{"id":3491807,"date":"2026-09-13T20:28:00","date_gmt":"2026-09-13T20:28:00","guid":{"rendered":"https:\/\/techingeek.com\/index.php\/2026\/09\/13\/the-9-most-talked-about-startups-from-y-combinators-recent-demo-day-per-vcs\/"},"modified":"2026-09-13T20:28:00","modified_gmt":"2026-09-13T20:28:00","slug":"the-9-most-talked-about-startups-from-y-combinators-recent-demo-day-per-vcs","status":"publish","type":"post","link":"https:\/\/techingeek.com\/index.php\/2026\/09\/13\/the-9-most-talked-about-startups-from-y-combinators-recent-demo-day-per-vcs\/","title":{"rendered":"The 9 most talked-about startups from Y Combinator\u2019s recent Demo Day, per VCs"},"content":{"rendered":"
<\/div>\n
\n

Another Y Combinator Demo Day was held on Thursday, and this cohort had a distinct feel. While each batch introduces a new wave of entrepreneurs, the startups showcased this week leaned significantly more toward deep tech compared to previous years.<\/p>\n

As is customary each quarter, TechCrunch reached out to early-stage VCs to identify the standout startups in this batch\u2014both their preferred selections and the discussions dominating the conversation. True to form, we compiled a list of startups highlighted by at least two investors as the most talked-about in the cohort.<\/p>\n

The technology represented in this batch felt \u201clike science fiction,\u201d per one investor’s description. However, there was a shared understanding that valuations were much more realistic than in some of the prior cohorts.<\/p>\n

Below, we present the top picks in alphabetical order:<\/p>\n

Automarine<\/strong>\u00a0<\/strong><\/p>\n

What it\u2019s developing: <\/strong>Nuclear-powered data centers situated on the ocean\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>With limited power supplies and local communities increasingly resistant to new data centers, Atomarine \u2014 co-founded by an MIT computer science and naval engineer alongside an MIT PhD in nuclear engineering \u2014 aims to address the computing shortage by situating data centers on barges at sea, leveraging seawater for near-free cooling. The startup plans to initiate a gas-powered pilot by 2028 and transition to floating nuclear power vessels by 2032. Atomarine claims to have garnered over $4 billion in customer interest via letters of intent, contributing to its position as one of the highest-valued startups in the batch, as noted by a VC.<\/p>\n

Dipole Labs\u00a0<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>High-speed optical networking hardware designed to be efficient and energy-saving for AI data centers<\/p>\n

Why it\u2019s a favorite: <\/strong>The concept revolves around GPU clusters, which often waste significant computation time waiting for data to transfer between chips. Within the networking component, data transitions from light to electricity and back, a process that consumes considerable power while generating excess heat. Dipole Labs asserts that it has engineered an optical switch that bypasses this conversion, allowing data to remain as light and travel directly to its destination. This solution is timely, given the prohibitive costs of GPUs and the push from data centers to maximize the efficiency of their computing hardware.<\/p>\n

Isengard Industries\u00a0<\/strong><\/p>\n

What it\u2019s developing:<\/strong> Jet-powered offense and defense drones that can be manufactured locally<\/p>\n

Why it\u2019s a favorite: <\/strong>Isengard is focused on mass-producing jet-powered attack and counter-drones directly in allied nations, at a fraction of the cost typically charged by prime contractors in the U.S. Co-founded by a former Australian Army officer and a defense entrepreneur who previously scaled another Ukraine-centric drone business to $60 million in revenue, Isengard is generating $10 million in revenue on its own. The startup has captured significant VC interest, claiming one of the highest valuations in this YC batch, according to two investors.<\/p>\n

Lamb Labs<\/p>\n

What it\u2019s developing: <\/strong>Custom inference chips with built-in AI model weights<\/p>\n

Why it\u2019s a favorite: <\/strong>Traditional AI hardware consumes large amounts of energy during inference phases while retrieving model weights from memory. Co-founded by an AI Ph.D. from Imperial College London and a theoretical physicist from Oxford, Lamb Labs seeks to create ultra-efficient chips by hardcoding AI model weights directly into silicon. Called \u201cModel Processing Units\u201d (MPUs), these custom chips aim to eliminate memory-bandwidth bottlenecks.<\/p>\n

Praxis AI<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Gathering real-world data for training robots\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>This company collaborates with businesses to gather video footage and data of humans performing various tasks, converting this material into training resources for companies that build robots. It claims to already work with publicly traded entities and has obtained video data from over 150 distinct environments. This endeavor could prove significant as businesses assess which tasks are best suited for humans and, as AI continues to evolve, which are better left to robots.\u00a0<\/p>\n

Nori<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Affordable robots for handling daily tasks\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>Since its launch just six weeks ago, this firm has almost half a million in sales. This is not surprising\u2014it offers a humanoid robot designed to assist with cleaning and folding laundry. Users can control their robot through a laptop app, with a price point of around $1,600, a bargain compared to other humanoid options like Neo, which costs about $20,000. A major question in robotics is whether it’s feasible to develop an affordable at-home robot capable of loading a dishwasher and performing routine household chores. Nori aims to address this challenge.\u00a0<\/p>\n

Cosmic Robotics\u00a0<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Self-operating robots capable of lifting substantial loads\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>The founders have an ambition to construct a city on Mars. The initial stage of this goal? Developing robotic technology for heavy-duty tasks. They claim their technology is already being utilized to install solar panels across the U.S. and has secured $25 million in contracts through 2027. Their vision is for this technology to aid in automating construction efforts for Mars colonization. The startup is racing against SpaceX’s timeline for Martian development, with hopes of commencing an exploratory mission by 2028.\u00a0<\/p>\n

Parasma<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Training human brain cells to potentially power computing\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>This company is investigating the most effective methods to meet the power demands of running models. Parasma is exploring the use of human brain cells as a potentially more energy-efficient alternative to current AI computing hardware.\u00a0<\/p>\n

Waddle Labs<\/p>\n

What it\u2019s developing:<\/strong> An API layer that generates robot control code<\/p>\n

Why it\u2019s a favorite:<\/strong> Investors and technology enthusiasts are anticipating a breakthrough moment for robotics akin to what ChatGPT achieved. This enthusiasm is fostering various strategies to create a general robotics model. Instead of training foundational models on raw videos or data derived from human teleoperation, Waddle Labs leverages a layer of LLM agents to generate code and manage robots directly. Founded by Harvard alumni, the startup aims to position itself as \u201cClaude Code for robotics.\u201d The startup asserts that by connecting any hardware to Waddle\u2019s API, developers can instruct the robot using natural language, enabling its AI agents to autonomously produce executable control code, verify its functionality, and set up the robot in approximately 20 minutes.<\/p>\n<\/div>\n

When you purchase through links in our articles, we may earn a small commission. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"

<\/div>\n
\n

Another Y Combinator Demo Day was held on Thursday, and this cohort had a distinct feel. While each batch introduces a new wave of entrepreneurs, the startups showcased this week leaned significantly more toward deep tech compared to previous years.<\/p>\n

As is customary each quarter, TechCrunch reached out to early-stage VCs to identify the standout startups in this batch\u2014both their preferred selections and the discussions dominating the conversation. True to form, we compiled a list of startups highlighted by at least two investors as the most talked-about in the cohort.<\/p>\n

The technology represented in this batch felt \u201clike science fiction,\u201d per one investor’s description. However, there was a shared understanding that valuations were much more realistic than in some of the prior cohorts.<\/p>\n

Below, we present the top picks in alphabetical order:<\/p>\n

Automarine<\/strong>\u00a0<\/strong><\/p>\n

What it\u2019s developing: <\/strong>Nuclear-powered data centers situated on the ocean\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>With limited power supplies and local communities increasingly resistant to new data centers, Atomarine \u2014 co-founded by an MIT computer science and naval engineer alongside an MIT PhD in nuclear engineering \u2014 aims to address the computing shortage by situating data centers on barges at sea, leveraging seawater for near-free cooling. The startup plans to initiate a gas-powered pilot by 2028 and transition to floating nuclear power vessels by 2032. Atomarine claims to have garnered over $4 billion in customer interest via letters of intent, contributing to its position as one of the highest-valued startups in the batch, as noted by a VC.<\/p>\n

Dipole Labs\u00a0<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>High-speed optical networking hardware designed to be efficient and energy-saving for AI data centers<\/p>\n

Why it\u2019s a favorite: <\/strong>The concept revolves around GPU clusters, which often waste significant computation time waiting for data to transfer between chips. Within the networking component, data transitions from light to electricity and back, a process that consumes considerable power while generating excess heat. Dipole Labs asserts that it has engineered an optical switch that bypasses this conversion, allowing data to remain as light and travel directly to its destination. This solution is timely, given the prohibitive costs of GPUs and the push from data centers to maximize the efficiency of their computing hardware.<\/p>\n

Isengard Industries\u00a0<\/strong><\/p>\n

What it\u2019s developing:<\/strong> Jet-powered offense and defense drones that can be manufactured locally<\/p>\n

Why it\u2019s a favorite: <\/strong>Isengard is focused on mass-producing jet-powered attack and counter-drones directly in allied nations, at a fraction of the cost typically charged by prime contractors in the U.S. Co-founded by a former Australian Army officer and a defense entrepreneur who previously scaled another Ukraine-centric drone business to $60 million in revenue, Isengard is generating $10 million in revenue on its own. The startup has captured significant VC interest, claiming one of the highest valuations in this YC batch, according to two investors.<\/p>\n

Lamb Labs<\/p>\n

What it\u2019s developing: <\/strong>Custom inference chips with built-in AI model weights<\/p>\n

Why it\u2019s a favorite: <\/strong>Traditional AI hardware consumes large amounts of energy during inference phases while retrieving model weights from memory. Co-founded by an AI Ph.D. from Imperial College London and a theoretical physicist from Oxford, Lamb Labs seeks to create ultra-efficient chips by hardcoding AI model weights directly into silicon. Called \u201cModel Processing Units\u201d (MPUs), these custom chips aim to eliminate memory-bandwidth bottlenecks.<\/p>\n

Praxis AI<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Gathering real-world data for training robots\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>This company collaborates with businesses to gather video footage and data of humans performing various tasks, converting this material into training resources for companies that build robots. It claims to already work with publicly traded entities and has obtained video data from over 150 distinct environments. This endeavor could prove significant as businesses assess which tasks are best suited for humans and, as AI continues to evolve, which are better left to robots.\u00a0<\/p>\n

Nori<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Affordable robots for handling daily tasks\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>Since its launch just six weeks ago, this firm has almost half a million in sales. This is not surprising\u2014it offers a humanoid robot designed to assist with cleaning and folding laundry. Users can control their robot through a laptop app, with a price point of around $1,600, a bargain compared to other humanoid options like Neo, which costs about $20,000. A major question in robotics is whether it’s feasible to develop an affordable at-home robot capable of loading a dishwasher and performing routine household chores. Nori aims to address this challenge.\u00a0<\/p>\n

Cosmic Robotics\u00a0<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Self-operating robots capable of lifting substantial loads\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>The founders have an ambition to construct a city on Mars. The initial stage of this goal? Developing robotic technology for heavy-duty tasks. They claim their technology is already being utilized to install solar panels across the U.S. and has secured $25 million in contracts through 2027. Their vision is for this technology to aid in automating construction efforts for Mars colonization. The startup is racing against SpaceX’s timeline for Martian development, with hopes of commencing an exploratory mission by 2028.\u00a0<\/p>\n

Parasma<\/strong><\/strong><\/p>\n

What it\u2019s developing: <\/strong>Training human brain cells to potentially power computing\u00a0<\/p>\n

Why it\u2019s a favorite: <\/strong>This company is investigating the most effective methods to meet the power demands of running models. Parasma is exploring the use of human brain cells as a potentially more energy-efficient alternative to current AI computing hardware.\u00a0<\/p>\n

Waddle Labs<\/p>\n

What it\u2019s developing:<\/strong> An API layer that generates robot control code<\/p>\n

Why it\u2019s a favorite:<\/strong> Investors and technology enthusiasts are anticipating a breakthrough moment for robotics akin to what ChatGPT achieved. This enthusiasm is fostering various strategies to create a general robotics model. Instead of training foundational models on raw videos or data derived from human teleoperation, Waddle Labs leverages a layer of LLM agents to generate code and manage robots directly. Founded by Harvard alumni, the startup aims to position itself as \u201cClaude Code for robotics.\u201d The startup asserts that by connecting any hardware to Waddle\u2019s API, developers can instruct the robot using natural language, enabling its AI agents to autonomously produce executable control code, verify its functionality, and set up the robot in approximately 20 minutes.<\/p>\n<\/div>\n

When you purchase through links in our articles, we may earn a small commission. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n","protected":false},"author":2,"featured_media":3491808,"comment_status":"open","ping_status":"closed","sticky":false,"template":"Default","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-3491807","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/posts\/3491807"}],"collection":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/comments?post=3491807"}],"version-history":[{"count":0,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/posts\/3491807\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/media\/3491808"}],"wp:attachment":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/media?parent=3491807"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/categories?post=3491807"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/tags?post=3491807"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}