Y Combinator regularly publishes RFS for each batch, where partners write short essays about the companies they would like to see. I love reading them as a way to glimpse the future: these are people who see thousands of applications a year and have a rough idea of where money will go and where markets will appear in 3-5 years.
https://www.ycombinator.com/rfs
This time there are 11 ideas. The most interesting ones, in my opinion:
- LLMs That Write Well. Jared Friedman believes that new models write worse than those from 2024-2025: the same reward signals that make them strong in coding "flatten" their voice and style. Large labs will continue to optimize for coding and agents, so the niche for a model that's pleasant to converse with is open
- The $15-an-Hour Expert. AI that turns an employee without a specialized education into an expert in taxes, insurance claims, medical billing, or immigration documents. Quote: "The last industrial revolution built the middle class from factory jobs. This one can build it from expert roles"
- Data Center Supply Chain. The bottleneck in building data centers has long been more than just GPUs and electricity. The lead time for large power transformers is three years, switchgear and liquid cooling are sold out, and electrical steel is made by only one plant in the US
- Startups in a World of Abundant Space Access. SpaceX is planning for 10,000 Starship launches a year. YC suggests building businesses that don't make economic sense today but will when launch costs drop by 100 times
- AI for the Air-Gapped Economy. Defense, banks, hospitals, and nuclear plants keep their most valuable data offline and can't use cloud APIs. Selling there is slow and painful, but they have the biggest budgets
- Own Your Intelligence. Open-weight models have caught up with the frontier on many benchmarks, and companies want to stop renting intelligence and have their own models, fine-tuned on their own data
It's noticeable that almost half of the list is about the physical world: robots made in America, funding for GPU clusters and data centers, supply chains. Software AI has become infrastructure, and the shortage has shifted to hardware, energy, and construction funding.
#ycombinator@rvnikita_blog #startups@rvnikita_blog #ai@rvnikita_blog #llm@rvnikita_blog #space@rvnikita_blog