YC Requests for Startups, Winter 2027: all 11 areas explained

The 11 areas

1. Startups in a World of Abundant Space Access
2. The $15-an-Hour Expert
3. Robots Made in America
4. LLMs That Write Well
5. Plaid for Government Benefits
6. Wall Street for the Physical Buildout
7. Vertical Foundation Models
8. AI for City Hall
9. Data Center Supply Chain
10. AI for the Air-Gapped Economy
11. Own Your Intelligence

YC’s own framing for this list: AI’s biggest bottlenecks are moving beyond the models, and it is excited about startups building the hardware, infrastructure, financial systems and specialized software that bring intelligence into the real economy.

1. Startups in a World of Abundant Space Access

YC expects the cost of getting mass to orbit to keep falling, possibly by a factor of 100 or more, as very large rockets fly often and get reused. It is asking for companies built for that world: products and services in orbit that are too expensive to run today but would work at a tenth or a hundredth of today’s launch price. The question to answer is what you would build if launch were nearly free.

Sounds like: Satellite services, in-orbit manufacturing, orbital infrastructure, or anything whose economics depend on cheap launch.

Making the fit explicit, if it is real for you:

• Say what launch price your numbers assume, and what the business looks like at today’s price.
• Name the first customer who would pay for it, and what they use instead.

2. The $15-an-Hour Expert

Instead of AI replacing workers, YC wants AI that makes a capable but uncredentialed person perform like an expert. Its examples are jobs where a license or a long apprenticeship has been the bottleneck: benefits, taxes, insurance claims, medical billing and immigration paperwork. The model is a company that hires sharp people, gives them AI tools and sells the expert-level result.

Sounds like: Services or software where AI plus a trained non-specialist replaces a credentialed specialist, in paperwork-heavy fields.

Making the fit explicit, if it is real for you:

• Name the specific job title you are upgrading and the credential that used to be required.
• Give one task your AI-assisted worker finishes, how long it took before, and who pays for it.

3. Robots Made in America

YC sees the limiting factor for robotics moving from software to hardware. Most robots used in the US are imported, and many depend on parts such as rare-earth magnets from a single country. It points to rules that reward robots with mostly US-made content, and asks for founders building US robots and the components inside them: magnets, motors, gears, sensors, grippers, chips, batteries and the machines that make them.

Sounds like: Robot makers, or suppliers of the parts and tooling robots need, with a plan to build in the US.

Making the fit explicit, if it is real for you:

• Say which part of the robot supply chain you make and who you sell it to.
• State where it is made and the share of cost that is domestic, with a number if you have one.

4. LLMs That Write Well

YC argues that today’s top models have improved at coding and math while their prose has become flatter and more robotic, because the training signals that reward code work against personality. The big labs are unlikely to prioritize style, so the gap is open for a team building a model people actually enjoy reading and talking to.

Sounds like: A model or training approach built for voice, tone and style, rather than another writing app on top of a general model.

Making the fit explicit, if it is real for you:

• Say what you change in the model or training, not just the interface.
• Say how you measure that readers prefer the writing, and name the readers who pay for it.

5. Plaid for Government Benefits

Applying for benefits means proving your income, household and address, then repeating it for each program and every renewal, while agencies check documents by hand and fraud still gets through. YC compares the fix to Plaid for banking: with the person’s consent, pull verified records straight from the source and let every program use them. It notes that states face new re-verification duties and that AI agents will need trustworthy data.

Sounds like: Infrastructure that verifies income, tax or residency from source records for benefit programs, with the applicant’s consent.

Making the fit explicit, if it is real for you:

• Name the program and the state or agency that would use it first.
• Say which source records you connect to and what an agency does today instead.

6. Wall Street for the Physical Buildout

Software is cheap to start, but robots, GPU clusters and data centers need large amounts of capital up front, and the financing for them is immature: banks may underwrite a robot like a forklift. YC says past buildouts, from railroads to telecom, produced large financial companies, and it wants founders who understand both the machines and the money to build the lenders, leasing firms, securitizations and hedges for this one.

Sounds like: Financial products or marketplaces for equipment-heavy AI assets: robot lenders, GPU leasing or hedging, data center financing.

Making the fit explicit, if it is real for you:

• Name the asset you finance and the customer who borrows against it.
• Explain why you can price the risk better than a bank, using data you have that they don’t.

7. Vertical Foundation Models

YC wants research-led teams to build specialized foundation models for fields like manufacturing, energy and biology. The target is a model that works with little or no task-specific training and matches or beats a general model while using far less compute, so it is cheap enough to run on everything. The hard parts are getting rights to good domain data and proving a clear win over existing specialist models.

Sounds like: A domain-specific model, not an app: it must be measurably better or cheaper than general models in its field.

Making the fit explicit, if it is real for you:

• Name the domain and the exact task your model does better than a general model.
• Give the benchmark or comparison, and say who has the data rights.

8. AI for City Hall

Getting a building or business permit can take months, and city staff spend their time checking drawings against code, matching applications to zoning and trading corrections. YC says AI can do much of that, that delay costs cities money and keeps projects from being built, and that selling software to local government has become easier. It suggests starting with permits, then moving into 311, public records, licensing and inspections.

Sounds like: Software for city or county agencies that automates review, intake or correspondence in permitting and similar services.

Making the fit explicit, if it is real for you:

• Name the type of permit or request and the kind of city office that buys your product.
• Say how long that step takes today and what changes with your tool.

9. Data Center Supply Chain

YC describes the AI data center buildout as one of the largest capital projects ever, and says the obvious shortages (chips, power, electricians) are joined by less visible ones. Large transformers have multi-year lead times, switchgear and liquid cooling gear are sold out, and one key electrical steel is made at very few plants. It wants companies that unlock any part of that supply chain.

Sounds like: Hardware, materials, manufacturing capacity or software that removes a specific shortage in building data centers.

Making the fit explicit, if it is real for you:

• Name the exact component or material and how long customers wait for it today.
• Say who buys it (a data center builder, utility or equipment maker) and what you do to shorten the wait.

10. AI for the Air-Gapped Economy

Some of the most valuable data sits in defense contractors, banks, hospitals and nuclear plants that cannot send it to an outside AI service. YC says founders keep hearing customers want the product running entirely inside their own building with no internet connection. It wants on-prem models, deployment tools, security accreditation and hardware, and notes the sales are slower but the budgets and moats are larger.

Sounds like: Products that run inside a customer’s own secured environment, including the accreditation and hardware to do it.

Making the fit explicit, if it is real for you:

• Name the kind of customer, the environment it runs in and the approval you need.
• Say what data stays inside and what the customer uses today.

11. Own Your Intelligence

Most companies building with AI rent their intelligence by sending requests to a frontier lab. YC thinks that is changing: open-weight models are now competitive, and companies in regulated industries especially want to own models they can improve on their own data and run where they choose. The missing piece is the work in between: turning workflows into evaluations, expertise into training data and production failures into improvements.

Sounds like: Tools or services that help a company train, evaluate and keep improving its own specialized model.

Making the fit explicit, if it is real for you:

• Name the workflow and the industry whose data you turn into a better model.
• Say what the customer owns at the end and how their model improves over time.

How to use this list

• Fit is a topic, not a trick. Name your real customer and the specific problem you solve. Do not rename your startup to match a heading.
• Keep the one-liner plain. Spaces count toward the 50-character limit; see how to write your one-liner.
• Explain, don’t claim. “We build X for Y because Z” beats “we are in YC’s wishlist category.”
• Not on the list? That is fine. Most of what YC funds is not, by its own account.

Source: YC’s Requests for Startups page, Winter 2027 section, checked October 28, 2026. Related: every application question, explained · how to get into YC · deadlines · the pitch library, filtered by area.

Common questions

Do I have to pitch one of these ideas to apply to YC?

No. YC says on the page itself that these ideas are only a fraction of what it funds, and that you do not need to work on one of them to apply. If one fits what you are already building, it can be extra validation, but it is not a requirement.

How often does the list change?

YC publishes a new Requests for Startups list for each batch. This page covers the Winter 2027 list, which we last checked on October 28, 2026. If YC updates its page, ours may lag, so check YC’s Requests for Startups page for the current version.

Does Placet check whether my startup fits one of the areas?

Yes, as an optional note in the review. If your answers clearly match an area, the review says so and suggests how to make the fit explicit. If nothing matches, it says nothing. A match is only a topic match: it does not predict anything about how YC will decide.

Are the example one-liners real startups?

No. Every example on this page is a made-up company written by us to show how a clear one-liner for that area could read. None of them is a real company or a YC example.