The 9 buzziest startups from Y Combinator’s latest Demo Day, according to VCs
The 9 buzziest startups from Y Combinator’s latest Demo Day, according to VCs
根据风投机构的评价，Y Combinator 最新 Demo Day 上最受关注的 9 家初创公司
Another Y Combinator Demo Day took place on Thursday, but this batch looked a little different. While every cohort brings a fresh wave of founders, the startups presenting this week skewed far more toward deep tech than in past years. As we do every quarter, TechCrunch asked early-stage VCs to name the hottest startups in this batch — both their top picks and the deals everyone was talking about. And as usual, we put together a list of startups flagged by at least two investors as the buzziest in the batch. The tech in this batch felt “like science fiction,” as one investor put it. But beyond the wild ideas, the consensus was that valuations were far more grounded than in recent cohorts. We’ve listed the top picks alphabetically below:
周四，Y Combinator 又举办了一场 Demo Day，但这一届看起来有些不同。虽然每一批初创公司都会带来一批新的创始人，但本周展示的项目在深度科技（deep tech）领域的比重远高于往年。按照每季度的惯例，TechCrunch 询问了早期风险投资人，请他们列出本届最热门的初创公司——包括他们的首选项目以及大家都在讨论的交易。像往常一样，我们整理了一份被至少两位投资人标记为“最受关注”的初创公司名单。正如一位投资人所言，这一届的技术感觉“就像科幻小说一样”。但除了这些天马行空的想法外，大家的共识是，估值比最近几届要务实得多。以下是按字母顺序排列的精选名单：
Automarine
What it’s building: Nuclear-powered data centers floating at sea Why it’s a fav: Power is in short supply and local communities are increasingly opposing the buildout of new data centers. Atomarine — co-founded by an MIT computer science and naval engineer and an MIT PhD in nuclear engineering — wants to solve the compute shortage by putting data centers on barges at sea, where seawater could provide near-free cooling. The startup plans to launch a gas-powered pilot by 2028 and plans to transition to floating nuclear power ships in 2032. Atomarine claims it has already secured over $4 billion in customer interest through letters of intent. That revenue potential has helped make it one of the highest-valued startups in the batch, one VC told TechCrunch.
Automarine
业务内容： 海上浮动核动力数据中心 入选理由： 电力供应短缺，且当地社区越来越反对新建数据中心。Atomarine 由一位麻省理工学院计算机科学与海军工程师和一位麻省理工学院核工程博士共同创立，他们希望通过将数据中心放置在海上的驳船上来解决计算短缺问题，海水可以提供近乎免费的冷却效果。该初创公司计划在 2028 年前启动一个燃气动力试点项目，并计划在 2032 年过渡到浮动核动力船。Atomarine 声称已通过意向书锁定了超过 40 亿美元的客户意向。一位风投告诉 TechCrunch，这种营收潜力使其成为本届估值最高的初创公司之一。
Dipole Labs
What it’s building: Effective and energy-efficient high-speed optical networking hardware for AI data centers Why it’s a fav: The idea is that GPU clusters waste a lot of compute time simply waiting for data to move between each chip. Within the networking layer, data is converted from light to electricity and back again in a process that burns lots of power and yields lots of heat. But Dipole Labs says it built an optical switch that skips the conversion so the data can stay as light and go directly where it needs to go. It’s a timely problem to tackle as GPUs are crazy expensive and data centers want the most out of their compute hardware without leaving any time to spare.
Dipole Labs
业务内容： 为 AI 数据中心提供高效、节能的高速光网络硬件 入选理由： 核心逻辑在于，GPU 集群浪费了大量计算时间，仅仅是在等待数据在芯片之间传输。在网络层中，数据需要在光电之间反复转换，这一过程消耗大量电力并产生大量热量。但 Dipole Labs 表示，他们构建了一种光交换机，可以跳过这种转换，使数据保持光信号状态并直接传输到目的地。这是一个亟待解决的及时问题，因为 GPU 极其昂贵，数据中心希望在不浪费任何时间的情况下最大限度地利用其计算硬件。
Isengard Industries
What it’s building: Locally producible, jet-powered strike and counter-drones Why it’s a fav: Isengard aims to mass-produce jet-powered attack and counter-drones directly within allied countries, at a fraction of what prime contractors charge to build them in the U.S. Co-founded by a former Australian Army officer and a defense entrepreneur who previously scaled another Ukraine-focused drone startup to $60 million in revenue, Isengard is already generating $10 million in revenue itself. The startup has generated strong VC buzz, garnering one of the loftiest valuations in this YC batch, according to two investors.
Isengard Industries
业务内容： 可在当地生产的喷气式攻击和反无人机系统 入选理由： Isengard 旨在直接在盟国境内大规模生产喷气式攻击和反无人机，其成本仅为美国主要承包商制造价格的一小部分。该公司由一位前澳大利亚陆军军官和一位国防企业家共同创立，后者曾将另一家专注于乌克兰的无人机初创公司规模扩大至 6000 万美元营收。Isengard 目前已实现 1000 万美元的营收。据两位投资人称，该初创公司在风投圈引起了强烈反响，获得了本届 YC 中最高的估值之一。
Lamb Labs
What it’s building: Custom inference chips with hardcoded AI model weights Why it’s a fav: Traditional AI chips burn massive amounts of energy during inference, fetching model weights from memory. Co-founded by an Imperial College London AI Ph.D. and an Oxford theoretical physicist, Lamb Labs wants to build ultra-efficient chips by hardcoding AI model weights directly into silicon. Dubbed “Model Processing Units” (MPUs), these custom chips eliminate memory-bandwidth bottlenecks.
Lamb Labs
业务内容： 带有硬编码 AI 模型权重的定制推理芯片 入选理由： 传统的 AI 芯片在推理过程中会消耗大量能量，用于从内存中获取模型权重。Lamb Labs 由一位伦敦帝国理工学院的 AI 博士和一位牛津大学理论物理学家共同创立，他们希望通过将 AI 模型权重直接硬编码到硅片中来构建超高效芯片。这些被称为“模型处理单元”（MPU）的定制芯片消除了内存带宽瓶颈。
Praxis AI
What it’s building: Collecting real-world data on which to train robots Why it’s a fav: This company partners with businesses to collect videos and data of humans doing work, which it then turns into training material for companies building robots. It says it’s already working with publicly traded companies and has captured video data in more than 150 different environments. This company could become important as businesses start experimenting more with which tasks are best for humans — and, as AI continues to refine itself, which tasks are better left to a robot.
Praxis AI
业务内容： 收集用于训练机器人的真实世界数据 入选理由： 该公司与企业合作，收集人类工作时的视频和数据，并将其转化为机器人制造公司的训练材料。据称，他们已经与多家上市公司合作，并在 150 多种不同的环境中捕获了视频数据。随着企业开始更多地尝试哪些任务最适合人类，以及随着 AI 的不断完善，哪些任务更适合交给机器人，这家公司可能会变得非常重要。
Nori
What it’s building: Creating affordable robots to handle everyday tasks Why it’s a fav: Launched just six weeks ago, this company already touts almost half a million in sales. That’s not shocking — it’s a humanoid robot that is promising to help clean and fold clothes. Users can also operate their robot via a laptop app. It’s priced at around $1,600, which is a steal compared to other humanoids like Neo, priced around $20,000. One of the biggest questions in robotics is if it is possible to create an affordable at-home robot that can actually stack a dishwasher and do everyday home tasks. Nori is another swing at taking on the task.
Nori
业务内容： 制造可处理日常任务的平价机器人 入选理由： 该公司仅在六周前成立，销售额已接近 50 万美元。这并不令人惊讶——它是一款人形机器人，承诺可以帮助清洁和折叠衣物。用户还可以通过笔记本电脑应用程序操作机器人。其售价约为 1600 美元，与售价约 2 万美元的 Neo 等其他人形机器人相比，简直是“白菜价”。机器人领域最大的问题之一是，是否有可能制造出一款负担得起的家用机器人，能够真正完成洗碗机装载和日常家务。Nori 是在这一挑战上的又一次尝试。
Cosmic Robotics
What it’s building: Autonomous robots that can lift heavy objects Why it’s a fav: The founders of this company want to build a city on Mars. The first step to that? Building robotic technology that performs heavy-duty tasks. It says its tech is already installing solar panels across the U.S. and that it has $25 million in contracts through 2027. The vision is that this technology can help automate construction for colonizing Mars. The startup is racing with SpaceX’s timeline for building on Mars, as it hopes to begin an exploratory mission by 2028.
Cosmic Robotics
业务内容： 可搬运重物的自主机器人 入选理由： 该公司的创始人希望在火星上建立一座城市。第一步是什么？构建能够执行重型任务的机器人技术。该公司表示，其技术已在美国各地安装太阳能电池板，并拥有到 2027 年为止的 2500 万美元合同。其愿景是，这项技术可以帮助实现火星殖民的建筑自动化。该初创公司正在与 SpaceX 的火星建设时间表赛跑，希望在 2028 年前开始执行探索任务。
Parasma
What it’s building: Training human brain cells to one day power compute Why it’s a fav: Another company trying to find the best way to tackle how much power and energy is needed to run models. This time, Parasma is looking at using human brain cells as an effective, more energy-efficient alternative to today’s AI computing hardware.
Parasma
业务内容： 培养人类脑细胞，以期未来为计算提供动力 入选理由： 这是另一家试图寻找最佳方案来解决运行模型所需巨大电力和能源问题的公司。这一次，Parasma 正在研究利用人类脑细胞作为一种有效且比当今 AI 计算硬件更节能的替代方案。
Waddle Labs
What it’s building: An API layer that writes robot control code Why it’s a fav: Investors and tech enthusiasts are hoping that the ChatGPT moment for robots is around the corner. That excitement is driving different approaches to creating a general robotics model. Instead of training foundation models on raw video or human teleoperation data, Waddle Labs uses a layer of LLM agents to write code and control robots directly. The startup, founded by Harvard graduates, is positioning itself as…
Waddle Labs
业务内容： 一个编写机器人控制代码的 API 层 入选理由： 投资人和科技爱好者们都希望机器人的“ChatGPT 时刻”即将来临。这种兴奋感推动了创建通用机器人模型的不同方法。Waddle Labs 没有在原始视频或人类远程操作数据上训练基础模型，而是使用一层 LLM（大语言模型）代理来编写代码并直接控制机器人。这家由哈佛大学毕业生创立的初创公司，将自己定位为……