Y Combinator's Latest Batch: A Glimpse Into the Future of Deep Tech

Deep Tech Takes Center Stage

When I first heard about this latest Y Combinator Demo Day, I was struck by something unusual: the startups weren't just talking about mobile apps or SaaS platforms. Instead, they were presenting ideas that could reshape entire industries — and in many cases, humanity itself.

The shift toward deep tech isn't just a trend — it's a fundamental reimagining of how we approach innovation. As one investor put it, the tech in this batch felt "like science fiction." But here's what separates these startups from pure fantasy: they have real solutions to pressing problems, whether that's energy scarcity, computing inefficiency, or robotics challenges.

Atomarine: Floating Nuclear Data Centers

One of the most intriguing concepts was Atomarine, a company building nuclear-powered data centers that float at sea. This isn't just about convenience — it's about solving a real problem in our computing infrastructure.

As demand for cloud services continues to grow, so does the strain on traditional data center facilities. Power shortages and local opposition to new construction projects have created a bottleneck that Atomarine aims to solve by bringing computing power directly to where it's needed most — on the ocean. The startup plans to launch its first gas-powered pilot by 2028, transitioning to nuclear-powered ships in 2032.

"The idea is that we're not just moving computation offshore, but solving a systemic issue with compute supply," said one investor who followed the company closely. "This could be a game-changer for data center efficiency and sustainability."

The fact that Atomarine has already secured over $4 billion in customer interest through letters of intent speaks volumes about the real-world appeal of this approach. It's not just theoretical — it's commercial, scalable, and addressing a genuine need in the industry.

Dipole Labs: Revolutionizing AI Networking

Next up is Dipole Labs, a company focused on creating effective and energy-efficient high-speed optical networking hardware for AI data centers.

This startup addresses one of the most critical bottlenecks in modern AI infrastructure: how quickly data moves between chips. Traditional systems convert data from light to electricity and back again, which is incredibly energy-intensive and generates significant heat. Dipole Labs' solution skips this conversion process entirely, allowing data to travel as light — directly where it needs to go.

As GPUs become more powerful but also more expensive, minimizing idle compute time becomes crucial. By optimizing the networking layer, Dipole Labs ensures that every second of processing power counts, making AI infrastructure not just faster, but also far more efficient.

Isengard Industries: Counter-Drones for the Modern Battlefield

Isengard Industries introduces a fascinating twist in defense technology by developing locally producible, jet-powered strike and counter-drones. 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 company's approach addresses the growing need for affordable, scalable drone technology that can be manufactured closer to where it's needed — reducing reliance on expensive, centralized production chains. This could have profound implications for national security and humanitarian efforts worldwide.

Lamb Labs: AI Chips Built for Efficiency

Lamb Labs tackles another critical challenge in AI computing by building custom inference chips with hardcoded AI model weights. Traditional AI chips waste enormous amounts of energy during inference, fetching model weights from memory — a process that's both slow and power-hungry.

Co-founded by an Imperial College London AI Ph.D. and an Oxford theoretical physicist, Lamb Labs has developed what they call "Model Processing Units" (MPUs). These chips eliminate memory-bandwidth bottlenecks by hardcoding AI model weights directly into silicon, dramatically improving efficiency and performance.

Praxis AI: Teaching Robots to See the World

Praxis AI focuses on collecting real-world data for robot training. The company partners with businesses to gather videos and data of humans performing various tasks, which it then turns into valuable training material for companies building robots.

This approach is particularly significant as businesses begin to experiment with which tasks are best suited for humans versus machines. As AI systems continue to evolve, having quality training data that reflects real-world scenarios becomes increasingly vital — and Praxis AI is positioning itself to be at the forefront of this effort.

Nori: Affordable Robotics for Everyday Life

Nori has made a bold move into consumer robotics with its humanoid robot priced at just $1,600 — a fraction of the cost of similar products like Neo, which retails for around $20,000. The company has already sold nearly half a million units in just six weeks.

The question many are asking is whether we can create affordable, practical robots that can handle everyday home tasks like cleaning and folding clothes. Nori's approach suggests there may be a path forward — one that makes robotics accessible to more people than ever before.

Cosmic Robotics: Robots for Mars

Cosmic Robotics isn't just thinking about Earth anymore — it's dreaming of Mars. The company's vision includes building robotic technology capable of performing heavy-duty tasks on the red planet, with the ultimate goal of supporting human colonization efforts.

The startup already has $25 million in contracts through 2027, and is racing against SpaceX's timeline for building on Mars. By the time we're ready to start colonizing other planets, these robots may be our first steps toward becoming an interplanetary species.

Parasma: Brain Cells as Compute Engines

Parasma is taking a different approach to AI efficiency by exploring how human brain cells might be used as an alternative to traditional computing hardware. This company is investigating whether using neural tissue could make AI systems more energy-efficient and powerful than current models.

This is science fiction turned real, but it's not just about futuristic concepts — it's about finding better ways to power our machines without sacrificing performance or increasing carbon footprints.

Waddle Labs: The ChatGPT Moment for Robotics

Finally, there's Waddle Labs, which is working to bring a ChatGPT-style experience to robotics. Instead of training foundation models on raw video or human teleoperation data, Waddle Labs uses LLM agents to write code and control robots directly.

The startup claims developers can tell the robot what to do in natural language, and its AI agents will autonomously generate executable control code — checking that it worked and setting up the robot in about 20 minutes. If successful, this could mark a turning point for general robotics models, similar to how large language models transformed natural language processing.

Conclusion: A New Era of Innovation

The startups featured at this YC Demo Day represent a new era of innovation — one where the focus is no longer on incremental improvements but on radical rethinking of fundamental systems. Whether it's how we power our data centers, how we train AI models, or even how we build robots for space exploration, these companies are pushing boundaries and changing the landscape.

What I find most compelling is that despite their ambitious goals, these startups have real-world applications today. They're not just theoretical exercises — they're addressing concrete challenges in energy, computing, security, and robotics with solutions that could transform industries.

The next few years will tell us if this batch of startups can deliver on their promises. But what's clear is that we're witnessing a shift toward more purposeful, impactful innovation — one that has the potential to solve some of our most pressing global issues.