Y Combinator is looking for founders who can make government procedures work faster. Think processing permits, checking applications, and connecting records across agencies feature among the problems in its latest startup wishlist.
The Silicon Valley accelerator’s Winter 2027 Requests for Startups cover 11 areas, including public administration, robotics and space businesses. This round goes even further than requesting improved AI models. YC sees growing opportunities in the equipment, financing and specialised software needed to put AI into everyday use.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
The requests explain what its investment team would like founders to tackle. They represent a fraction of YC’s interests, and applicants can pursue ideas outside the list.
Government Services As A Business Opportunity
In AI for City Hall, YC partner Gustaf Alströmer takes aim at a frustration many have encountered when looking to build a home or start a business: waiting for a permit from the municipality. In some American cities, he writes, getting a permit can take longer than building the house itself.
Municipal teams are tied up manually checking drawings against building codes, comparing applications with zoning rules and sending corrections back and forth. Alströmer wants startups to use AI to help alleviate this administrative bottleneck.
Those delays, he states, have significant financial consequences. Each additional month makes projects more expensive, and some developments never proceed. Cities also miss out on the tax income that new buildings would generate.
YC estimates that if all US building permits were approved in a tenth of the time they currently take, “US cities would make $16 billion a year in additional tax revenue.”
Alströmer wants founders to build AI tools that help municipal employees handle this workload and move applications through the process more quickly. His request extends beyond building permits to licensing, inspections and public records.
“The conventional wisdom says you can’t sell software to government,” he writes. “That used to be true. It isn’t anymore.”
Delays are only one part of the administrative burden. His second request, Plaid for Government Benefits, addresses the information applicants need to repeatedly submit to different agencies.
People supply proof of income, residence and household circumstances for separate benefits, then repeat the process at each renewal. Alströmer proposes connecting programmes to details such as verified income, tax and residency records, with the applicant’s consent.
Where The Requests Meet Cyprus
Cypriot authorities have identified related problems. In an April 2026 presentation, Larnaca’s building licensing authority highlighted incomplete applications and delays in consultations with government departments. It also stated that not all the departments were registered on the electronic permitting platform Hippodamus.
Businesses are also being invited to develop solutions to other public-service needs. A public-transport challenge under Cyprus’s AI in Government programme announced it was looking for better analysis of travel demand to improve bus routes and timetables. The €650,000 call attracted 27 proposals, according to the Research and Innovation Foundation. The programme supports prototype development and testing with public authorities. Participants retain ownership of their innovation for future commercial use, allowing them to develop a product that is not limited to its initial application.
Specialist Models And Company Knowledge
In Vertical Foundation Models, YC’s Diana Hu calls for models developed for fields such as manufacturing, energy, and biology. She wants them to match or outperform general-purpose systems while requiring substantially less computing power to produce results.
These models should work without extensive retraining for every task, potentially using only a handful of examples. Hu points to Google’s TimesFM-3, which forecasts several related sets of measurements together. Users supply historical information as context, rather than commissioning a separate training process for each forecast.
Access to suitable data is central to her proposal, alongside reliable ways to measure whether a model’s outputs are correct.
In Own Your Intelligence, YC’s Jon Xu moves the focus from industry to an individual company. Improvements in models with publicly available weights, such as the numerical values learned during training, make it a more suitable option for businesses to adapt and operate their own systems, he argues.
His interest lies in the continuing work required to make those systems useful. That includes converting business tasks into performance tests, preparing training material from company expertise and using failures to guide improvements. Those capabilities must carry forward when the underlying model changes.
Xu expects companies to continue buying access to leading models for their hardest problems, while simultaneously developing more specialised systems that accumulate knowledge of their own operations.
AI That Runs Without An Internet Connection
Some customers are concerned about where their software runs. In AI for the Air-Gapped Economy, YC’s Brad Flora describes founders encountering organisations that want AI installed entirely within their own premises, without an internet connection.
The most likely customers he points to are defence contractors, banks, hospitals and nuclear facilities. Their sensitive information can make external AI services unsuitable.
Serving these organisations would involve models, deployment software, hardware and security accreditation. Flora sees the demanding implementation and sales process as part of what could help companies establish a more sustainable position in this market.
Models That Write Well
Writing quality in this round has received a request of its own. In LLMs That Write Well, YC partner Jared Friedman argues that improvements in coding and mathematics have come with a deterioration in prose.
“The same reward signals that make models good at code actively flatten voice and style,” he writes.
Friedman sees room for a startup that makes natural, distinctive writing a priority. He wants founders to build a language model that writes naturally and has a distinctive voice.
Helping Workers Handle Specialist Tasks
Flora’s The $15-an-Hour Expert considers businesses in which AI helps employees undertake work that previously required specialist knowledge.
He envisions companies hiring capable people without conventional credentials and supporting them with AI. Suggested applications include insurance claims, medical billing, tax administration and immigration paperwork.
The model combines an employee’s ability to work with customers and use AI effectively with access to knowledge that would previously have required more specialised experience. Flora wants founders to build companies around that expanded capability, creating services through which workers can handle more demanding tasks.
Robots And The Equipment Behind AI
The list also addresses the physical machines and components needed to extend AI’s use.
In Robots Made in America, Hu argues that the push to improve robot models is also a push towards hardware. As systems become better at transferring skills between robots, the availability of machines and their components becomes increasingly important.
Hu is bringing forward the gap in US manufacturing. Most robots used in the country are imported from Japan and Europe, she states, while even American assembly can depend on components such as Chinese-made magnets. That dependence can interrupt production. China’s restrictions on rare-earth exports in April 2025 left some carmakers struggling to obtain magnets and forced them to reduce or temporarily halt production.
New US rules add to the pressure to manufacture locally. In July 2026, the Federal Communications Commission restricted new equipment approvals for certain foreign-produced advanced mobile robots, with exemptions available through conditional approval.
Hu therefore wants founders to build the supply chain within the US as well as the finished robots. Her call includes manufacturers of motors, magnets, sensors, grippers, chips and battery packs, along with the machinery needed to produce those parts. A startup supplying a critical component could serve several robot manufacturers.
Friedman’s Data Center Supply Chain examines another set of physical constraints. He describes shortages extending beyond graphics processors to transformers, electrical switching equipment, circuit breakers and liquid-cooling systems.
His interest reaches further back into the materials needed to manufacture that equipment. The proposed businesses would help computing infrastructure get built and operated, addressing bottlenecks throughout its supply chain.
Financing The Machines
In Wall Street for the Physical Buildout, Flora asks who will pay for this expansion. Robots and computing installations require substantial spending before they begin earning money.
“Somebody has to pay for all that steel and silicon up front,” he writes.
He highlights founders developing lending for robots that, in his example, earn $500,000 a year each. These are machines with earning potential, whose purchase still needs to be funded.
Flora is looking for lenders and leasing companies serving this market. Leasing would allow customers to use equipment and pay over time rather than purchase it outright.
His argument is that the financing market has yet to catch up with the technology, leaving founders to develop arrangements themselves. Just as earlier waves of railway, electricity and telecommunications construction supported financial businesses, he expects the physical expansion of AI to need its own specialists.
Building For Cheaper Access To Space
Ankit Gupta’s Startups in a World of Abundant Space Access asks founders to consider companies whose economics could change as launch costs fall.
The YC partner bases his request on the development of SpaceX’s Starship and the prospect of substantially cheaper transport into orbit. He asks what businesses might become viable if launch costs were reduced by a factor of 100 or more.
Gupta wants founders to begin developing those businesses ahead of the anticipated price reductions, rather than wait until cheaper access is established.
Applying From Cyprus
Cyprus-based startups can apply to YC without an existing US headquarters. Acceptance would, however, bring additional requirements.
A company incorporated in Cyprus would need to establish a parent company in a jurisdiction accepted by YC, with its existing company becoming a subsidiary. Founders would also need to attend the three-month programme in San Francisco, although they can base themselves elsewhere afterwards.
The Winter 2027 programme runs from January to March. Its on-time application deadline is 2 November 2026 at 8pm Pacific Time, equivalent to 6am in Cyprus on 3 November. Applicants meeting that deadline are due to receive a decision by 11 December. Late applications are considered, although YC does not guarantee a response timetable.
The full requests, application information and participation requirements are available on YC’s website.
