7 batches. ~1,800 companies. One that changed everything.
414 companies. 50% international founders. The first batch with a $500K standard deal. YC had never been more ambitious, or more exposed.
Then the funding winter hit. Then interest rates climbed. Then the market decided it had no patience for early-stage experiments with no path to revenue.
That dot grid to the left: each dot is a company. The grey ones never made it.
Between S22 and S23, the AI share of YC went from 15% to 57%. Not a gradual shift. A detonation.
Founders who submitted S23 applications had seen GPT-4. Founders who submitted S22 applications had not. That one release date is the dividing line between two completely different YC eras.
W22 had 50% international founders. By W25, that number had collapsed to 10%. Garry Tan took over, reversed the pandemic-era global experiment, and rebuilt YC around physical presence in San Francisco.
"The world's accelerator became a zip code."
Cursor graduated from W23. Today it is worth $29.3 billion. The entire W22 batch, all 414 companies, cannot match it.
This is not a YC success story. It is a YC outlier story. The returns are not distributed. They are concentrated in one or two founders who were right about which specific layer of AI would matter, before it was obvious.
Note: Perplexity AI is frequently misattributed to YC. It is not a YC company. The W23 story belongs to Cursor alone.
Cursor did not win because it was the best-executed AI coding tool. It won because it was the earliest credible one, from a team with the right technical depth, at the moment the category was forming.
The founders who won from W22-W25 understood which specific layer of AI would matter before that layer was confirmed.