<- all tokdocs

YC's "Own Your Intelligence" Request for Startups, Explained

By , founder of Mavric and Apso

View on TikTok ->

"Own Your Intelligence" is an entry in Y Combinator's Winter 2027 Requests for Startups, written by Jon Xu, asking for startups that help companies train and own specialized AI models instead of renting frontier models. A TikTok from @y_combinator reads the request almost word for word. Its central claim, that open-weight models are "competitive with the top frontier models on meaningful benchmarks", names no benchmark or model, and the cost and history claims around it carry no figures.

What is YC's "Own Your Intelligence" request?

The YC Requests for Startups page lists "Own Your Intelligence" under the Winter 2027 heading, with Jon Xu as author. The page shows no publication date. A FourWeekMBA write-up dated 2026-10-28 quotes its opening line, "Today, nearly every company building with AI rents its intelligence." The video, posted 2026-10-30, opens with the same sentence.

The video's lower third begins "Jon X" and "General", which fits a title card for Xu. The YC page calls him a partner.

What is the argument?

The monologue runs in four steps:

  1. Companies rent intelligence from frontier labs.
  2. That "was an easy trade when the best models were years ahead of everything else," and "nothing you did with your own data could close the gap."
  3. Open-weight models have now caught up on "meaningful benchmarks."
  4. Companies will want to own specialized models trained on their own workflows, data and failures.

The video does not argue that renting ends. It says "Companies will keep renting frontier intelligence for their hardest problems."

Are open-weight models really competitive with frontier models?

The video says "Open weight models are now competitive with the top frontier models on meaningful benchmarks." It names no model, benchmark, score or date, and neither does the YC page text or the FourWeekMBA summary.

The wording is hedged. "Competitive" on "meaningful benchmarks" does not claim open models match frontier models on every task, and the later line about renting for the hardest problems is consistent with that. To test the claim yourself, pick the benchmarks that match your work and compare current open-weight and frontier results from a leaderboard or model card.

When were frontier models "years ahead"?

The video gives no year, model or size for that gap. The claim explains why the economics have supposedly changed, but nothing in the video or the sources dates when the gap closed.

Is renting AI more expensive in regulated industries?

The video says the issue "matters most in regulated industries, where strict data governance and lockdown infrastructure make renting even more expensive for the performance they get." It names no cost figure, industry, regulation or customer, and the YC page does not add numbers. Read it as YC's judgment.

What kind of startup is YC asking for?

The most concrete passage lists the work: "Someone has to turn workflows into evals, expertise into training data, production failures into improvements, and carry all of it forward with each new base model release." That is four jobs:

  • building evaluations from a company's real workflows
  • turning expert knowledge into training data
  • feeding production failures back into training
  • moving all of it to each new base model when one is released

The video closes with "If you're helping companies own their intelligence, we want to hear from you." It gives no URL, deadline or example company, and it shows no demo of a model or training run. The YC page is where the request lives.

Is "own your intelligence" a YC-only idea?

No. Search results surfaced related commentary from Sequoia, also titled "Own Your Intelligence", and from Fireworks AI arguing a similar thesis. I did not read those pages, so they show only that other firms use the phrase.

Bottom line

The video is YC reading its own Winter 2027 request: help companies turn workflows, expertise and failures into models they own. The four jobs it lists are a concrete description of the opportunity. The supporting claims, that open-weight models are competitive on benchmarks, that frontier models were once years ahead and that regulated industries pay more to rent, come with no benchmarks, dates or figures.

Sources

Published October 30, 2026. Based on a TikTok by @y_combinator. Full video notes (markdown)