What Y Combinator wants to fund now, and what it tells the Brazilian founder

Y Combinator has published the Fall 2026 edition of its Request for Startups (RFS), the public list of ideas that the fund's partners would like to see founders tackle. The document itself makes a point of warning that these ideas represent "just a fraction" of what YC funds, and that no one needs to work on them to apply. Even so, treating the RFS as mere curiosity underestimates the signal. When the organization that has already put capital into Stripe, Coinbase, Airbnb and Deel decides to write down, with each partner's full name attached, where it thinks the next return cycle lies, it is drawing a capital-allocation map, and the money that follows YC (including Brazilian money) tends to move in the same direction a few quarters later.

The throughline of this edition is explicit in the opening text: "AI is moving into the physical world". The thesis is no longer the chatbot on the screen. It's AI leaving the browser and moving into education, healthcare, defense, infrastructure, logistics and the factory floor itself. It's worth reading each request not for the idea itself, but for the business bet hidden behind it.

The categories, and the bet behind each one

Some entries in the RFS are more revealing than they seem at first glance:

• The Primer (Andrew Miklas): an AI tutor that teaches children to read, write and do math "at the quality of a dedicated private tutor and at consumer scale." The bet is that personalized education, historically an expensive privilege, becomes an off-the-shelf product. The business angle: it starts as something a parent buys, but it's an entry point for bigger ambitions.
• A Cloud for Small Software (Pete Koomen): infrastructure for "small software," tools built by agents for one or a few users. The direct thesis is that AWS and Azure were designed for "Big Software" and carry too much complexity for this new use case. Here YC is pointing to a market gap below the cloud incumbents.
• Multiplayer AI (Aaron Epstein): the argument is historical. Google Docs beat Word and Figma beat Photoshop by being multiplayer, and AI "hasn't had its multiplayer moment yet." The bet is that agent sessions shared by an entire team will replace today's thousand private chat threads.
• AI-Powered Consumer Products for 1 Billion People (Raphael Schaad): this request carries the document's most concrete number. Today "the magic" can cost $1,000 a month in tokens per user, but that cost, according to the text, "is falling 10x per year." The thesis is that this cost decline is what will unlock AI's consumer moment, and whoever builds first owns the category.
• AI for the Aging Population (Max Kolysh): backed by a demographic projection, one in five people in the US will be over 65 by 2030, with millions of caregiving positions unfilled and 53 million family members already providing unpaid care.
• AI-Native Compliance Infrastructure and Self-Maintaining APIs: two B2B infrastructure entries. The first wants to rebuild financial compliance, today "stitched together with spreadsheets," as an AI-native problem. The second proposes agents that don't just announce API changes, but open the pull request with the fix, a "Dependabot for APIs."

There are also more exotic bets, such as Compute at Sea (data centers floating on the ocean to escape the energy and land-based licensing bottleneck) and The Future of American Defense, written, for the first time, by the sitting Secretary of the US Army, promising capital and a proving ground for founders building for ground combat. This entry is the most politically charged of the batch and the least transferable to the Brazilian context.

What's actually useful for those building in Brazil

Here it's necessary to separate what is a universal signal from what is specific to the American market. Defense, US population aging and state-by-state American compliance are agendas that respond to local pain points, and importing them without translation would be exactly the mistake that turns analysis into copying.

Two categories, however, have direct resonance here. The first is crypto as a payment rail. The text The Best Time to Build in Crypto (Nemil Dalal) is emphatic: YC expects that "eventually every YC startup will use crypto rails, from fundraising to payments, even if most never know it." And it names BlindPay and Infinia by name, building the developer interface and the on/off-ramps for Latin America, alongside Aspora for remittances to India. In other words: the RFS itself acknowledges that stablecoins and remittances make more sense where banking infrastructure is expensive and fragmented, exactly the Latin American case. For a Brazilian fintech founder, this is the single most actionable line in the whole document.

The second is the token-cost-decline thesis. If the "10x per year" figure cited in Consumer Products holds, consumer-facing AI products that would be unviable today because of inference cost become viable within a one-to-two-year horizon. That's roadmap planning, not philosophy: you can design a product aiming at tomorrow's cost, not today's.

The counterpoint the founder needs to make

Before treating the RFS as prophecy, some skepticism is warranted. First, YC has an interest in directing dealflow: the more founders attacking the themes it has already decided to fund, the cheaper and more efficient its funnel becomes. The document is, in part, acquisition marketing. Second, the RFS gets things wrong. Lists of "ideas we'd like to see" are notoriously bad at predicting which categories actually produce outliers, because venture returns come precisely from what nobody asked for. YC itself admits these ideas are a fraction of what it funds, which is an elegant way of saying that the best businesses usually come from outside the list.

Third, and most important for Brazil: Brazilian capital that "follows YC" follows with a lag and a context discount. An American defense thesis or an ocean data-center thesis doesn't have an equivalent exit thesis here. Copying the category without the corresponding market, regulator and buyer is a recipe for burning runway.

The takeaway for decision-makers

The value of the Fall 2026 RFS isn't in serving as a checklist of ideas to clone. It's in revealing the direction of conviction held by the world's largest early-stage allocator: vertical AI, applied to the physical world and to large, under-digitized industries, no longer the generic LLM wrapper. For the Brazilian founder deciding where to point the technical roadmap or preparing the next pitch, the useful question isn't "which of these ideas do I build." It's "which of these theses has a real analog in the Brazilian market, with local pain, a local buyer and a local payment rail." Crypto and stablecoins for Latin America, with two examples cited in the document itself, is the most obvious candidate. Everything else requires translation, and poorly done translation costs dearly when it's time to raise money.