The Deep Tech Revolution at Y Combinator: Inside the Buzziest Startups of the Latest Demo Day

By TechCrunch Venture Capital Reporting Team

Another Y Combinator Demo Day concluded on Thursday, offering a window into the evolving ambitions of early-stage entrepreneurship. While every cohort brings a fresh wave of founders tackling diverse market segments, the startups presenting this week skewed far more toward "deep tech" and science-fiction-adjacent engineering than in past cohorts.

As is customary every quarter, early-stage venture capitalists were polled to identify the hottest startups of the batch—highlighting both personal top picks and the deals dominating back-channel conversations. Investors consistently described the cohort’s technological focus as “like science fiction,” yet noted a refreshing silver lining: unlike the hyper-inflated, speculative valuations of recent years, this batch’s pricing remained notably grounded.

Here is an in-depth look at the top-performing, most heavily flagged startups from the latest Y Combinator batch, categorized by their structural impact on AI infrastructure, defense, robotics, and biological computing.

Main Facts: A Shift Toward Hard Science and Infrastructure

The overarching theme of the latest Y Combinator Demo Day is an unmistakable pivot away from surface-level software-as-a-service (SaaS) and toward foundational, heavy-infrastructure challenges. With the global compute crunch, surging energy demands, and geopolitical tensions redefining the tech landscape, founders are addressing physical bottlenecks rather than digital ones.

Key highlights from the investor consensus include:

• Infrastructure at Sea and in Silicon: Startups like Atomarine and Dipole Labs are tackling data center power shortages through radical means, ranging from floating nuclear barges to high-speed optical networking.
• Defense and Dual-Use Tech: Companies such as Isengard Industries are capitalizing on localized, cost-effective manufacturing of advanced military hardware.
• The Robotics Boom: From affordable consumer humanoids (Nori) to automated Mars-colonization infrastructure (Cosmic Robotics) and LLM-driven robot control layers (Waddle Labs), robotics took center stage.
• Biological Frontiers: Experimental startups like Parasma are pushing the boundaries of computing by exploring human brain cells as energy-efficient computational alternatives.

Chronology: The Evolution to Deep Tech

The journey toward this deep-tech-heavy YC cohort has been building for several quarters, accelerated primarily by the generative AI boom.

• Late 2022 to 2023: The launch of mainstream generative AI models triggered an unprecedented surge in demand for compute power, GPUs, and electricity. Data centers strained under the weight of AI inference loads.
• 2024: Power grids began showing strain, and local communities increasingly pushed back against the physical buildout of massive onshore data centers. Simultaneously, geopolitical conflicts accelerated the demand for domestic and allied defense manufacturing, particularly autonomous drones.
• The Recent Demo Day (Thursday): Culminating months of Y Combinator acceleration, the latest batch showcased a direct response to these macro-challenges. Founders unveiled hard-science solutions designed to bypass traditional silicon, land-based power limitations, and expensive robotics training pipelines.

Supporting Data: The Standout Startups and Their Metrics

Investors flagged several companies by name, noting that they commanded some of the highest valuations and strongest pre-launch traction in the batch.

1. Atomarine: Floating Nuclear Data Centers

• The Problem: Global power supplies are constrained, and local communities are increasingly blocking the construction of new land-based data centers.
• The Solution: Co-founded by an MIT computer science and naval engineer alongside an MIT PhD in nuclear engineering, Atomarine builds nuclear-powered data centers that float at sea. Seawater provides near-free cooling.
• Traction & Valuation: The startup plans a gas-powered pilot by 2028, followed by a transition to floating nuclear power ships in 2032. Atomarine has already secured over $4 billion in customer interest through letters of intent (LOIs), making it one of the highest-valued startups in the batch.

2. Dipole Labs: Optical Networking for AI

• The Problem: GPU clusters waste massive amounts of compute time waiting for data to move between chips. Traditional networking layers constantly convert data between light and electricity, burning power and generating immense heat.
• The Solution: Dipole Labs developed an optical switch that skips conversion entirely, allowing data to remain as light and travel directly where it needs to go.

3. Isengard Industries: Affordable Jet-Powered Drones

• The Problem: Prime defense contractors charge exorbitant prices to build strike and counter-drones in the U.S., limiting scale for allied nations.
• The Solution: Co-founded by a former Australian Army officer and a defense entrepreneur who previously scaled a Ukraine-focused drone startup to $60 million in revenue, Isengard aims to mass-produce jet-powered attack and counter-drones locally within allied countries.
• Traction: The company is already generating $10 million in revenue and has secured a top-tier valuation among investors.

4. Lamb Labs: Custom Chips for AI Inference

• The Problem: Traditional AI chips burn excessive energy during inference simply fetching model weights from memory.
• The Solution: Founded by an Imperial College London AI Ph.D. and an Oxford theoretical physicist, Lamb Labs creates "Model Processing Units" (MPUs) that hardcode AI model weights directly into silicon, eliminating memory-bandwidth bottlenecks.

5. Praxis AI: Training Data for Robotics

• The Problem: Building generalized robots requires massive amounts of real-world training data showcasing humans performing physical labor.
• The Solution: Praxis partners with businesses to capture video and operational data across diverse environments—already working with publicly traded companies across more than 150 different settings.

6. Nori: Affordable At-Home Robotics

• The Problem: Humanoid robots are traditionally prohibitively expensive (often retailing around $20,000), making everyday household automation economically unviable.
• The Solution: Launched just six weeks prior to Demo Day, Nori introduced a humanoid robot priced at roughly $1,600 designed to handle chores like cleaning and folding clothes, controllable via a laptop app. The company has already recorded nearly $500,000 in sales.

7. Cosmic Robotics: Heavy-Duty Automation for Earth and Mars

• The Problem: Scaling automated construction requires heavy-duty robotics capable of operating reliably in harsh, unstructured environments.
• The Solution: With a long-term vision of helping build a city on Mars by 2028, Cosmic Robotics builds autonomous heavy-duty robots. Its technology is already deployed installing solar panels across the U.S. and boasts $25 million in contracts through 2027.

8. Parasma: Biological Computing

• The Problem: Silicon-based AI hardware demands unsustainable amounts of energy and electricity to scale.
• The Solution: Parasma is researching methods to train human brain cells to eventually serve as an ultra-energy-efficient alternative to traditional computing hardware.

9. Waddle Labs: An API for Robot Control Code

• The Problem: Training foundation models on raw video or human teleoperation data is slow and fragmented; developers lack an easy way to command hardware using natural language.
• The Solution: Founded by Harvard graduates, Waddle Labs uses a layer of Large Language Model (LLM) agents to write executable robot control code. Billed as "Claude Code for robotics," the API allows developers to plug in any hardware, issue a natural language command, and have the system generate code, test it, and configure the robot in about 20 minutes.

Official Responses and Investor Perspectives

The overarching sentiment from the venture capital community during this Demo Day was one of cautious optimism tempered by awe. While the concepts border on science fiction—ranging from floating nuclear reactors in the ocean to biological brain-cell computing—investors emphasized that the economic fundamentals driving these startups are deeply rooted in current market failures.

"Power is the ultimate bottleneck for the next decade of computing," noted one early-stage VC regarding startups like Atomarine and Dipole Labs. "Founders aren’t just building apps anymore; they are rewriting the physical laws of how data centers and automation operate."

Furthermore, investors highlighted the discipline shown in round valuations. Unlike the runaway multiples seen during the 2021 tech boom, founders this cycle appeared more pragmatic, aligning their valuation expectations with the capital-intensive, long-horizon nature of deep tech.

Implications: What This Means for the Future of Tech

The heavy tilt toward deep tech, defense, and infrastructure at Y Combinator signals a fundamental maturation of the startup ecosystem.

1. The Infrastructure Imperative: As generative AI capabilities scale, the limiting factor is no longer algorithmic brilliance, but energy, cooling, and silicon efficiency. Companies that solve these physical bottlenecks will capture immense market share over the next decade.
2. Dual-Use Normalization: The inclusion of defense-focused startups like Isengard highlights a broader cultural shift within Silicon Valley. Venture capital is increasingly welcoming national security and allied defense tech as core pillars of modern innovation.
3. The Democratization of Robotics: With low-cost consumer humanoids like Nori and developer-friendly APIs like Waddle Labs entering the market, the "ChatGPT moment for robotics" may transition from theoretical discourse to commercial reality sooner than anticipated.

Ultimately, this YC batch demonstrates that the next generation of venture-backed giants will not be built on software screens alone, but on the atomic, mechanical, and biological systems that power the modern world.