bit more than two years, I live in an afterhype world. So I have either lived in the hype world or the afterhype world. And I think this is also a very common symptom for everyone here in the promised land of the Silicon Valley where we always live in this hype or after hype world And we including myself has ignored a bigger world normal business world And in fact, the answer isn't this traditional investment activity. The normal business world is the answer to go through the hype cycles. This is what I believe. I'm talking to my team. What if people don't believe in atomic driving anymore? if people feel that there is an after-hype repercussion. We can still make money, we can still find investment, if we're thinking about ourselves in a very humble way, thinking that we're running a chain restaurant business. Of course, 90% of all the economics are happening in just like a normal world. Not hype, not after hype. OK, no secret recipe. We're not KFC, we're just like a typical restaurant. And no one needs to take a leap of faith with us. Like, our food is not superior. And there's no valuation inflation. And of course, limited social media attention. But even with all of that, can we make money? Here, this is a formula. If you want to take photos, please take this one. This is probably the most important thing, your survival formula for the after world Restaurant value equals food value minus operating cost and minus advertising cost if you can solve that value you can say okay my restaurant makes money now I need more money to build more restaurants to make a chain that is very convincing to normal investors we don't need to find those consensus investors we don't need consensus or or venture consensus venture capital but let's transform that restaurant formula for autonomy product value People are talking about autonomous driving. Oh, beautiful. Remove the driver, right? But people forgot about operating cost and customer acquisition cost, especially if you are selling the technology. A lot of us today, we're simplifying our roadmap of technology growth, or company growth, or enterprise growth by thinking about, okay, let's talk about thinking about selling technology to millions of the customers, assuming that we have the customer here tomorrow. and they don't really think about getting into there. But I feel more and more true that many of these hard technologies requires the founder of enterprise, not a startup they're planning to sell to big companies next year a founder of the enterprise to really focusing on making money as if we building the chain restaurant especially on the operating cost For autonomous driving even cleaning the sensors is not a trivial job But many people have overlooked that. Give you one data point. In my last company, we spend $20 per mile in operating our fleet. And we're making $2 per mile. Yes, the technology is cool. But people will only pay for coolness in the hype. We've got to go back to the normal business. And I'm not sure if today, I don't want to speculate about Waymo's cost, but my understanding is that the total cost combined of the operation is still not making money. And I'm glad my new company bought auto. We are already making a tiny bit of money by removing the driver here in the operating cost. And the last thing is customer acquisition cost. This is especially true for robotics. Like, I don't really feel that my parents would allow a robot to be invited, to be introduced to their home and they're living with a robot? No way. They're not prepared. So how can we get customers if we're selling the robots to others? Similarly, how can we sell the autonomous driving technology to a truck manufacturer, car manufacturer, or a fleet of trucks? We have to be the owner who operates the trucks. Again, this is like a chain [14:59] We are here providing the food. We're not here providing the recipe. We're not here providing the half-baked food, right? That's a half-processed food. That's an issue in China right now, right? But thinking about this, if you are a restaurant, you fully believe you are a chain restaurant, you got to provide the food for the customer because that's the easiest way for you to make money. You can solve the customer acquisition cost and solving operation cost. OK. I'll leave several minutes for the quick Q&A. And for those who want to build hard tech, live long and prosper.