14 Billion dollar AI Ideas YC Is Betting On in 2026

Added: 2026-05-07

[00:00:00]Every 3 months, the most powerful startup accelerator in the world publishes a document that tells you exactly which billiondoll companies they want to fund. Not vaguely, not we like AI or we're excited about health tech specifically with the reasoning behind each idea written by the partners themselves. The document is called the request for startups, the [music] RFS, and almost nobody talks about it outside of Silicon Valley startup circles. Here is what makes this interesting for you specifically. Y Combinator has funded Airbnb, Stripe, Dropbox, Reddit, [music] Coinbase, OpenAI, Door Dash. Their portfolio is worth more than $600 billion combined. When the people who backed those companies publish a document saying we want [music] someone to build this, that is not an opinion.

[00:00:37]That is a directional signal backed by the most successful pattern recognition engine in the history of technology entrepreneurship. And the summer 2026 RFS dropped a few weeks ago. 14 specific categories, 14 areas where YC's partners believe the next generation of defining companies will be built. So today I want to do two things. First, decode what YC actually wants [music] and why. Second, and this is the part most people skip entirely, talk about which of these ideas are genuinely buildable right now with or without being in Silicon Valley.

[00:01:05]Because the most important thing about the RFS is not that it tells you what to build. It is that it tells you what the smartest investors in the world currently believe the market is missing.

[00:01:12]Let's decode. You already know what YC is. Airbnb, Stripe, Dropbox, Reddit, Coinbase, OpenAI, Door Dash. Over 5,600 companies funded. portfolio valued at over $600 billion. So when they publish a document saying here is exactly what we want founders to build, the correct response is to read it very carefully.

[00:01:30]Now let me walk through what is actually on the summer 2026 RFS. I am going [music] to group these into what I think are the three clearest themes because reading them in sequence you start to see YC's worldview very clearly. Theme one AI is becoming the company not the feature. The first category is AI native service companies. And this is the one I want you to pay [music] close attention to because it is the most immediately actionable. YC partner Gustaf Ulstr writes, "Historically services became SAS software. Then they became AI co-pilots. [music] What we are excited about now is the next step. AI native companies that do not sell software.

[00:02:03]They sell the service." Think about what that means. Insurance [music] brokerage, accounting, tax and audit, compliance, healthcare administration. Every one of these is a massive industry that runs primarily on human labor. has already been partially digitized [music] through SAS tools and is now ready for a model where you do not buy software to do the work. You just buy the outcome, the work gets done. You do not know or care how the total spend on services is many times [music] larger than the spend on software outsourced which means the switching cost is lower. You are not replacing a product someone loves.

[00:02:35][music] You are replacing a vendor relationship someone would happily change if the new vendor was better and cheaper. The second category in this theme is the company brain. Tom Blumfield, founder of Monzo, writes about this one. He argues [music] that the biggest blocker to AI automation in companies is no longer the models. It is domain knowledge. [music] Every company has critical knowledge scattered across emails, Slack threads, people's heads, old support tickets. AI cannot operate on that. What every company needs is a living map of how it actually works. So AI agents can execute reliably and consistently [music] rather than guessing. The third is the AI operating system for companies, which is related but different. If the company brain is about capturing institutional knowledge, the AIOS is about making the entire company queriable and creating a closed feedback loop between decisions and outcomes. And the fourth in this theme is dynamic software interfaces. The idea that software interfaces of the future will be unique to each user. Your email client looks like a task list. My email client looks like a calendar. The underlying software is the same, but the interface is personalized down to the individual user level by AI agents.

[00:03:38][music] The next wave of software companies will be built by people who understand that AI can now customize the experience, [music] execute the task and hold the context, not just assist with it. Team two, physical infrastructure and hardware have been neglected for a decade. That is ending. The second cluster of RFS categories is about the physical world and it is the cluster that I think most people in India completely underestimate. The [music] first is AI for low pesticide agriculture. Gary Tan YC CEO writes this one personally. His argument is that the combination of AI vision, cheap sensors, precision robotics, [music] and biological alternatives to synthetic chemicals has created a genuine technological moment. A company that can cut pesticide use by 90% while helping farmers grow more food is, [music] in his words, a generational company.

[00:04:24]Agriculture is one of the biggest markets in the world. The green revolution happened in India in the 1960s [music] and saved hundreds of millions of people from starvation. It was built on exactly the same logic. New technology applied to an ancient problem at scale. Gary Tan is saying the third green revolution is now [music] possible and AI is the enabling technology. The second physical category is the hardware supply chain. Building hardware in the US takes weeks where building hardware in Shenzhen [music] takes a day. The same iteration speed gap that software companies solve with cloud infrastructure has not been solved for hardware. YC wants companies that can compress the hardware iteration [music] cycle dramatically. The third is counterwarm defense. A drone costs $500.

[00:04:59]A Patriot missile cost $3 million. One cheap Iranian drone swarm recently took out an AWS data center and nothing stopped it. YC partner Tyler Bosman argues that drone defense is no longer a weapons problem. It is a real-time distributed systems problem. The winning company will look more like Cloudflare than Rathon. [music] And then there are two space categories. Electronics and inference chips for space and industrial capabilities on the [music] moon. These are longer time horizon plays, but they are on the list because reusable rockets from SpaceX have so dramatically lowered the cost of putting things in space that infrastructure that was economically unthinkable [music] 5 years ago is now approaching viability. What is the thread running through all of these? YC is saying that the AI revolution is not just a software revolution. It is about to go physical. The companies that apply AIdriven precision and intelligence to agriculture, defense, hardware, manufacturing, and eventually space [music] are the next generation of trillion dollar businesses. Let me take a break here. I will come right back with the most directly actionable part of this theme. Three, the infrastructure layer for AI agents does not yet exist.

[00:05:57]This is the one that is hardest to explain but might be the biggest opportunity on the entire list. See, in 2023 and 2024, the dominant framing for AI startups was build an agent that does something useful. Browse the web, write code, answer customer service questions.

[00:06:10]And thousands of companies built exactly that. But there is a problem. These AI agents are running on top of software that was designed for humans clicking buttons in a browser which is slow, inconsistent, and brittle. It is like trying to run a delivery service when all your drivers are using maps designed for pedestrians. YC partner Aaron Epstein writes, "The next trillion users on the internet will not be people, they will be AI agents. And we need to make something agents want. That means every major category of software needs to be rebuilt with machine readable interfaces, APIs, MCPs, CLIs, documentation that allows agents to discover, sign up for, and use new tools without a human in the loop. Every form, every button, every dashboard that currently assumes a human is using it needs an agent native version. The second category in this theme is inference chips for agent workflows.

[00:06:54]Current GPUs hit 30 to 40% of peak utilization on agentic workloads because those workloads are bursty and bounce between memory bound model calls Ibound tool use and [music] CPUbound orchestration. Nobody has built silicon optimized for the way agents actually execute. Nvidia bought Grock for $20 billion because they saw this gap. YC wants someone to close it. The third is SAS challengers and this is the most classically counterintuitive one on the list. AI has collapsed the cost of producing software by 10 to 100 times.

[00:07:21]The moat that protected legacy SAS companies, millions of lines of code built over decades is gone. A fivep person team can now go after chip design software, ERPs, [music] industrial control systems, supply chain management, the things that look untouchable for decades. YC partner Jared Freriedman explicitly says, "Do not start with project management tools.

[00:07:40][music] Go after the products that seem invulnerable." And the fourth is supply chain 2 for semiconductors. A single advanced AI chip goes through 1,400 process steps, crosses a dozen countries, and takes 5 months to build.

[00:07:51]This supply chain is currently managed with spreadsheets, SAP, and phone [music] calls, real-time allocation tracking, multi-ter risk monitoring, export compliance. Almost none of this tooling exists today. The through line across all four, YC believes the AI application layer is mostly being built.

[00:08:08]The infrastructure that AI applications run on is not. Now, here is the catch, and this is the most important thing I'm going to say. The RFS is brilliant at telling you where opportunities exist.

[00:08:18]It is not a guarantee that building in these areas will make you successful.

[00:08:21]And there is a very specific trap that many founders fall into when they read documents like this. The trap is that the RFS describes outcome level categories. Build an AI native service company in [music] insurance. Build the company brain for every business. These are directions, not product [music] specs. The founders who succeed in these categories are not the ones who read the RFS and say, "Great, I will build an AI insurance broker." They are the ones who spend 6 months inside a specific industry talking to customers before writing a single line of code. Here's the thing about these categories that most people [music] miss. AI native service companies in accounting, tax, compliance, and healthcare administration. The person who builds the winning product here is not a generalist AI engineer who thought this sounds like a big market. It is someone who spent 5 years inside a compliance firm watching the exact steps that a human does every day and knows precisely where the errors [music] happen, where the delays happen and what the client actually cares about. Domain expertise is the moat, not the model. AI for agriculture. The problems Gary Tan describes in his RFS entry. [music] Farmers stuck in a loop of increasing chemical use with diminishing returns are not abstract problem. They are the daily reality for farmers in the US Midwest, in Brazil, in Southeast Asia, in subsaharan Africa. The founder with genuine on the ground context in a specific agricultural region has a structural advantage over anyone building this in a vacuum. The technology is now accessible. [music] The insight has to come from proximity to the problem. The chips act is standing up new American fabs in Arizona, Texas, Ohio, and New York. Each one needs a supply chain [music] nearly from scratch. The person who understands both semiconductor manufacturing processes and modern [music] supply chain software deeply enough to build the tooling is almost certainly someone who has lived inside one of these systems, not someone who read a market report. and the SAS [music] challengers category. YC says go after the products that seem invulnerable. The ERPs, the industrial control systems, [music] the people with the deepest institutional knowledge of what is broken inside these systems are the engineers [music] who spent years at the companies that built them or the enterprises that bought them. Not the person with the freshest AI skills, the person with the most specific frustration. [music] So what do you actually do with all of this? Here's the honest answer. If you are a founder deciding what to work on, [music] the RFS gives you a reality check. Is your idea in a category where smart well-resourced people see a genuine gap or are you in a category that is already crowded with wellunded competitors [music] and no clear reason why your approach wins? If you are an investor in India trying to understand which sectors will attract global capital over the next 5 years, the RFS [music] is a better signal than most market reports because it is written by people who are actively deploying money, not just analyzing [music] trends. And unlike most documents that tell you the future is exciting, this one tells you specifically what is [music] missing, why it is missing, and what kind of company would exist if someone fixed it.

[00:11:00]That is a treasure map. Most people just forget to follow it. I want to ask you a very specific question in the comments.

[00:11:06][music] Which of these 14 categories do you think is the most underrated, not the most hype? The one where you think the actual gap between what exists today [music] and what needs to exist is the largest and where the winning company has not been built yet. I want to know how you are thinking about it. Drop your answer below. I will [music] see you in the next one. Bye-bye.