Another Y Combinator Demo Day batch leans heavily into deep tech with grounded valuations.

The latest Y Combinator Demo Day, held on Thursday, showcased a distinct shift in the startup landscape, with the presenting cohort demonstrating a pronounced inclination towards deep tech ventures. Unlike previous batches, this group of founders introduced innovations that one investor described as feeling "like science fiction," pushing the boundaries of what’s currently feasible. While the ideas were undeniably ambitious and often groundbreaking, a notable consensus among early-stage venture capitalists was that the valuations assigned to these startups were considerably more grounded than those observed in recent cohorts, reflecting a more cautious yet appreciative investment climate.

As is customary each quarter, TechCrunch engaged with early-stage VCs to identify the most promising startups from this batch. Investors were asked to name their top picks, as well as the deals that were generating widespread buzz within the community. From these insights, TechCrunch compiled a curated list featuring startups that were flagged by at least two investors, ensuring a robust and validated selection of the cohort’s most impactful and talked-about companies. These selections highlight the innovative spirit of the batch and the areas drawing significant investor attention.

The technological advancements presented in this cohort were a primary topic of discussion, with many remarking on their futuristic nature. Beyond the sheer audacity of these concepts, the tempered valuations offered a pragmatic counterpoint, suggesting a maturing market where revolutionary ideas are still highly sought after, but with a more disciplined approach to investment. The following startups, listed alphabetically, represent the top picks from this unique Y Combinator Demo Day:

Automarine
Automarine is pioneering the development of nuclear-powered data centers designed to float at sea. This innovative approach directly addresses two critical challenges facing the rapidly expanding data center industry: a severe shortage of power and increasing community opposition to the construction of new land-based facilities. The startup, co-founded by an MIT computer science and naval engineer alongside an MIT PhD in nuclear engineering, proposes a solution where data centers are housed on barges. This maritime setting offers a unique advantage, as seawater can provide a nearly free and abundant cooling source, significantly reducing operational costs and environmental impact. Atomarine’s ambitious roadmap includes launching a gas-powered pilot facility by 2028, with plans to transition to floating nuclear power ships by 2032. The company has already demonstrated significant market validation, claiming to have secured over $4 billion in customer interest through letters of intent. This substantial revenue potential has positioned Atomarine as one of the highest-valued startups in the current YC batch, according to venture capitalists tracking the cohort. Their vision for decentralized, energy-independent computing infrastructure represents a bold leap forward in addressing global compute demands.

Cosmic Robotics
Cosmic Robotics is focused on building autonomous robots capable of lifting heavy objects, with an audacious long-term vision: to construct a city on Mars. The founders believe that developing advanced robotic technology for heavy-duty tasks on Earth is the foundational first step toward automating construction for future off-world colonization efforts. Their technology is already being deployed in real-world applications, with the company actively installing solar panels across the United States. Cosmic Robotics has secured an impressive $25 million in contracts extending through 2027, underscoring the immediate commercial viability of their solutions. The startup is strategically aligning its development timeline with SpaceX’s ambitious plans for Mars, aspiring to begin an exploratory mission by 2028. This dual focus on terrestrial utility and extraterrestrial ambition makes Cosmic Robotics a compelling deep tech venture, tapping into both immediate market needs and the long-term future of space exploration and habitation.

Dipole Labs
Dipole Labs is tackling a critical bottleneck within AI data centers by developing effective and energy-efficient high-speed optical networking hardware. The core problem it addresses is the significant amount of compute time wasted by GPU clusters waiting for data to traverse between individual chips. Within the networking layer of traditional data centers, data undergoes a constant conversion process from light to electricity and back again. This conversion is not only energy-intensive, generating considerable heat, but also introduces latency, slowing down processing. Dipole Labs claims to have engineered an optical switch that eliminates this conversion step, allowing data to remain as light and travel directly to its destination. This innovation is particularly timely given the exorbitant cost of GPUs and the imperative for data centers to maximize the utilization of their compute hardware. By minimizing wasted compute time and improving energy efficiency, Dipole Labs offers a solution that could significantly enhance the performance and sustainability of modern AI infrastructure, addressing a "billion-dollar bottleneck" that plagues high-performance computing.

Isengard Industries
Isengard Industries is making significant strides in the defense sector by building locally producible, jet-powered strike and counter-drones. The company’s mission is to offer these advanced defense assets to allied countries at a fraction of the cost typically charged by prime contractors in the U.S. This localized production model not only enhances affordability but also strengthens supply chain resilience for allied nations. The startup benefits from an experienced leadership team, 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 Industries has quickly demonstrated its commercial prowess, already generating $10 million in revenue. This rapid traction, combined with its strategic market positioning, has garnered strong venture capital interest, leading to one of the loftiest valuations observed in this YC batch, according to two investors. Their approach addresses the growing demand for accessible, high-performance unmanned aerial systems in an evolving geopolitical landscape.

Lamb Labs
Lamb Labs is addressing the substantial energy consumption of AI inference by developing custom inference chips featuring hardcoded AI model weights. Traditional AI chips expend massive amounts of energy retrieving model weights from memory during the inference process, which is the stage where a trained AI model makes predictions or decisions. The company, co-founded by an Imperial College London AI Ph.D. and an Oxford theoretical physicist, proposes a radical solution: building ultra-efficient chips by directly embedding, or "hardcoding," AI model weights into the silicon itself. These innovative custom chips, dubbed "Model Processing Units" (MPUs), are designed to eliminate the memory-bandwidth bottlenecks that plague conventional architectures. By removing the need to constantly fetch data from external memory, Lamb Labs aims to drastically reduce the energy footprint and increase the speed of AI inference. This advancement could have profound implications for the sustainability and scalability of AI, making powerful AI models more accessible and environmentally friendly for deployment across various applications.

Nori
Nori is attracting significant attention with its focus on creating affordable robots capable of handling everyday household tasks. Launched just six weeks prior to Demo Day, the company has already reported nearly half a million dollars in sales, a testament to the immediate appeal of its offering. Nori’s flagship product is a humanoid robot designed to assist with chores like cleaning and folding clothes, which users can operate conveniently via a laptop application. Priced at approximately $1,600, Nori’s robot stands out in the market for its accessibility, especially when compared to other humanoid robots like Neo, which typically retail for around $20,000. One of the enduring challenges in robotics has been to develop an affordable, at-home robot that can reliably perform common domestic tasks, such as stacking a dishwasher. Nori represents a bold new attempt to tackle this challenge, aiming to bring practical and accessible robotics into the everyday lives of consumers, potentially democratizing home automation.

Parasma
Parasma is exploring one of the most intriguing and forward-thinking solutions to the escalating power and energy demands of AI models: training human brain cells to one day power compute. This pioneering company is delving into the realm of bio-computing, researching whether biological neural networks can serve as a more effective and energy-efficient alternative to current AI computing hardware. As artificial intelligence models grow in complexity and scale, the energy required to train and run them becomes an increasingly significant concern, both economically and environmentally. Parasma’s research into leveraging the inherent efficiency of biological systems for computation represents a radical departure from conventional silicon-based approaches. While still in its early stages, this venture highlights a profound commitment to finding sustainable and novel solutions for the future of computing, potentially unlocking unprecedented levels of energy efficiency and processing power by tapping into the very architecture of intelligence itself.

Praxis AI
Praxis AI is addressing a critical need in the burgeoning field of robotics: collecting real-world data specifically tailored for training robots. The company partners with businesses to capture videos and other forms of data depicting humans performing various tasks in their work environments. This raw data is then meticulously processed and transformed into high-quality training material for companies developing and deploying robots. Praxis AI has already demonstrated significant traction, actively collaborating with publicly traded companies and having successfully captured video data in over 150 diverse environments. This service is poised to become increasingly vital as businesses deepen their experimentation with automation, discerning which tasks are best suited for human execution and which can be more efficiently handled by robots as AI continues to refine its capabilities. By providing rich, contextualized real-world data, Praxis AI plays a pivotal role in accelerating the development of more capable, adaptable, and intelligent robotic systems across various industries.

Waddle Labs
Waddle Labs is positioning itself at the forefront of the general robotics movement, aiming to create an API layer that writes robot control code using advanced Large Language Model (LLM) agents. The startup is driven by the widespread excitement surrounding the potential for a "ChatGPT moment" in robotics, where robots can be controlled and programmed with unprecedented ease and flexibility. Unlike traditional approaches that train foundational models on raw video or human teleoperation data, Waddle Labs employs a novel strategy. Its system uses a layer of LLM agents to directly generate and execute control code for robots. Founded by Harvard graduates, the company describes its offering as "Claude Code for robotics," drawing an analogy to the sophisticated code generation capabilities of modern LLMs. Waddle Labs claims that by integrating any hardware with its API, developers will be able to command robots in natural language. Its AI agents will then autonomously generate the executable control code, verify its successful execution, and set up the robot for its task in approximately 20 minutes, promising a significant leap in the accessibility and versatility of robotic programming.

Marina Temkin is a venture capital and startups reporter at TechCrunch. Prior to joining TechCrunch, she wrote about VC for PitchBook and Venture Capital Journal. Earlier in her career, Marina was a financial analyst and earned a CFA charterholder designation. You can contact or verify outreach from Marina by emailing [email protected] or via encrypted message at +1 347-683-3909 on Signal.

Dominic-Madori Davis is a senior venture capital and startup reporter at TechCrunch. She is based in New York City. You can contact or verify outreach from Dominic by emailing [email protected] or via encrypted message at +1 646 831-7565 on Signal.