Science of Apps - Launchpad · Explainer

Y Combinator startup ideas:
what YC wants funded in 2027

Y Combinator's startup ideas for Winter 2027 are 11 Requests for Startups, focused on bringing AI into the real economy: AI for city hall, a 'Plaid for government benefits', robots made in America, data-centre supply chains, vertical foundation models, on-premises AI for air-gapped organisations, LLMs that write well and more. YC says you don't need one of these ideas to apply.

By , FounderUpdated 12 min read

Share

Quick answer: What startup ideas does Y Combinator want?

  1. 11 requests in YC's Winter 2027 Requests for Startups (RFS).
  2. Theme: AI moving into factories, data centres, government and space.
  3. Optional: YC says the requests are a fraction of what it funds.
  4. Paul Graham's rule: look for problems, ideally your own.
  5. Just an idea is fine: YC says about 40% of funded companies in each batch are just an idea.

Key takeaways

  • YC's RFS tells you which problems partners think are opening up now, and why.
  • The Winter 2027 list is less about chatbots and more about the physical and public economy.
  • Paul Graham's 2012 essay still explains why sitting down to think up ideas rarely works.
  • A request is a direction, not a business; you still need customers who confirm the problem.
  • Test a YC-style idea with buyers before you apply or build.

Every batch, Y Combinator publishes a short list of problems its partners want founders to work on: the Requests for Startups, or RFS. It's the closest thing to a public wish list from one of the best-known accelerators, and each request explains why the problem is opening up now, the part of an idea that's hardest to judge alone. This guide covers all 11 Y Combinator startup ideas for the Winter 2027 batch, what Paul Graham's classic essay says about getting ideas, what YC's FAQ says about applying with just an idea, and how to test a YC-style idea free. Everything was checked on YC's own pages on 18 November 2026. Jump to the 11 requests, Paul Graham's essay or testing your idea.

Y Combinator Requests for Startups page with tabs from Winter 2027 back to Summer 2024 and an intro saying AI's biggest bottlenecks are moving beyond the models
YC's Requests for Startups, Winter 2027, with earlier editions in tabs.Image: Y Combinator · Screenshot shown for commentary, 18 Nov 2026 · source

What startup ideas does Y Combinator want in 2027?

YC frames the Winter 2027 list around one idea: AI's biggest bottlenecks are moving beyond the models, into hardware, infrastructure, finance and specialised software for the real economy.

#RequestYC partnerThe problem in one line
1Startups in a World of Abundant Space AccessAnkit GuptaBusiness models that only work once launch costs fall
2The $15-an-Hour ExpertBrad FloraAI that lets workers without credentials do specialist work (tax, claims, billing)
3Robots Made in AmericaDiana HuRobots and their domestic supply chain: magnets, motors, gears, sensors, batteries
4LLMs That Write WellJared FriedmanA model tuned for natural writing, not just code and maths
5Plaid for Government BenefitsGustaf AlströmerConsent-based retrieval of verified income, tax and residency records
6Wall Street for the Physical BuildoutBrad FloraLending, leasing and hedging for robots, GPU clusters and data centres
7Vertical Foundation ModelsDiana HuModels for manufacturing, energy and biology
8AI for City HallGustaf AlströmerPermitting, code review, 311, licensing and inspections
9Data Center Supply ChainJared FriedmanBottlenecks in transformers, switchgear, cooling and electrical steel
10AI for the Air-Gapped EconomyBrad FloraOn-premises AI and accreditation for offline organisations
11Own Your IntelligenceJon XuTools for companies to build their own models on proprietary data

Table: Y Combinator's Requests for Startups, Winter 2027 (ycombinator.com/rfs, checked 18 Nov 2026)

The 11 Y Combinator ideas, explained

Below, each request in plain words, who it suits, and a first step that costs nothing. The "who it suits" and "first step" notes are ours, not YC's.

1. Abundant space access

If reusable rockets cut the cost per kilogram by orders of magnitude, new businesses become possible. Suits founders from aerospace. First step: list three businesses that fail at today's launch prices and work at a tenth of them.

2. The $15-an-hour expert

AI tools that let people without formal credentials do specialist work in taxes, insurance claims and medical billing. Suits founders who have done that work. First step: shadow someone doing it for a day.

3. Robots made in America

Robots plus the domestic supply chain behind them, as new US content rules arrive. YC's page notes China produced 94% of sintered rare-earth magnets in 2024. Suits hardware founders. First step: map one robot's bill of materials by country.

4. LLMs that write well

Jared Friedman argues training for code and maths has made model prose more robotic. Suits ML researchers. First step: blind-test outputs with readers who don't know which model wrote what.

5. Plaid for government benefits

Pull verified records from government sources, with consent, so people qualify for several programmes at once. YC's page points to Medicaid re-checks every six months starting in 2027. First step: list the documents one programme asks for.

6. Wall Street for the physical buildout

Lenders, leasing firms, securitisation and hedging for robots, GPU clusters and data centres. Suits finance founders. First step: price one lease.

7. Vertical foundation models

Models for manufacturing, energy or biology that work with little task-specific training. Suits founders with proprietary data access. First step: secure a data partner before writing code.

8. AI for city hall

Speeding up permitting, code review, 311, licensing and inspections. First step: read a month of your city's public 311 data and count the request types.

9. Data-centre supply chain

Shortages of transformers, switchgear, cooling and electrical steel. Suits founders from utilities, construction or procurement. First step: interview three procurement leads about lead times.

10. AI for the air-gapped economy

On-premises AI, deployment tools and security accreditation for organisations that can't connect to the internet, such as defence contractors and hospitals. First step: ask an IT lead what blocked their last AI pilot.

11. Own your intelligence

Help companies build and improve their own specialised models on their own data, instead of only renting frontier models. First step: offer one fine-tune as a service.

Startup ideas, Y Combinator style: what the list tells you

Three patterns run through the Winter 2027 requests, and they're useful even if you never apply:

  • "Why now" is explicit. Every request names the change that makes it possible: cheaper launches, new content rules, Medicaid re-checks, data-centre demand.
  • Unglamorous problems. Permits, switchgear, billing appeals. Paul Graham called the habit of avoiding such work "schlep blindness".
  • Industry access matters. Most requests are hard to start without having worked in the field, which is YC's top recipe for ideas.

From a request to a wedge: a worked example

A request is too broad to build. The work is narrowing it to one buyer, one task and one first sale. Here is how that might look for "AI for city hall". This is a hypothetical illustration, not a real company.

Diagram of 6 steps for from a request to a wedge: a worked example: 1. Pick one task; 2. Pick one buyer; 3. Define the result; 4. Do it by hand first; 5. Charge for the manual version; 6. Then build.
From a request to a wedge: a worked example
  1. Pick one task — of permitting, code review, 311, licensing and inspections, choose the one with the longest queue in your city. Public dashboards and council minutes often show it.
  2. Pick one buyer — the city may buy slowly; the applicant (an architect or small builder) may pay faster to avoid resubmissions.
  3. Define the result — for example, "an application that passes the first review", not "AI for permits".
  4. Do it by hand first — review five real applications against the published code and note every reason one would bounce.
  5. Charge for the manual version — if builders pay for a pre-check done by a person, software that does it faster has a market.
  6. Then build — automate the checks you did most often, and keep a person on the rest.

The same narrowing works for every request on the list: "data-centre supply chain" becomes "lead-time alerts for one class of transformer"; "the $15-an-hour expert" becomes "a guided workflow for one type of insurance appeal". The narrower the first version, the sooner you learn whether anyone pays.

Paul Graham: how to get startup ideas

Paul Graham's essay "How to Get Startup Ideas" (November 2012) is still one of the clearest answers to the question. Its argument, in brief:

Checklist: Paul Graham: how to get startup ideas — Look for problems; Organic, not brainstormed; Live in the future; A deep well beats a shallow hole; Beware schlep blindness; Turn off the unsexy filter.
Paul Graham: how to get startup ideas
  • Look for problems — the best ideas come from problems, ideally your own; most startups fail by solving problems nobody has
  • Organic, not brainstormed — sitting down to think up ideas tends to produce plausible-sounding bad ones
  • Live in the future — be at the leading edge of a changing field and you'll notice what's missing
  • A deep well beats a shallow hole — a few people who want something urgently beat many who want it a little
  • Beware schlep blindness — founders avoid tedious problems, which leaves them less contested
  • Turn off the unsexy filter — whether a problem seems glamorous says little about its value

His most-quoted line: "Live in the future, then build what's missing." YC's Jared Friedman later turned the same thinking into seven recipes, best first; our guide to how to find startup ideas summarises them.

Y Combinator Startup Library page How to get startup ideas by Jared Friedman
Jared Friedman's talk in YC's Startup Library.Image: Y Combinator · Screenshot shown for commentary, 17 Nov 2026 · source

Applying to YC with an idea

Just an idea?
About 40% of funded companies per batch, YC says
Students
Can apply; full-time during the batch if accepted
Standard deal
$500,000 ($125k for 7% + $375k uncapped MFN SAFE)
Winter 2027
Jan–Mar, in person in San Francisco
Catch
On-time deadline (2 Nov) has passed; late applications considered
Read YC's FAQ
Y Combinator Frequently Asked Questions page with sections for Logistics, Should I apply?, Incorporating and Late Applications
YC's FAQ covers applying with an idea, students and late applications.Image: Y Combinator · Screenshot shown for commentary, 18 Nov 2026 · source
  • Cost: YC charges no fees; its standard deal is described on its deal page.
  • When: the Winter 2027 batch runs January to March; YC's apply page says late applications are considered, without a promised response time.
  • Rules: the batch is in person in San Francisco, starting with a three-day retreat.
  • Other options: see our startup accelerator list for programmes with different terms.

Which YC request fits your background?

Your backgroundRequests to look at
Aerospace or hardwareSpace access; robots made in America
Finance or lendingWall Street for the physical buildout
Government, social work, benefitsPlaid for government benefits; AI for city hall
Billing, tax, insurance operationsThe $15-an-hour expert
ML researchLLMs that write well; vertical foundation models
Enterprise IT or securityAI for the air-gapped economy; own your intelligence
Utilities, construction, procurementData-centre supply chain

Table: Matching YC's Winter 2027 requests to founder backgrounds (Science of Apps editorial)

How to test a YC-style idea before you build

Diagram of 6 steps for how to test a yc-style idea before you build: 1. Read the request twice; 2. Find ten people with the problem; 3. Sell the manual version; 4. Put up one page; 5. List it free; 6. Apply with evidence.
How to test a YC-style idea before you build
  1. Read the request twice — note the "why now" and who has the problem.
  2. Find ten people with the problem — ex-colleagues first; ask about the last time it happened.
  3. Sell the manual version — a report, an audit or a done-for-you service.
  4. Put up one page — what it does, for whom, and a way to get in touch.
  5. List it free — a Launchpad profile gives the idea a permanent, findable page.
  6. Apply with evidence — what ten users told you is stronger than any pitch.

The Launchpad is free to list, forever, and ranks by relevance, editorial picks and community interest, never payment. Once your page is live, a free baseline diagnosis checks your landing page, discoverability and speed. When there's a product behind it, a full investigation tests it end to end in a real browser on desktop and mobile, with every finding approved by an analyst.

Areas of your product tested
22
Automatic measurements, desktop and mobile
39
Expert checks behind your login
74
Verdicts that rank what to fix first
5
Before demo day, or before applying

See your product the way a partner will

List free, then order the report that fits your stage.

Economy

See what new customers see

  • Store listing and homepage, screen by screen
  • Scorecard against 60 Laws of UX
  • The fixes worth most
≈ 40 pagesDelivered within 3 hours
Most popular
Premium

The full diagnosis and the plan

  • Every screen, every finding, every flow
  • All 60 laws with evidence and fixes
  • 8 strategy chapters and a 90-day roadmap
≈ 95 pagesDelivered within 24 hours

Private to your team. Never affects ranking.

Mistakes with Y Combinator ideas

Checklist: Mistakes with Y Combinator ideas — Copying the title, not the problem; No industry access; Building before selling; Treating the RFS as required; Ignoring the schlep.
Mistakes with Y Combinator ideas
  • Copying the title, not the problem — "AI for city hall" is a direction; pick one task in one city
  • No industry access — most of these requests need buyers you can reach this month
  • Building before selling — a manual service proves demand faster than a model
  • Treating the RFS as required — YC says plainly you don't need these ideas to apply
  • Ignoring the schlep — the tedious part (accreditation, procurement) is often the moat

Ready to launch?

How we know

The 11 requests, their authors and the facts in them come from YC's Requests for Startups page (Winter 2027 edition); the application facts from YC's FAQ, deal and apply pages; the essay summary from Paul Graham's "How to Get Startup Ideas" (November 2012). All were read on 18 November 2026. "Who it suits" and "first step" notes are our editorial suggestions. Science of Apps runs the Launchpad and has no affiliation with Y Combinator.

Key terms, defined

  • Requests for Startups (RFS): YC's published list of problems it would like founders to tackle.
  • Batch: a YC cohort, such as Winter 2027, that runs together for three months.
  • SAFE: a simple agreement for future equity, YC's standard investment instrument.
  • MFN: "most favoured nation", a term that converts on the best terms given to later investors.
  • Schlep blindness: Paul Graham's term for founders' tendency to avoid tedious problems.
  • Air-gapped: a system with no connection to the internet, common in defence and healthcare.

Frequently asked questions

What startup ideas does Y Combinator want?

For Winter 2027, Y Combinator lists 11 Requests for Startups: space access, the $15-an-hour expert, robots made in America, LLMs that write well, Plaid for government benefits, Wall Street for the physical buildout, vertical foundation models, AI for city hall, the data-centre supply chain, AI for the air-gapped economy, and owning your own intelligence. YC says it funds much more than these.

Do I need to work on a YC Request for Startups to get in?

No. YC's RFS page says the requests represent just a fraction of what it funds, and that you don't need to work on these ideas to apply. If one excites you, YC suggests taking it as extra validation. Most YC companies apply with their own idea.

What does Paul Graham say about how to get startup ideas?

In his November 2012 essay, Paul Graham says the way to get startup ideas is to look for problems, preferably ones you have yourself. He argues that deliberately brainstorming produces plausible but bad ideas, that the best ideas often look like toys or schleps, and that you should live in the future, then build what's missing.

Can I apply to Y Combinator with just an idea?

Yes. YC's FAQ says it would be a mistake to wait, and that on average about 40% of the companies it funds in each batch are just an idea. It recommends applying as soon as you have a founding team and an idea you are excited about.

How much does Y Combinator invest?

YC's standard deal is $500,000: $125,000 on a post-money SAFE for 7% of the company, plus $375,000 on an uncapped MFN SAFE that converts on the terms of the lowest-cap SAFE issued before the priced round. YC says the investment isn't milestone-based and it charges no fees.

Where can I see older Y Combinator ideas?

The RFS page has tabs for earlier editions, back to Summer 2024 when we checked on 18 November 2026. YC's Startup Library also hosts Jared Friedman's talk 'How to get startup ideas', which lists seven recipes for generating ideas; our guide to finding startup ideas summarises it.

Can students apply to Y Combinator?

Yes. YC's FAQ says you can apply while a full-time student or employee, but founders are expected to work full-time on the company during the batch if accepted. It also offers Early Decision for students who want to join the batch after graduating.

Sources

  1. Y Combinator — Requests for Startups, Winter 2027 (checked 18 Nov 2026)
  2. Y Combinator — Frequently Asked Questions (checked 18 Nov 2026)
  3. Y Combinator — The YC deal (checked 18 Nov 2026)
  4. Y Combinator — Apply: Winter 2027 batch (checked 18 Nov 2026)
  5. Paul Graham — How to Get Startup Ideas, November 2012 (checked 18 Nov 2026)
  6. Y Combinator Startup Library — How to get startup ideas (Jared Friedman) (checked 18 Nov 2026)
  7. Science of Apps — Launchpad: list your product free (checked 18 Nov 2026)
  8. Science of Apps — Product investigation: 22 areas (checked 18 Nov 2026)
  9. Checked November 18, 2026. Figures that change — fees, salaries, prices — are reviewed every six months; if something is out of date, tell us.
Pass it on

Know someone who'd enjoy this?

Send it to them — one tap, no sign-up. Good ideas travel best between friends.

Keep exploring

Where to go from here

Join free

Get the ideas, not the noise.

Sign up with Google and we’ll send you short UX psychology tips by email — why people tap, stall and leave, and what to do about it. Unsubscribe anytime. Prefer something to keep? Get the free Laws of UX guide as a PDF.

Get the free Laws of UX PDFFree · no spam · unsubscribe anytime
Free, forever

Launch day ends. Your listing shouldn't.

Building on one of YC's requests? List it free on the Science of Apps Launchpad for a permanent, findable profile, early feedback, and your product tested end to end with a detailed report.

Free PDF · 74 pages · 2 editions

The Laws of UX, in one field guide

60 research-backed principles in 6 categories — what each one means, why it works on the human brain and how to use it on your own screens. Written by Sushant Singh, founder of Science of Apps, who has scaled consumer apps to lacs of users and spent 10+ years leading product across startups and digital businesses in the US, EU & India.

Pick your edition
Free PDF · 90 pages · 2 editions

100+ Subscription Paywalls, decoded

110 real paywalls from top apps, taken apart, rated out of 5 and decoded with the Laws of UX — plus the frameworks, copy and checklists to ship your own. By Sushant Singh, founder of Science of Apps.

  • 110 paywalls rated
  • 34 Laws of UX
  • 12 frameworks

Which paywall scored 4.8 out of 5, and which scored 2.6? It’s all inside.

Pick your edition
Free PDF · 188 pages · 2 editions

Unmistakable: the logo book

How the world’s greatest logos are made, chosen and kept — and how to make yours. 98 short lessons and 19 true stories for founders, designers and the curious. By Sushant Singh, founder of Science of Apps.

  • 98 lessons
  • 19 true stories
  • 60+ real logos

Why did a $35 sketch become the Swoosh, and why did Gap lose its nerve in six days? It’s all inside.

Pick your edition