Where the Smart Money Is Going: Y Combinator Names 15 AI Categories Worth Billions

Y Combinator has funded OpenAI, Airbnb, Stripe, DoorDash, and Coinbase. When it publishes a list of what it wants to back, sophisticated investors pay attention. The Summer 2026 Requests for Startups is the most expansive the organization has produced in years, and its contents reveal a clear shift in where YC believes the largest AI opportunities now sit.

The through-line across all 15 categories is this: AI has moved past being an add-on to existing software. YC is now looking for companies where AI is the core operating layer, and in many cases, the product itself. Here is a full breakdown of each category, what problem it addresses, and why it matters for enterprise decision-makers and investors watching this space.

01 AI for Low-Pesticide Agriculture

Written by YC CEO Garry Tan personally, this category starts with a problem that has no easy solution under the current agricultural model. Farmers are caught in a loop: pests and weeds adapt to chemicals, so farmers spray more, costs rise, margins fall, and the pipeline for new synthetic solutions is slower and more expensive than it has ever been.

What changed is a combination of factors arriving at once. AI can now identify individual weeds and pests in real time. Cameras and sensors are cheap enough to deploy across large areas. Robotics can treat a single plant rather than blanketing an entire field. And breakthroughs in biology, including microbes, RNA-based solutions, and engineered plants, are making it possible to replace entire classes of synthetic chemicals. Tan’s view is that a company capable of cutting pesticide use by 90% while simultaneously improving yields would be a generational business. Agriculture is one of the largest markets in the world, and adoption of a working solution would move fast.

02 AI-Native Service Companies

This is arguably the most consequential category for enterprise investors to understand. YC partner Gustaf Alströmer draws a clear line between what AI startups built from 2023 to 2025 and what comes next. Most of that period produced tools that helped people do their existing jobs faster. The next step is companies that skip the human entirely and just do the work.

The distinction matters financially. Global spending on services is many times larger than spending on software. And because a large portion of services are already outsourced, replacing them with an AI-native operation is structurally more straightforward than displacing software. YC is particularly focused on insurance brokerage, accounting, tax and audit, compliance, and healthcare administration. These are large, mature markets where the work is well-defined and currently performed at considerable cost by human labor.

03 AI Personalized Medicine

Two independent cost curves are collapsing simultaneously, and YC believes their convergence creates a major opening. The cost of generating personalized diagnostics is falling rapidly. Genome sequencing has dropped faster than Moore’s Law for years, and new diagnostic tools are entering the market that can detect health conditions far earlier than was previously possible. At the same time, the cost of producing personalized genetic therapies is also falling. Delivery technologies like mRNA now make treatments designed for an individual patient technically feasible, and the FDA has expressed increasing openness to allowing patients access to these approaches.

YC sees AI as the layer that connects these capabilities, analyzing genome scans, electronic health records, wearable data, and diagnostic results to produce accurate, patient-specific recommendations. The ambition is a shift away from population-level medicine toward individualized care at scale.

04 Company Brain

Tom Blomfield, founder of Monzo and a YC partner, identifies the central problem holding back AI adoption inside companies: the models are no longer the bottleneck. The bottleneck is institutional knowledge that exists nowhere in a structured form.

Every organization has critical know-how scattered across people’s heads, email threads, Slack channels, support tickets, and shared drives. The company functions because experienced employees remember, roughly, where that knowledge lives. AI cannot operate this way. Blomfield wants a system that extracts this knowledge from its fragmented sources, organizes it, keeps it current, and turns it into something AI can actually use to take action reliably. This is not a document search tool or a chatbot layered over files. It is a living operational map of how a company actually works, from how refunds are handled to how engineers respond to incidents. His view is that every company in the world will need one.

05 Counter-Swarm Defense

YC partner Tyler Bosmeny opens this category with a concrete event: a swarm of cheap drones recently took out an AWS data center, and no one stopped them. The economics of drone warfare have turned decisively against defenders. A Patriot missile costs three million dollars. An FPV attack drone costs approximately five hundred. All the cost leverage sits with the attacker.

Current counter-drone infrastructure was built for individual threats. It will not scale to coordinated swarms that are cheap, autonomous, and jam-resistant. Bosmeny is looking for founders building across the full stack: high-capacity interceptors that can neutralize many drones at once, software that fuses data from every sensor into a single real-time picture, non-kinetic defenses, and attacks on the autonomous software that controls the swarms. His framing is that winning in this space will look less like traditional defense procurement and more like operating a distributed software system at scale.

06 Dynamic Software Interfaces

Every user of a software product today sees essentially the same interface, with minor cosmetic differences. YC partner Ankit Gupta believes this is about to change. AI coding systems have reached a point where individual users could, in theory, become their own customization engineers, radically reshaping the software they use for their specific needs rather than accepting a one-size-fits-all design.

Gupta imagines an email client that looks like a task list for one user and an events calendar for another, both built on the same underlying infrastructure. The opportunity he sees is in the software layer that makes this possible: how developers ship flexible components that users can modify deeply, and what the entire stack of software delivery looks like when the end interface is no longer fixed.

07 Electronics in Space

Reusable rockets from SpaceX and Stoke Space are about to increase humanity’s capacity to put things in orbit by a significant order of magnitude. YC’s view, written by Philip Johnston, is that the immediate bottleneck will be compute. Specifically, the market for inference chips designed for use in space, optimized for weight, heat management, and radiation tolerance, is going to be very large, very soon, and currently has almost no supply.

This is one of the most specifically targeted categories in YC’s history. The organization is explicitly calling out engineers currently working in chip design at SpaceX or NVIDIA as the target audience, people who already understand both the technical constraints and the market that is forming.

08 Hardware Supply Chain

YC partner Nicolas Dessaigne frames this as an iteration speed problem. In Shenzhen, a hardware team can go from a design file to a physical part in one day. In the United States, the same process routinely takes weeks. That gap compounds across every development cycle and puts American hardware companies at a structural disadvantage.

YC is funding more hardware companies than ever, covering medical devices, home robotics, and space hardware. But the supporting infrastructure for fast hardware development in the US barely exists. The accelerator wants startups that can produce parts dramatically faster, enable rapid iteration cycles, and tightly integrate design, manufacturing, and logistics into something closer to what exists today in Shenzhen.

09 Industrial Capabilities in Space

Written by Adi Oltean, this is the most long-horizon category on the list. The moon contains silicon, aluminum, iron, and titanium. Electrolysis works more efficiently in low gravity. Three-dimensional printing from molten lunar material is more practical there than on Earth because structures do not require the same physical supports. YC wants founders working on building the industrial base that would make sustained human presence in space economically self-sustaining rather than dependent on constant resupply from Earth.

10 Inference Chips for Agent Workflows

Most AI chips today are designed around a simple model: a prompt goes in, a response comes out. AI agents work differently. They loop continuously, calling tools, branching based on results, backtracking, and holding context across many steps. This creates a fundamentally different demand on hardware.

YC partner Diana Hu notes that current GPU architectures hit only 30 to 40 percent of peak utilization on these workloads because the processing is bursty and uneven. Nobody has yet designed chips specifically for the agent execution pattern: fast switching between models, memory optimized for the context that agents carry across long tasks, and architecture that treats the full execution loop as the unit of work rather than a single inference call. NVIDIA’s $20 billion acquisition of Groq reflects how seriously the market is taking this gap.

11 SaaS Challengers

YC partner Jared Friedman makes a direct case: the vulnerabilities that investors have been pricing into legacy software company valuations are real, and that creates an opening. AI has cut the cost of producing software by a factor of ten to one hundred. The complexity that once protected large SaaS businesses, millions of lines of code built over decades, no longer constitutes a defensible barrier the way it did.

Friedman’s advice is not to start with easy targets. He wants founders going after products that currently appear untouchable: chip design software, enterprise resource planning systems, industrial control software, and supply chain management tools. The comparison he draws is to the previous generation of great software companies, which were built by replacing on-premise software with cloud delivery. The next generation will be built by replacing cloud SaaS with software that is AI-native from the ground up.

12 Software for Agents

YC partner Aaron Epstein makes an observation that carries significant implications for enterprise software buyers: the next trillion users of the internet will not be people. They will be AI agents. And almost all of the software these agents currently interact with was designed for humans using keyboards and mice.

This creates a practical problem. Agents browsing websites, managing CRM systems, making purchases, and conducting research are doing so on interfaces built around human behavior. The result is slow, inconsistent, and brittle. Epstein wants startups building software designed from the start for machine consumption, interfaces, APIs, and data structures that work with how agents actually operate.

13 Selling to Huge Companies

YC partners Harshita Arora and Brad Flora present a shift in enterprise sales dynamics that deserves attention from large organization procurement teams as much as from investors. Their observation is that a two or three person startup can now, for the first time, build something that a Fortune 10 company finds genuinely useful before the startup has been operating for a year.

This was not true before. Enterprise-grade software required years of development, significant teams, and deep integration work. AI has compressed that timeline substantially. Arora and Flora note that YC companies are already landing pilots and multi-million dollar contracts within their first year, and that large enterprise buyers are moving faster than they historically have because the problems being solved are real and the solutions are working.

14 Semiconductor Supply Chain

Also written by Diana Hu, this category starts with a specific fact: a single advanced AI chip goes through approximately 1,400 manufacturing steps, crosses a dozen countries, and takes five months to produce. The supply chain that supports this process is managed primarily through spreadsheets, legacy ERP software, and phone calls.

In 2021, a $300 chip shortage halted the production of $210 billion worth of vehicles. The structural problem has gotten more acute since then. TSMC’s advanced packaging capacity is currently the single largest bottleneck in AI compute production. NVIDIA has locked up over 60 percent of it. Export controls change on a quarterly basis. New American fabrication plants are being built in Arizona, Texas, Ohio, and New York under the CHIPS Act, each requiring supply chains to be constructed nearly from scratch. The tooling that one would expect to exist for managing this complexity largely does not.

15 AI Operating System for Companies

The final category, also attributed to Diana Hu, addresses what happens at the company level as AI agents begin handling more internal work. The best AI-native companies, in Hu’s framing, have made their entire organization legible to an AI layer. Every meeting is recorded. Every support ticket is tracked. Every customer interaction is captured and connected to an intelligence system that can learn from it and act on it.

This transforms a company from a collection of disconnected processes into something that can operate with genuine continuity and institutional memory. The companies that have built versions of this internally have cut development time and increased output substantially. The product that enables any organization to build this layer does not yet exist as a commercial offering. That gap is what Hu wants founders to close.

Taken together, the Summer 2026 list represents a meaningful departure from YC’s previous priorities. Earlier editions focused on applying AI to familiar business workflows. This one covers chips, defense systems, space manufacturing, agricultural robotics, and the infrastructure that AI itself runs on. Eight of the fifteen categories require significant capital, specialized hardware expertise, or both. The remaining seven are software opportunities, but they are deeper and more structurally ambitious than anything the previous batch identified.

For investors, the practical reading is straightforward. YC is placing its reputation on the view that the AI opportunity is no longer primarily about productivity software. The next phase is about rebuilding services, hardware, and physical systems with AI as the foundation. The companies that get there first, in categories that are hard to enter, will be the ones that look obvious in retrospect.