Y Combinator updates its RFS list and points AI outside the screen

Editor's note: the final paragraph of this article states that the AI-driven financial compliance and physical data/robotics theses, as well as the crypto thesis, have 'portfolio examples operating' in Latin America. This does not hold up: the source (Y Combinator's RFS page) only cites regional examples (BlindPay, Infinia, Aspora) for the crypto thesis, as the article itself correctly acknowledges in the previous section. The compliance and physical data/robotics theses do not mention any company operating in the region. We kept the original text and are flagging the inaccuracy here so readers don't take the final sentence as equivalent to the earlier fact-check.

Y Combinator has maintained, for years, a public list called Request for Startups (RFS): a set of theses written by the accelerator's partners and by portfolio founders, explicitly stating which category they want to put money into for the next batch. The page gathers the full history, from Summer 2024 to the most recent batch, Fall 2026, and it's precisely this update that marks the relevant turning point: for the first time in YC's history, one of the theses was written by a sitting member of the American government, the U.S. Secretary of the Army, Daniel P. Driscoll.

This isn't just a curiosity. It's an indicator that the accelerator is willing to publicly formalize its bet on defense as an investment category as legitimate as SaaS or fintech. The Fall 2026 list sums up the shift in a single title: "AI is moving into the physical world." Of the thirteen theses published in the batch, most move away from pure software and target hardware, robotics, physical data, and infrastructure.

What YC Is Asking for Now

The Fall 2026 list covers a broad spectrum. It's worth looking at the categories side by side to understand where capital is migrating to:

Beyond these, the batch also calls for AI-driven financial compliance infrastructure, physical data collected by robots and sensors, human identity verification against deepfakes, and APIs that self-correct when the provider changes the spec.

The Number Behind the Consumer Bet

Not every RFS thesis is about heavy hardware. The one titled "AI-Powered Consumer Products for 1 Billion People," signed by Raphael Schaad, bets that the next cycle of mass consumer products, on the scale of Instagram or DoorDash, hasn't appeared yet because the cost of running an agent per user is still high, but is falling fast.

Today, the magic can run $1,000 a month in tokens for each user, but that is falling 10x a year. Follow the curve, and you can predict the consumer moment: it lands very soon.

Raphael Schaad, author of the RFS "AI-Powered Consumer Products for 1 Billion People" at Y Combinator

The $1,000/month-per-user figure is cited without an external source, it's the thesis author's own estimate, but it serves as a yardstick for when the business model becomes viable: if the decline curve really is 10x a year, the break-even point for mass consumer products stops being hypothetical within a few cycles.

The Most Concrete Clue for Those Building in Latin America

Among the thirteen theses, the crypto one is the only one that explicitly cites startups operating in the region. Nemil Dalal, author of the RFS "The Best Time to Build in Crypto," argues that a down market is the best time to build, because it pushes out speculators and attracts founders focused on real infrastructure. He cites BlindPay and Infinia as examples of YC portfolio startups building the developer interface and the ramps (conversion between local currency and crypto) for Latin America, and Aspora doing the same for remittances to India.

This is a direct signal: YC already has a declared thesis and capital allocated to crypto infrastructure aimed at the region, even with sector prices down. For whoever decides on product in Brazil, this is more concrete than any American hardware category cited in the list, because it points to exactly the kind of ramp and rail the accelerator is already funding here.

The Counterpoint: An Agenda List Is Not a Market Guarantee

The RFS is written by the people who will decide the checks for the next batch, which means it reflects the appetite of YC's partners, not necessarily proven market demand. The "Compute at Sea" thesis, for example, assumes that moving data centers to the ocean solves the energy and licensing bottleneck, but it doesn't bring a single pilot, a single operating company, or a single cost-per-megawatt figure. It's a conviction bet, not a data extrapolation.

The defense thesis also deserves caution outside the U.S.: the "tear up the old procurement manual" model cited by Driscoll depends on a specific Pentagon reform, with no direct equivalent in Brazil's defense procurement process. Brazilian dual-use startups don't currently find the same venture capital channel willing to fund war hardware that YC is signaling in the U.S.

Still, one of the theses brings a data point concrete enough to validate the pain, even without a formal benchmark: Harsha Gaddipati, in proposing "Self-Maintaining APIs," cites his own experience at AWS.

When I worked at AWS, over 30% of our service downtime was due to external api/package changes going unnoticed.

Harsha Gaddipati, author of the RFS "Self-Maintaining APIs" at Y Combinator

This number has an identifiable source (the author's own professional experience at AWS), even though it's not an independent study, and it describes a problem that any platform team in Brazil also faces: dependency on an external API breaking without warning.

What This Means for Decision-Makers in Brazil

The practical lesson isn't "copy the American category." It's reading the RFS as a map of where top-tier capital will look for founders over the next twelve months, and cross-referencing that with what already has local signal. AI-driven financial compliance, physical-world data collected by sensors and robotics, and crypto rails for remittances and currency conversion are categories that YC has already named with portfolio examples operating in the region.

Defense hardware and maritime compute depend on a regulatory and capital ecosystem that Brazil doesn't yet have. The yardstick for the Brazilian founder isn't asking "is this on YC's list," it's asking "does this same problem, without the American context, still hurt here."

Source

1. Y Combinator, Request for Startups (RFS): https://www.ycombinator.com/rfs