Follow Beinsure
Y Combinator has published 11 Requests for Startups for its Winter 2027 batch, outlining investment interests across artificial intelligence, robotics, insurance and financial technology, industrial infrastructure and government services.
The startup accelerator is focusing on businesses that address practical constraints on AI adoption, particularly in industries where hardware availability, financing, data security and regulatory requirements limit deployment.
- Y Combinator has identified 11 startup opportunities for Winter 2027, targeting AI infrastructure, robotics, financial services, government technology and enterprise software.
- YC is seeking companies addressing hardware shortages, industrial financing, AI deployment in regulated industries and the development of specialized foundation models.
- The accelerator highlights opportunities in insurance claims automation, government benefits verification, domestic robotics manufacturing and data center supply chains.
“AI’s biggest bottlenecks are moving beyond the models,” YC said, identifying hardware, infrastructure, financial systems and specialized software as areas where new companies could help bring AI into factories, data centers, government agencies and space-related businesses.
The proposals cover everything from AI-assisted professional services and domestic robot manufacturing to specialized foundation models, secure enterprise AI and financial products for emerging infrastructure assets.
AI applications and specialized foundation models
Several of YC’s proposals focus on improving AI capabilities and making advanced models more useful for businesses with specific operational requirements.
1. LLMs That Write Well
YC partner Jared Friedman argues that improvements in coding and mathematical reasoning have come at the expense of natural-language writing quality.
According to Friedman, some current large language models produce more robotic writing than systems released in 2024 and 2025. He attributes the problem partly to training methods that reward performance on coding tasks but reduce stylistic variation and natural expression.
Friedman believes major AI laboratories will continue prioritizing coding and agent-based workflows, creating an opportunity for startups developing language models specifically optimized for writing and conversational quality.
2. Vertical Foundation Models
YC is interested in specialized AI models designed for manufacturing, energy, biology and other technical industries.
Diana Hu, a YC partner, argues that smaller models capable of performing specific tasks with limited training examples could match or outperform general-purpose systems while requiring substantially less computing power.
She points to Jev, which produces typed decisions and probabilities rather than generated text without requiring a separately trained classifier for every task.
Another example is Google’s TimesFM-3, a 330 mn-parameter forecasting model that can predict related time series simultaneously without task-specific fine-tuning. The model produces forecasts in a single forward pass, using each customer’s historical data as context.
YC sees opportunities for research teams that can obtain high-quality industry data, develop verifiable training rewards and demonstrate advantages in accuracy and inference costs over existing models.
3. Own Your Intelligence
Another proposal focuses on helping businesses develop and operate proprietary AI capabilities rather than depending entirely on external model providers.
YC partner Jon Xu argues that competitive open-weight models are making it increasingly practical for companies to maintain specialized systems trained on their own data and expertise.
The opportunity extends beyond deploying an open-source model. Businesses need tools to convert operational workflows into evaluations, incorporate employee knowledge into training data and use production failures to improve model performance.
These capabilities must also survive changes to the underlying foundation models.
YC expects companies to continue purchasing access to frontier models for demanding tasks while increasingly owning specialized systems that reflect their internal operations.
The approach is particularly relevant to regulated industries, where data governance, infrastructure restrictions and the cost of external model access can influence AI deployment decisions.
4. AI for the Air-Gapped Economy
Defense contractors, banks, hospitals and nuclear facilities often operate systems that cannot transmit sensitive information to external cloud-based AI services.
YC wants startups developing on-premises models, secure deployment tools, specialized hardware and security accreditation services for these environments.
Brad Flora, a YC partner, said companies building enterprise AI products frequently encounter customers who want the technology but require it to operate entirely within their own infrastructure, sometimes without an internet connection.
Although procurement and implementation can be more difficult in these markets, YC believes their security requirements, substantial technology budgets and operational complexity create opportunities for specialized providers.
Robotics, space infrastructure and data center supply chains
YC also identifies opportunities in the physical infrastructure required to support AI adoption, including industrial manufacturing, computing facilities and space transportation.
5. Startups in a World of Abundant Space Access
YC partner Ankit Gupta points to SpaceX’s projection of a future with 10,000 Starship launches annually as a potential foundation for new commercial activities in orbit.
According to Gupta, an expendable Starship is already approaching the cost of a reusable Falcon 9 launch while offering substantially lower costs per kilogram of payload.
As Starship production scales and the spacecraft becomes reusable, he expects launch costs per kilogram to fall by several orders of magnitude.
YC wants founders to consider business models that may become economically viable if the cost of accessing space declines by 100 times or more.
Its proposal encourages companies to begin developing those businesses before lower launch costs create a larger orbital economy.
6. Robots Made in America
YC sees an opportunity to expand domestic robotics manufacturing as advances in foundation models improve robots’ ability to transfer learned skills between different machines.
The US installed 38,400 industrial robots in 2025, an increase of 12%, according to figures cited by YC. Most robots used domestically are still imported from Japan and Europe, while US-assembled machines often rely on foreign components.
China produced 94% of the world’s sintered rare-earth magnets in 2024. Export restrictions introduced in April 2025 subsequently forced some automakers to reduce production.
Regulatory requirements have also increased pressure to localize manufacturing.
YC notes that the Federal Communications Commission stopped authorizing new models of connected mobile robots in July unless they were manufactured domestically or qualified for an exemption.
Under the requirements described by YC, more than 65% of a robot’s component cost must be domestic, rising to 75% in 2029.
The accelerator is interested in businesses manufacturing rare-earth magnets, motors, strain-wave gears, encoders, force sensors, grippers, chips, battery packs and the production equipment needed to make these components.
7. Data Center Supply Chain
YC describes the expansion of AI data centers as an exceptionally large capital investment program, creating demand for equipment and materials throughout the construction supply chain.
Demand for GPUs, electricity and skilled electricians is already well established, but the accelerator identifies several less visible constraints.
Large power transformers can have delivery lead times of three years, while switchgear, circuit breakers and liquid-cooling equipment face substantial supply shortages.
The availability of raw materials is another concern. YC notes that grain-oriented electrical steel, used in transformer cores, is produced at only one US plant, which cannot meet current demand.
Jared Friedman argues that these shortages create business opportunities for manufacturers, suppliers and service providers capable of accelerating data center construction.
Financial technology, insurance and AI-enabled professional services
YC’s financial and insurance technology proposals concentrate on developing new financing structures and using AI to reduce the cost of specialist work.
8. The $15-an-Hour Expert
Brad Flora sees potential for AI to enable workers without traditional professional credentials to perform tasks previously requiring expensive specialists.
Rather than focusing exclusively on replacing employees, the concept involves combining capable generalist workers with AI systems that provide specialized knowledge and decision support.
YC cites complex mergers and acquisitions, supply chain optimization and international trade consulting as examples of work that could become more accessible through AI assistance.
The accelerator also identifies millions of administrative and specialist roles involving employee benefits, tax preparation, insurance claims, medical billing and immigration paperwork.
These activities frequently require extensive training, familiarity with regulations or professional qualifications.
YC is interested in companies developing business models that allow AI-assisted workers to deliver specialist services at substantially lower costs.
9. Wall Street for the Physical Buildout
The expansion of robotics, GPU computing clusters and data centers requires financing structures that differ from those traditionally used for software companies.
Unlike conventional software startups, businesses constructing physical AI infrastructure must invest heavily in equipment before generating revenue.
YC argues that existing financial institutions often lack suitable methods for valuing and financing these assets.
Banks may evaluate industrial robots using conventional equipment-lending models, despite differences in their income-generating capabilities.
Flora cites emerging financing concepts involving robots that generate $500,000 in annual revenue and options contracts tied to GPU computing capacity.
YC wants companies developing specialized lending, leasing, securitization and hedging products for the physical AI economy.
The accelerator draws comparisons with earlier infrastructure expansions, including railroads, electricity and telecommunications, which created demand for financial institutions serving new asset classes.
AI for government benefits and municipal administration
Public-sector technology accounts for two additional startup opportunities, with YC focusing on fragmented administrative systems, manual verification and lengthy approval processes.
10. Plaid for Government Benefits
YC proposes a shared verification infrastructure that would allow government agencies to access reliable income, tax and residency information with an applicant’s consent.
The concept is modeled on Plaid’s approach to connecting financial applications with banking information.
Government benefit applicants currently face repeated requests to verify income, household composition and residency across different programs. Agencies often operate separate systems, requiring employees to review documents manually.
YC argues that this fragmented approach contributes to application delays and creates opportunities for fraud.
The accelerator cites a Government Accountability Office estimate suggesting that fraud across federal programs may cost the US government as much as $500 bn annually.
The increased availability of AI-generated documents could complicate verification further, including through the fabrication of convincing pay stubs.
A shared verification service could retrieve authenticated records directly from original sources, allowing applicants to complete verification once and reuse that information across multiple programs.
YC also points to requirements beginning in 2027 for states to recheck the eligibility of millions of Medicaid enrollees every six months.
It expects reliable data infrastructure to become increasingly important as AI agents begin handling government benefit enrollment and related administrative tasks.
11. AI for City Hall
YC sees opportunities for AI systems that automate municipal permitting, building-plan reviews, zoning checks, licensing and inspections.
Gustaf Alströmer, a YC partner, argues that many US cities continue to rely on manual processes that delay construction projects and increase development costs.
City employees routinely review architectural drawings against building codes, verify zoning compliance and exchange correction requests with applicants.
YC estimates that reducing building-permit approval times to 10% of their current duration could generate an additional $16 bn in annual tax revenue for US cities by accelerating construction and the resulting property tax payments.
The accelerator identifies approximately 19,000 cities in the US as a potential market for government-focused software.
Beyond construction permits, it sees applications in 311 service requests, public records management, business licensing and inspection processes.
Alströmer also challenges the assumption that selling software to government agencies necessarily limits startup growth. YC says several companies in its portfolio supplying AI technology to government customers have achieved particularly rapid growth.
Y Combinator’s Winter 2027 investment priorities
The 11 proposals share a focus on applying AI beyond general-purpose chatbots and conventional software products.
YC is directing attention toward the infrastructure needed to deploy intelligent systems in industrial, financial and public-sector environments, including domestic hardware manufacturing, secure enterprise deployment, specialized models and financing for capital-intensive technology.
Its Requests for Startups also address operational constraints that AI models alone cannot resolve, such as equipment shortages, fragmented government databases, regulatory requirements and limited access to specialized industry information.
The Winter 2027 list represents selected areas of interest rather than a restriction on eligible startup applications. YC continues to accept founders working on other business concepts, with the 11 published requests indicating opportunities it is particularly interested in supporting.
