By Titik Puspa 
The latest Y Combinator Demo Day, held on Thursday, presented a notable shift in focus, with startups leaning heavily into deep technology—a trend that distinguished this cohort from its predecessors. This quarterly showcase, a crucial event for early-stage venture capitalists, saw a significant increase in companies tackling complex, foundational scientific and engineering challenges. TechCrunch, as is its custom, polled a range of early-stage VCs to identify the most promising ventures from this batch, discerning both their top selections and the startups generating the most industry buzz. The collective feedback painted a picture of a cohort pushing the boundaries of what was previously considered science fiction, yet with valuations that appeared more pragmatic and grounded than in recent funding cycles.
The investor consensus highlighted the "science fiction-like" nature of the technologies presented, a testament to the ambitious scope of the founders. This advanced technological orientation, coupled with a more measured approach to valuation, has created a fertile ground for investment and innovation. Below is an in-depth look at the startups that captured the attention of investors and the broader tech community, presented alphabetically.
Automarine: Powering the Future of Data Centers at Sea
Automarine is pioneering a radical solution to the growing global demand for computing power and the increasing opposition to new data center construction: nuclear-powered data centers floating at sea. This ambitious venture, co-founded by individuals with deep expertise from MIT in computer science, naval engineering, and nuclear engineering, aims to alleviate the compute shortage by deploying data centers on barges positioned offshore. The strategic advantage of this approach lies in the readily available and cost-effective cooling provided by seawater. The company has outlined an aggressive roadmap, with plans to launch a gas-powered pilot project by 2028, followed by a transition to floating nuclear power ships by 2032. The potential for massive revenue generation has not gone unnoticed; Automarine claims to have already secured over $4 billion in customer interest through letters of intent. This substantial market validation has contributed to its status as one of the highest-valued startups in the current YC batch, according to industry analysts. The convergence of escalating data demands, environmental concerns, and the inherent challenges of terrestrial infrastructure development positions Automarine’s innovative approach as a potentially transformative solution for the future of cloud computing and data processing.
Dipole Labs: Revolutionizing AI Data Center Networking
Dipole Labs is addressing a critical bottleneck in the explosive growth of artificial intelligence: the energy-intensive and time-consuming process of data transfer within AI data centers. Their innovation lies in developing effective and energy-efficient high-speed optical networking hardware designed specifically for these demanding environments. The core problem Dipole Labs tackles is the inefficiency inherent in current GPU clusters, where significant compute time is wasted waiting for data to traverse between chips. Traditional networking methods involve converting data from light to electricity and back, a process that consumes considerable power and generates substantial heat. Dipole Labs’ solution bypasses this conversion by employing an optical switch that allows data to remain in its light form, traveling directly to its destination. This direct optical path drastically reduces latency and energy consumption. As the cost of GPUs continues to soar and data centers strive to maximize the utility of their existing hardware, Dipole Labs’ technology offers a compelling pathway to enhanced performance and reduced operational expenses. Their approach is particularly timely, enabling AI hardware to operate at peak efficiency and minimizing idle time, thereby unlocking greater computational power from expensive resources.
Isengard Industries: Accelerating Defense Drone Production
Isengard Industries is poised to disrupt the defense sector by focusing on the localized, mass production of jet-powered strike and counter-drones. Their mission is to provide allied nations with advanced aerial capabilities at a significantly lower cost than traditional prime contractors typically charge for manufacturing in the U.S. The company’s founding team brings a wealth of relevant experience: one co-founder is a former Australian Army officer with deep insights into military operational needs, while the other is a defense entrepreneur who previously scaled a Ukraine-focused drone startup to an impressive $60 million in revenue. This combination of strategic vision and proven execution capability is already yielding results. Isengard Industries is currently generating $10 million in revenue and has garnered significant attention from venture capitalists, securing one of the highest valuations in this YC batch, according to multiple investors. The geopolitical landscape, marked by increasing global instability and the growing importance of drone warfare, makes Isengard’s ability to rapidly deploy cost-effective, high-performance drones a strategically vital offering. Their model of localized production not only reduces costs but also enhances supply chain resilience and allows for quicker adaptation to the specific needs of individual allied forces.
Lamb Labs: Custom Chips for Ultra-Efficient AI Inference
Lamb Labs is tackling the significant energy demands of AI inference with the development of custom chips that feature hardcoded AI model weights. Traditional AI chips expend vast amounts of energy during inference as they constantly fetch model weights from external memory. This process creates a substantial performance bottleneck and contributes to the overall energy footprint of AI systems. Co-founded by an AI Ph.D. from Imperial College London and a theoretical physicist from Oxford, Lamb Labs proposes a paradigm shift by embedding AI model weights directly into silicon. These custom chips, termed "Model Processing Units" (MPUs), are designed to eliminate the memory-bandwidth bottlenecks that plague conventional architectures. By hardcoding the weights, the need for constant memory retrieval is obviated, leading to dramatically improved energy efficiency and processing speeds. This innovation is particularly crucial as the deployment of AI models moves beyond large data centers to edge devices and resource-constrained environments, where power consumption is a paramount concern. Lamb Labs’ approach promises to make AI inference more accessible, sustainable, and cost-effective across a wide range of applications.
Praxis AI: Curating Real-World Data for Robot Training
Praxis AI is dedicated to solving a fundamental challenge in the advancement of robotics: the acquisition of high-quality, real-world data necessary for effective robot training. The company partners with businesses to capture video and other data streams of humans performing various tasks. This rich dataset is then meticulously curated and transformed into training material for companies developing robotic systems. Praxis AI’s ability to collect data across more than 150 different environments, and its current engagement with publicly traded companies, underscores the practical value of its service. As businesses increasingly explore the integration of robots into their operations, the question of which tasks are best suited for human or robotic execution becomes critical. Praxis AI’s work directly supports this transition by providing the essential data fuel for AI models to learn and refine their capabilities, enabling robots to perform tasks more efficiently and effectively. Their contribution is vital for accelerating the adoption of robotics in diverse industries, from manufacturing and logistics to customer service and healthcare.
Nori: Bringing Affordable Humanoid Robotics to Everyday Tasks
Nori is making waves in the robotics industry with its focus on creating affordable humanoid robots capable of handling everyday domestic tasks. Launched just six weeks prior to its Demo Day presentation, the company has already achieved nearly half a million dollars in sales, a testament to the market’s appetite for such solutions. Nori’s flagship offering is a humanoid robot designed to assist with household chores, including cleaning and folding clothes. Users can control their robot via a user-friendly laptop application, providing an intuitive interface for managing domestic tasks. What sets Nori apart is its aggressive pricing strategy; the robot is priced at approximately $1,600, a stark contrast to other high-profile humanoids like Neo, which retail for around $20,000. The perennial question in robotics is whether an affordable, functional at-home robot can be realized. Nori represents a significant stride towards answering that question, offering a practical and accessible solution for users seeking assistance with mundane yet time-consuming household responsibilities. The company’s rapid market traction suggests a strong demand for domestic robots that are both capable and economically viable.
Cosmic Robotics: Engineering for Martian Colonization and Beyond
Cosmic Robotics harbors an ambitious long-term vision: to build a city on Mars. To achieve this monumental goal, the company is focusing on developing robotic technology capable of performing heavy-duty tasks. The immediate application of their innovative tech is already underway, with their robots currently installing solar panels across the United States. Cosmic Robotics has secured $25 million in contracts extending through 2027, demonstrating the commercial viability of its solutions in terrestrial markets. The ultimate aim is to leverage this advanced robotic capability to automate construction efforts for future Martian colonies. The startup’s timeline is closely aligned with the ambitious space exploration goals of organizations like SpaceX, with plans for an exploratory mission to Mars targeted for 2028. Cosmic Robotics’ dual focus on immediate commercial applications and long-term, transformative goals positions them as a significant player in the future of both terrestrial infrastructure development and extraterrestrial colonization. Their work highlights the critical role of robotics in enabling humanity’s expansion beyond Earth.
Parasma: Harnessing Human Brain Cells for Future Compute
Parasma is exploring a truly groundbreaking approach to address the immense power and energy requirements of modern AI models: utilizing human brain cells to power future computation. This venture is part of a broader trend of seeking novel and more energy-efficient alternatives to current AI computing hardware. By investigating the potential of biological computing, Parasma aims to develop a solution that could offer significant advantages in terms of energy efficiency and processing power. The idea of integrating biological components with computational systems represents a frontier of scientific and technological exploration. While still in its early stages, Parasma’s research into harnessing the inherent computational power and efficiency of human brain cells holds the potential to revolutionize the field of artificial intelligence and computing as we know it. This innovative direction underscores the diverse and often unconventional paths being explored by deep tech startups to solve fundamental challenges in the field.
Waddle Labs: The "Claude Code for Robotics"
Waddle Labs is positioning itself as a transformative force in robotics, aiming to replicate the "ChatGPT moment" for robot control. The company’s core innovation is an API layer that writes robot control code, abstracting away the complexities of programming for diverse hardware. Instead of relying on traditional methods like training foundation models on raw video or human teleoperation data, Waddle Labs employs a layer of Large Language Model (LLM) agents to directly generate and execute robot control code. Founded by Harvard graduates, the startup describes its offering as "Claude Code for robotics," emphasizing its user-friendly, natural language interface. Waddle Labs claims that by simply plugging any hardware into their API, developers will be able to instruct robots in plain English, and the AI agents will autonomously generate, verify, and deploy executable control code, setting up the robot in approximately 20 minutes. This capability promises to dramatically lower the barrier to entry for robot programming, accelerating the development and deployment of robotic systems across various industries and applications. The ability to control complex machinery through simple, intuitive commands could unlock a new era of accessibility and innovation in robotics.
