Y Combinator startups always shake up the tech market, but the latest Demo Day, held on Thursday, was unlike anything we have seen in recent years.
This edition brought a clear shift toward deep tech — that category of heavyweight technology that demands cutting-edge science and serious engineering to get off the ground. We are not talking about another pretty app or another SaaS platform with a slick dashboard. We are talking about projects that challenge the laws of physics, that require labs, scientists, and years of research before a single line of code is ever written. That is the level of what showed up in this batch.
Household robots, nuclear-powered data centers floating on the ocean, human brain cells as computing hardware… yeah, it sounds like a sci-fi script. And that is exactly what investors said when asked what caught their attention this time around. One of the VCs interviewed by TechCrunch summed it up nicely: the technology on display felt like it came straight out of a movie. But it was not fiction. It was real, right there in front of them, with working prototypes and founding teams who knew exactly what they were talking about.
But there was another surprise, too: unlike previous editions, valuations — meaning how much each startup was being priced at — were more realistic and grounded. Less hype, more substance. The market seems to have matured after a few years of bubble territory, and the founders who showed up this edition got the memo.
As it does every quarter, TechCrunch talked to early-stage investors to put together the list of the most buzzed-about startups from the batch — the ones that popped up on at least two different VCs’ radars. Here are the picks, listed in alphabetical order. 👇
Atomarine: nuclear data centers floating on the ocean
The idea behind Atomarine is bold: put data centers on barges in the middle of the ocean, where seawater can provide cooling practically for free. The company was co-founded by a naval engineer and computer scientist with an MIT degree alongside a PhD in nuclear engineering, also from MIT.
The context helps explain the enthusiasm. Energy is in short supply and local communities are increasingly pushing back against building new data centers on land. Atomarine wants to solve that shortage of computing capacity by taking everything offshore. The plan is to launch a gas-powered pilot by 2028 and transition to ships with floating nuclear energy by 2032. The startup says it has already secured more than 4 billion dollars in customer interest through letters of intent — a number that helped place it among the highest-valued companies in the batch, according to one of the VCs.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Cosmic Robotics: robots to colonize Mars
The founders of Cosmic Robotics have a massive dream: build a city on Mars. And the first step toward that is developing robots capable of autonomously performing heavy-duty tasks.
The company says its technology is already being used to install solar panels across the United States and that it has 25 million dollars in signed contracts through 2027. The vision is that this same technology could automate the construction needed to colonize the red planet. The startup is racing against SpaceX’s timeline for building on Mars and hopes to kick off an exploratory mission by 2028.
Dipole Labs: lasers to speed up AI data centers
Dipole Labs tackles an expensive bottleneck inside artificial intelligence data centers. GPU clusters waste a ton of processing time just waiting for data to move from one chip to another. At the networking layer, information gets converted from light to electricity and then back to light again — a process that eats up a lot of energy and generates a lot of heat.
The company’s solution is an optical switch that skips that conversion entirely, allowing data to stay in the form of light and travel straight to where it needs to go. It is an urgent problem, since GPUs are absurdly expensive and data centers want to squeeze every last second of capacity out of their hardware.
Isengard Industries: jet-powered drones produced locally
Isengard Industries wants to mass-produce jet-powered attack and defense drones directly in allied countries, at a fraction of what major defense contractors charge to build them in the United States.
The company was co-founded by a former Australian Army officer and a defense entrepreneur who previously scaled another drone startup focused on Ukraine to 60 million dollars in revenue. Isengard itself already generates 10 million dollars in revenue and landed one of the highest valuations in the batch, according to two investors who spoke with TechCrunch.
Lamb Labs: chips with AI weights etched into silicon
Traditional artificial intelligence chips burn enormous amounts of energy during inference, constantly fetching model weights from memory. Lamb Labs wants to fix that in a radical way: by etching model weights directly into the silicon.
Co-founded by a PhD in AI from Imperial College London and a theoretical physicist from Oxford, the company calls its chips Model Processing Units, or MPUs. The idea is to eliminate memory bandwidth bottlenecks and create ultra-efficient processors custom-built to run specific models.
Nori: affordable robots for everyday tasks
Launched just six weeks ago, Nori has already racked up nearly half a million dollars in sales. And the reason is no surprise: it is a humanoid robot that promises to help clean the house and fold laundry. Users can also control the robot through a laptop app.
The price is the real game-changer. Nori costs around 1,600 dollars, a bargain compared to other humanoids like the Neo, which is priced at roughly 20,000 dollars. One of the biggest questions in robotics right now is whether it is possible to build an affordable home robot that can actually load the dishwasher and handle common household chores. Nori is the latest attempt to answer that question.
Parasma: human brain cells for computing
Parasma is another company trying to solve the energy consumption problem that comes with running artificial intelligence models. Its approach, though, is quite unconventional: using human brain cells as a more energy-efficient alternative to current computing hardware.
It is one of the most experimental projects in the batch and the kind that makes anyone raise an eyebrow. But with the demand for energy efficiency growing nonstop, biological bets like this one are starting to attract the attention of investors willing to look way beyond the conventional.
Praxis AI: real-world data to train robots
Praxis AI positions itself in a fundamental part of the robotics pipeline: data. The company partners with businesses to collect videos and information of humans performing tasks, turning that material into training datasets for companies building robots.
According to the startup, it already works with publicly traded companies and has captured video data across more than 150 different environments. This company could grow in importance as businesses start experimenting with which tasks are better suited for humans and, as AI gets more refined, which tasks are better left to robots.
Waddle Labs: the Claude Code for robotics
Investors and tech enthusiasts are rooting for robotics’ ChatGPT moment to be right around the corner. That enthusiasm is fueling different approaches to building a general-purpose robotics model, and Waddle Labs has picked its own path.
Instead of training foundation models on raw video or human teleoperation data, Waddle uses an LLM-based agent layer to write code and control robots directly. Founded by Harvard alumni, the startup positions itself as the Claude Code for robotics. The promise is that by connecting any hardware to Waddle’s API, developers can simply tell the robot what to do in plain language. The AI agents then generate executable control code, verify that it works, and configure the robot in about 20 minutes.
What this batch says about the future of startups
For several years running, Y Combinator was dominated by software startups. It made sense: the cost of entry was low, scalability was massive, and the returns could be wild. But the landscape has changed. Artificial intelligence became increasingly accessible, which leveled the playing field in a big way. Suddenly, building an AI product became something that almost any team with access to the right APIs could pull off, and the market got flooded with similar solutions.
What this latest batch showed is that the pendulum has swung back the other way. Investors are looking for real barriers to entry, and deep tech delivers exactly that. When you are developing a modular nuclear reactor, building a chip with weights etched into silicon, or creating a robot that can actually handle household chores, you are in territory where very few can follow. The technical knowledge required is steep, the development timeline is long, and the capital involved is significant. But when it works, the competitive advantage is nearly impossible to overcome.
Another factor that stood out was the quality of the founding teams. Many of the startups presented were created by former researchers from top-tier universities like MIT, Oxford, Imperial College, and Harvard, as well as ex-engineers and former military officers. These were not first-time entrepreneurs taking a shot in the dark — they were specialists who identified a specific problem and decided to build something from scratch. That combination of deep technical expertise with an entrepreneurial mindset was one of the things that impressed VCs the most.
Zooming out from this batch, it is clear we are looking at a cycle shift in the startup ecosystem. The golden era of pure software is far from over, but the frontier of what is possible to build is expanding in a way we have not seen in a long time. Artificial intelligence is serving as an enabling layer that makes it feasible to develop increasingly sophisticated physical products.
When the most influential accelerator in the world starts betting on robots, floating nuclear energy, and biology-based chips, that is not a coincidence — it is a careful read on where the real opportunities are taking shape. For anyone following the industry closely, the message is clear: the next generation of major tech companies might not be built around an app or a digital platform. It might be built around a robot that folds laundry, a data center that floats on the ocean, or a chip that mimics the human brain. It sounds like science fiction, but that is exactly what showed up at the latest Y Combinator Demo Day — and investors are taking it very seriously. 🚀
