YC’s 11 Startup Requests: Bringing AI into the Real Economy
All 11 Winter 2027 startup requests, with a corresponding official YC video in every section.
Better models do not automatically make factories, hospitals, or city governments work better. Equipment, financing, trusted data, and accountable workflows still determine whether AI can deliver useful results.
01 / What becomes viable when space access gets cheaper?
YC asks founders to consider businesses that could work if launch costs fell dramatically and access to orbit became abundant. This is a scenario for future infrastructure, not evidence that the predicted costs have already been achieved.
Our interpretation: Look for a clear customer need that is sensitive to transport costs, and test whether the company can survive while the infrastructure is still developing.
02 / Give ordinary workers access to expert capabilities
Rather than simply replacing workers, this request explores equipping capable people without specialist credentials to perform work previously reserved for experts. YC cites benefits, taxes, insurance claims, medical billing, and immigration paperwork.
Our interpretation: The hourly rate in the title is a framing device, not a verified universal cost. Licensing and responsibility still matter. Measure delivery quality and the cost of professional review.
03 / Robot progress needs components and manufacturing
YC believes better robot foundation models may shift the bottleneck toward hardware. The request focuses on US manufacturing and components such as magnets, motors, gears, encoders, force sensors, grippers, chips, and battery packs.
Our interpretation: A reliable component that can be manufactured at scale can be a business in its own right. US localization policy is the context for this proposal and should not be assumed to apply elsewhere.
04 / There is room for models that write well
YC argues that training rewards favoring coding and agentic tasks can flatten voice and style, leaving an opportunity for models focused on the writing experience. This is YC’s assessment, not a settled finding about every model.
Our interpretation: Test whether the product preserves an author’s voice, editorial intent, and expressive range. Fluency alone is insufficient; evaluation needs feedback from real writers and readers.
05 / Build trusted data connections for public benefits
Applicants repeatedly prove income, residency, and household circumstances across disconnected programs. YC proposes a Plaid-like layer that, with consent, retrieves verified information from authoritative sources for reuse.
Our interpretation: The work extends beyond document recognition to authorization, provenance, correction, and cooperation between agencies. Measure both applicant effort and manual verification work.
06 / Finance the physical AI buildout
Robots, GPU clusters, and data centers require substantial upfront capital. YC wants lenders, leasing businesses, securitization, and hedging tools adapted to these assets.
Our interpretation: Understand equipment revenue, useful life, and resale value. Software revenue assumptions cannot simply be transferred to equipment finance; utilization and technology changes affect risk.
07 / Build foundation models for specific industries
Manufacturing, energy, and biology may benefit from specialized models that handle new tasks directly or with a few examples. YC emphasizes clear advantages in accuracy and inference compute.
Our interpretation: A vertical model needs more than an industry prompt. Data rights, trustworthy evaluation or rewards, and measurable gains over existing approaches are the foundations of a lasting advantage.
08 / Bring AI into everyday city administration
Permits, plan reviews, inspections, public records, and licensing involve repetitive information work. YC encourages startups to improve these workflows and shorten processing time.
Our interpretation: Faster processing must remain accountable. Begin with completeness checks, rule retrieval, and organizing revision requests, then verify that staff can review the basis for each conclusion.
09 / Data centers need more than GPUs
YC highlights transformers, switchgear, circuit breakers, liquid cooling, and electrical steel. Its argument is that large-scale compute construction creates demand throughout the supply chain.
Our interpretation: Ask what builders are actually waiting for. Delivery times, manufacturing yield, and installation capacity may matter more than whether a product carries an AI label.
10 / Serve customers whose data cannot leave the building
Some defense, finance, healthcare, and nuclear-industry customers require data and systems to remain inside their facilities. YC is interested in models, deployment tools, hardware, and secure delivery for isolated environments.
Our interpretation: Downloading a model is only the beginning. Updates, permissions, logging, troubleshooting, and offline maintenance must be part of the product. Customers need a system they can operate over time.
11 / Help companies own intelligence that keeps improving
YC expects some companies to want models they can improve on their own data, run where they choose, and update on their own schedule. This can coexist with renting frontier models for difficult tasks.
Our interpretation: Turn workflows into evaluations, expertise into training data, and production failures into improvements. Preserve those gains when the base model changes, so the business can maintain its capabilities.
Turn the list into a map of customer problems
Ask why a customer still cannot complete a job even when a capable model exists. The obstacle may be equipment, financing, reliable data, internal deployment, or expert review. Start with a familiar workflow, record its time, errors, and cost, and test whether one specific step improves. YC notes that these requests represent only part of what it funds; applicants do not need to choose one of them.