Y Combinator Reveals 15 Startup Ideas It Wants Founders to Build in Summer 2026

Y Combinator has released its Summer 2026 Requests for Startups (RFS). It covers 15 categories spanning AI, hardware, defense, agriculture, and space. The batch runs July through September in San Francisco. YC is investing $500,000 in every accepted company on standard terms.

The opening line of the RFS sets the tone as it says, “AI has stopped being a feature and started being the foundation.” Several ideas came directly from active YC founders describing what they are seeing on the frontier. This is not a list of trends. It is a list of gaps that are now solvable.

Here is a breakdown of all 15 categories. This is what YC wants, why now, and what it looks like in practice.

1. AI-Native Service Companies

Gustaf Alströmer leads this category with a simple economic argument. The total global spend on services dwarfs the spend on software. Most of those services are already outsourced, which makes them structurally easy to replace.

The progression so far has been that services became SaaS, then SaaS became AI copilots. YC now wants the next step. Companies that do the work, not companies that sell software to help people do the work. Instead of giving a client a tool to handle insurance brokerage, you become the broker.

Target sectors include insurance, accounting, tax, audit, compliance, and healthcare administration. A startup called Lexi is already exploring AI-native legal services along these lines. The unit economics are fundamentally different when you own the outcome.

Related Post: Y Combinator-backed Rulebase Secures $2.1M for Agent Coworker Software

2. SaaS Challengers

Jared Friedman makes a direct case that if AI has made legacy software incumbents vulnerable, this is the biggest startup opportunity in a decade. AI has collapsed the cost of producing software by 10 to 100 times. The moat that once protected companies like Salesforce, with millions of lines of code built over decades, is now gone.

YC encourages founders to think beyond easy targets like project management tools. The real prize is chip design software, ERPs, and industrial control systems. These are codebases that have been functionally untouchable for 20 years.

Attack strategies include cloning a product and selling it at one-tenth the price, bundling ten-point solutions into one AI-native suite, or building open-source replacements and monetising through hosting. The previous generation replaced on-premise software with cloud. This generation replaces legacy SaaS with AI-native software.

3. Company Brain

Tom Blomfield, who founded Monzo, identifies what he calls the missing primitive of the AI era. Every company has critical knowledge scattered across Slack threads, old emails, support tickets, and people’s heads. Humans navigate this vaguely. AI agents cannot.

A “company brain” would pull knowledge from all these fragmented sources, structure it, keep it current, and turn it into an executable file for AI agents. The example he gives is how does your company handle refunds? What are the rules for pricing exceptions? How do engineers respond to incidents? All of that should be legible to an AI system that can then act on it safely and consistently. This is a living map of how a company actually operates.

4. Software for Agents

Aaron Epstein opens with a striking framing as he says that the next trillion users on the internet will not be people. They will be AI agents. And almost no software is built for them.

Agents are already browsing the web, doing research, and managing workflows. But they are doing it on top of interfaces designed for humans clicking buttons, which is slow, brittle, and inconsistent. What agents need instead is machine-readable infrastructure. These are  APIs, MCPs (Model Context Protocols), and CLIs. They need thorough documentation so they can discover, sign up for, and start using new tools without a human in the loop.

Every major software category used today, from communication tools, CRMs, and analytics platforms, needs an agent-first rebuild. The opportunity is not to build the next agent. It is building what agents depend on.

Related Post: Avelis Health Secures $500k From Y Combinator to Reduce Patients’ Medical Bills

5. The AI Operating System for Companies

Diana Hu describes what she is seeing in the best AI-native companies, and it is that they have made their entire operation queryable. Every meeting recorded, every ticket tracked, every customer interaction captured and legible to an intelligence layer that learns from it.

This turns a company from an open loop, make a decision, check results weeks later, into a closed loop that monitors, compares, and adjusts in real time. She has seen teams using this approach cut sprint time in half.

The problem is that building it today requires brutal custom integration work. There is no product that connects Slack, Linear, GitHub, Notion, and call recordings into a single intelligence layer. That is the gap.

6. AI-Native Discovery Engines

Scientific discovery has run on the same loop for centuries. They hypothesise, experiment, interpret, repeat. Frontier AI models have now reached PhD-level performance on many scientific reasoning benchmarks. The shift YC is betting on is from AI research assistants to closed discovery loops, systems that propose hypotheses, run experiments, analyse results, and iterate without constant human direction.

This is already beginning in drug discovery, materials science, and protein engineering. Intelligent systems are starting to run full design-make-test-analyse cycles. YC does not want research copilots. It wants companies whose core output is scientific progress.

7. AI Personalized Medicine

Two revolutions are converging. The cost of personalized diagnostics, genome sequencing, wearable health data, EHR records is falling faster than Moore’s Law. At the same time, the cost of manufacturing personalised genetic therapies, delivered via mRNA vectors, is also collapsing. The FDA is showing more openness to patient access to these procedures.

YC partner Ankit Gupta argues that an agent harness can now take a patient’s genome scan and wearable data and generate highly accurate, user-specific treatment recommendations. Five years ago this was science fiction. Today, it is engineering. YC sees an entire ecosystem of startups needed with diagnostic tooling, therapy delivery, and the intelligence layer connecting them.

Related Post: Fearless Fund Makes 7-Figure Investment in Y Combinator-backed Zimi

8. Dynamic Software Interfaces

Every user today sees the same interface. Netflix personalises content but not layout. An email client looks identical whether the user is a retired accountant or a university student. The exception has always been enterprise software, where costly “forward deployed engineers” customise tools per client.

YC’s bet is that coding agents are now capable enough to give every user that same level of customisation. Your email client might look like a task list. A student’s might look like a calendar. Both run on shared underlying primitives that a software team designs and ships. This requires rethinking the entire stack of software delivery, potentially shipping source code instead of packaged binaries.

9. Supply Chain 2.0 for Semiconductors

A single advanced AI chip crosses roughly 1,400 process steps, passes through a dozen countries, and takes five months to manufacture. This supply chain is still managed with spreadsheets, SAP, and phone calls.

In 2021, a $300 chip held up a $50,000 car. $210 billion worth of vehicles were not built. Today, TSMC’s advanced packaging is the single biggest bottleneck in AI compute, with NVIDIA having locked up over 60 percent of it. HBM memory is booked through 2026. Export controls change quarterly.

The CHIPS Act is simultaneously standing up entirely new American fabs in Arizona, Texas, Ohio, and New York.  Each needs a supply chain built from scratch. Almost none of the expected tooling exists. There is a need for real-time allocation tracking, multi-tier risk monitoring, and export compliance automation. This requires deep domain knowledge of wafer allocation, which is exactly why it is a startup opportunity rather than a feature inside an existing ERP.

10. Counter-Swarm Defense

Tyler Bosmeny opens with a sobering fact that a swarm of cheap drones recently took out an AWS data centre. Sadly, nothing stopped them.

A Patriot missile costs $3 million. An FPV drone costs $500. All the cost advantage sits with the attacker. The next threat is not a single drone but coordinated autonomous swarms of thousands of cheap, jam-resistant, and lethal drones. Current counter-drone systems are a fragmented collection of radars, cameras, jammers, and interceptors that do not talk to each other.

YC wants the counter-swarm stack. These need to be high-capacity interceptors that neutralise fifty drones at once, software fusing every sensor on a site into a single real-time picture. And non-kinetic defences like aerosols that foul rotors or systems that attack the autonomy stack directly. The core insight, Bosmeny notes, is that drone defence now looks less like operating a weapon and more like running a real-time distributed system.

Related Post: Y Combinator-Backed DryMerge Raises $2.2 Million

11. Inference Chips for Agent Workflows

Most AI chips were designed for one scenario, which is prompt in, response out. However, agents work differently. They loop, calling tools, branching, backtracking, holding context across dozens of steps. That is a fundamentally different hardware problem.

Current GPUs achieve only 30 to 40 percent of peak utilisation on agent workloads because the work is bursty, bouncing between memory-bound model calls, I/O-bound tool use, and CPU-bound orchestration. NVIDIA acquired Groq for $20 billion because it saw this gap coming. Google built TPU v7 specifically for inference. But nobody has yet designed silicon for the agent loop itself with fast context switching between models, native speculative decoding, and memory built for KV caches that persist across an entire execution graph. Diana Hu notes that Groq’s real insight was its compiler, not its chip, and that will likely be true for whoever builds this next.

12. AI for Low-Pesticide Agriculture

YC CEO Garry Tan leads this category. The problem is a structural loop. More chemicals lead to adapting pests, which lead to more chemicals, higher costs, and shrinking margins. The pipeline for new chemical solutions is slower and more expensive than ever.

Three simultaneous changes have broken the deadlock. AI can now identify individual weeds and pests in real time. Sensors and cameras are cheap enough to deploy across entire fields. Robotics can treat a single plant rather than blanket an entire field with a  spray. Biology is also catching up, as microbes, peptides, and RNA-based solutions can now replace entire classes of synthetic chemicals.

The prize is a company that cuts pesticide use by 90% while increasing yields. Agriculture is one of the largest markets in the world. Adoption at that return on investment would be explosive.

13. Hardware Supply Chain

YC is funding more hardware companies than at any previous point in its history, from medical devices, home robots, and space systems. But American hardware iteration is still far too slow compared to China.

In Shenzhen, a team can go from a design file to a finished physical part in a day. In the US, that loop routinely takes weeks. That gap compounds every sprint cycle. YC is already backing early examples, such as Hlabs (W26), which is building actuators, and Prototyping.io (P26), which turns designs into mechanical parts in days. But the overall stack of dense supplier networks, rapid turnaround, and tight integration of design and manufacturing barely exists in the US. The next generation of great hardware companies will be built on much faster iteration loops.

14. Electronics in Space

Reusable rockets from SpaceX and Stoke Space are about to dramatically increase humanity’s capacity to put objects in orbit. That means an enormous new demand for computing capacity in space.

YC specifically wants inference chips built for the orbital environment. These will be slightly optimised for mass, slightly optimised for thermal management, and slightly hardened against radiation. Not radically exotic, but they should be tuned for the context. This RFS is addressed, almost explicitly, to chip designers currently working at SpaceX or NVIDIA.

Related Post: YC-backed Harper Raises $47M to Build AI Insurance Brokerage

15. Industrial Capabilities in Space

The longest time horizon on the list. YC partner Adi Oltean wants to see founders building the industrial base for long-term space presence. This could be extracting silicon, aluminium, iron, and titanium from the moon through electrolysis, and 3D printing complex structures from molten lunar regolith. Low gravity and the absence of an atmosphere make some of these processes more efficient than their Earth equivalents. This is a decade-long bet, aimed at founders and investors thinking across that horizon. If you are working on something like that in the space, then Y Combinator would like to hear from you.

What All 15 Categories Have in Common

Reading the full list, three clear themes emerge.

First, AI is now infrastructure, not a feature. Every category assumes AI as the operating layer, not the product. Second, YC is pushing AI into the physical world; agriculture, hardware, space, and defence are all on the list alongside software. Third, the mandate throughout is to replace rather than assist. Sell the service. Do the work. Build what agents depend on.

YC is also explicit that this list is not exhaustive, as they fund far more companies outside the RFS than inside it. This is a signal about where the most patient capital in Silicon Valley believes the next generation of important companies will come from.

Apply to YC Summer 2026 (late applications still accepted): ycombinator.com/apply