Y Combinator’s latest Demo Day offered a revealing look at where early-stage technology is heading next.
The Summer 2026 YC batch was notable not simply because of the number of AI startups, but because many founders were tackling problems underneath the AI boom: electricity, compute infrastructure, semiconductors, robotics, physical-world data and even biological computing.
That shift is important.
The first wave of generative AI created companies around software applications and AI assistants. The next wave increasingly appears to be about building the physical infrastructure that makes intelligent machines possible.
After speaking with early-stage investors, TechCrunch identified nine startups that generated particularly strong interest during the latest YC Demo Day. We reviewed those companies and their publicly available information to understand what makes them interesting and what broader technology trends they represent. (TechCrunch)
Here are the nine startups worth watching.
1. Atomarine: Taking AI Data Centers Offshore
AI requires enormous amounts of computing power, and computing power requires electricity, cooling, land and grid capacity.
Atomarine is approaching that problem from an unusual direction: move the data center offshore.
The YC-backed startup is developing floating compute infrastructure supported by nuclear power ships. Its concept involves standardized data-center barges that can operate at sea, where seawater can provide cooling and where operators can potentially avoid some of the land, power-grid and community constraints associated with large data-center projects. (Y Combinator)
Atomarine says it plans to test a gas-powered pilot before moving toward nuclear-powered floating infrastructure.
The bigger idea is more interesting than the vessel itself: AI infrastructure may increasingly move to wherever power and cooling are easiest to obtain.
That makes Atomarine a company worth watching as AI data-center demand continues to put pressure on traditional infrastructure.
2. Dipole Labs: Rethinking How AI Data Centers Move Data
The AI hardware conversation usually focuses on GPUs.
But as AI clusters become larger, the network connecting those GPUs can become just as important.
Dipole Labs is building optical networking technology designed to move data through AI clusters more efficiently. Its approach keeps data in the optical domain rather than repeatedly converting between light and electrical signals, potentially reducing latency, heat and energy consumption. (Y Combinator)
This is an example of a less visible part of the AI infrastructure race.
The question is no longer simply:
How do we build a faster AI chip?
It is increasingly:
How do we make thousands of expensive AI chips work together without wasting their capacity?
If AI clusters continue scaling, networking could become one of the industry’s most important bottlenecks.
3. Isengard Industries: Building Defense Technology Locally
Isengard Industries is taking a different route into the physical-AI economy.
The startup focuses on locally producible defense and counter-drone systems, with the broader goal of allowing allied countries to manufacture these systems domestically rather than relying entirely on large traditional defense contractors. YC lists the company as part of its Summer 2026 batch. (Y Combinator)
The company says it already has revenue and a substantial pipeline.
What makes Isengard particularly notable from a startup-investment perspective is the intersection of AI, robotics, manufacturing and defense supply chains.
The defense startup market is increasingly moving beyond software dashboards and analytics. Investors are looking at companies that can build actual hardware and manufacturing capacity.
That is a much harder problem — but potentially a much larger market.
4. Lamb Labs: Putting AI Models Directly Into Silicon
AI inference consumes significant amounts of computing power.
One reason is that conventional systems constantly move model parameters between memory and processing hardware.
Lamb Labs is pursuing a radically different approach: custom chips that hardcode AI model weights directly into silicon.
The company calls these processors Model Processing Units, or MPUs. YC describes the architecture as an attempt to reduce the memory bottleneck associated with conventional GPU-based inference. (Y Combinator)
The significance goes beyond creating another AI chip.
If AI becomes embedded in everything from robots to industrial machines and consumer devices, energy efficiency will become increasingly important.
The winners may not always be the companies building the biggest models.
Some may be the companies figuring out how to run those models using dramatically less energy.
5. Praxis: Turning Human Work Into Robotics Training Data
Robots need data.
Lots of it.
And unlike an AI chatbot, a robot cannot learn everything it needs to know simply by reading the internet.
Praxis is working on this physical-world data problem by capturing how humans perform real tasks and turning that information into training data for robotics companies. YC describes the company as transforming real-world work into robotics training data.
That could become an important layer in the emerging robotics ecosystem.
Imagine a factory where workers perform thousands of different physical tasks. Every movement, decision and interaction can potentially become useful information for training machines.
This creates a new category of infrastructure:
data collection for physical AI.
As robotics expands beyond controlled laboratory environments, companies capable of collecting high-quality real-world training data could become increasingly valuable.
6. Nori: A $1,688 Bet on Consumer Robotics
Humanoid robots have captured enormous attention, but price remains one of the biggest barriers to widespread adoption.
Nori is trying to attack that problem directly.
The YC startup is developing a relatively affordable bimanual mobile robot designed to perform everyday tasks such as cleaning, restocking and folding clothes. YC currently lists the robot at $1,688, with the company reporting more than $440,000 in sales shortly after launch. (Y Combinator)
That price point is what makes Nori particularly interesting.
Instead of positioning robots purely as expensive industrial machines, the company is betting on a future where robotics becomes more like consumer electronics.
There is also a second strategic advantage.
If thousands of affordable robots are deployed in homes and businesses, those machines could generate enormous amounts of real-world data.
In other words, selling the robot could also help improve the robotics model behind the robot.
7. Cosmic Robotics: Building for Earth Before Mars
Cosmic Robotics has one of the more ambitious long-term visions in the YC batch: helping build a city on Mars.
But the company isn’t waiting for Mars to become a commercial market.
Its autonomous heavy-lift robots are already being used for infrastructure work on Earth, including solar installations. YC says Cosmic-1 has been deployed at large solar projects and that the company is expanding toward data-center construction and lunar robotics.
This is an interesting startup strategy.
Rather than building technology exclusively for a future space economy, Cosmic is finding a terrestrial market where the same technology can generate revenue, deployment experience and data.
That creates a useful principle for deep-tech startups:
Build for the future, but find a customer today.
8. Parasma: Could Biological Cells Become Computers?
Perhaps the most unconventional company on the list is Parasma.
Instead of trying to make silicon computers more brain-like, Parasma is investigating whether biological brain cells themselves can become a computing substrate.
The company says it is training human brain cells for computation and has demonstrated biological systems performing tasks including gameplay and token prediction. (Y Combinator)
It sounds futuristic because it is.
But the underlying question is serious:
Can biological systems provide a more energy-efficient way to perform certain types of computation?
Modern AI systems require enormous amounts of energy and computing infrastructure. Biological brains operate using dramatically less energy while remaining extraordinarily capable at learning and adapting.
Parasma is still an extremely early-stage experiment, but its work represents a broader trend: the search for alternatives to conventional computing architectures.
9. Waddle Labs: Giving Robots a Natural-Language Interface
The final company on our list is attacking robotics from the software side.
Waddle Labs is developing an API that allows developers to give robots instructions using natural language. Its AI agents then generate the code required to execute the task and can verify and adjust their work.
The company’s pitch is essentially:
What if programming a robot could feel more like prompting an AI coding assistant?
That could dramatically lower the barrier to robotics development.
Today, making a robot perform a new task can require specialized robotics engineers, control systems and extensive programming.
If general-purpose AI agents can translate human instructions into executable robot behavior, robotics development could become much more accessible.
And that could be one of the missing pieces required for robots to move from specialized machines to general-purpose platforms.
What These Nine Startups Tell Us About the Next Technology Wave
The most interesting part of this YC batch isn’t any individual company.
It is the pattern.
Across these startups, five themes stand out.
1. AI is moving into the physical world
Robotics, autonomous construction, industrial automation and defense systems are becoming increasingly connected to AI.
The next generation of AI startups may therefore look less like traditional SaaS companies and more like hardware + software businesses.
2. Compute infrastructure is becoming a startup opportunity
Power generation, cooling, networking and specialized chips are becoming strategic constraints for AI.
Atomarine, Dipole Labs and Lamb Labs are attacking three different pieces of that problem.
3. Robotics needs a new data layer
Robots need physical-world training data.
Companies like Praxis are attempting to create that infrastructure, while companies such as Nori can potentially generate data through real-world deployments.
4. AI efficiency matters as much as AI capability
The industry spent years asking:
Can we make AI models bigger?
The next question may increasingly be:
Can we make AI dramatically more efficient?
That explains the interest in optical computing, specialized silicon and biological computing.
5. Deep tech is becoming more commercially ambitious
These aren’t simply research projects.
Several companies in this batch are already talking about revenue, contracts, deployments or commercial partnerships.
That matters because deep tech has historically faced a difficult gap between scientific breakthrough and commercial scale.
The new generation of founders appears increasingly focused on shortening that gap.
Why YC Summer 2026 Matters for Startup Investors
Y Combinator Demo Day is often treated as a showcase of individual startups.
But it can also be viewed as an early indicator of where venture capital is moving.
The Summer 2026 batch suggests that investors are looking beyond another layer of AI applications.
The bigger opportunities may be hiding underneath the AI economy:
- Energy
- Semiconductors
- Data-center infrastructure
- Optical networking
- Robotics
- Physical-world data
- Defense manufacturing
- Autonomous construction
- Alternative computing architectures
That doesn’t mean every company in these categories will succeed. Deep-tech startups face enormous technical, regulatory and capital requirements.
But the direction is becoming clearer.
AI is no longer just a software story.
It is increasingly an infrastructure story, a hardware story and a physical-world story.
And if this YC batch is any indication, some of the most interesting startups of the next decade may be building things that would have sounded almost impossible only a few years ago.
Frequently Asked Questions
What is Y Combinator Demo Day?
Y Combinator Demo Day is an event where startups from the accelerator’s latest batch present their businesses to an invite-only audience of investors and media. YC says its Demo Day audience is approximately 1,500 investors and media members.
Which YC Summer 2026 startups are attracting investor attention?
Among the startups highlighted following the latest Demo Day are Atomarine, Dipole Labs, Isengard Industries, Lamb Labs, Praxis, Nori, Cosmic Robotics, Parasma and Waddle Labs. The companies span AI infrastructure, robotics, semiconductors, energy, defense and biological computing.
Why are robotics startups becoming important?
Robotics is benefiting from advances in AI models, computer vision, simulation and hardware. Startups are now attacking different parts of the robotics stack, from training data and robot control software to affordable physical machines and autonomous construction.
Is AI moving beyond software?
Yes. The current startup landscape increasingly includes AI infrastructure, robotics, semiconductors, energy systems and other technologies that connect AI to the physical world. YC’s Summer 2026 company directory reflects this broadening of AI into areas such as robotics, hard tech and semiconductors.
What should founders learn from the latest YC batch?
One clear lesson is that ambitious technology can still be commercially grounded. The strongest deep-tech opportunities often combine a long-term technological vision with a near-term customer problem that can generate revenue, data or deployment experience.
Sources & further reading: Y Combinator company profiles and Demo Day information; TechCrunch’s September 13, 2026 reporting on the YC Summer 2026 batch. (TechCrunch)
Internal links for Query Feedz: This article can naturally link to Query Feedz’s existing coverage of [AI and technology], [startups and funding], and its recent article on ten startups that raised $1.47 billion in one day, as well as the article on OpenAI’s Astra model. (Query Feedz)
