Grass founder: training data remains the first product for frontier AI labs
toptickcrypto · x · 2026-09-29
Grass founder Andrej published a blog post, Abundant Intelligence, explaining why the company stays in the training-data business even as inference demand explodes. Grass supplies data to most leading AI labs, using millions of residential internet connections to reach content that modern crawlers can't access.
Key points:
- Public web access for AI crawlers has kept deteriorating over the past four years, making data one of the most critical inputs for scaling intelligence.
- Models use data in two ways: training (learning) and inference (informing decisions), with inference demand growing exponentially.
- Still, training data is Grass's first product; the long-term goal is giving AI a live connection to the world's information, forming a training-to-inference data flywheel.
More from Infra
- Google to launch experimental satellite next week to test AI data centers in space — bratton · 2026-09-29
- Sandboxes will shift from cloud devboxes to one ephemeral environment per command, with snapshots in between — ivan_bezdomny · 2026-09-29
- User ditches Whoop subscription, runs the band fully locally with AI on open-source zhoop fork — SumitGup · 2026-09-29
- Prediction: AI training and inference move to the edge, miners chasing cheap energy like Bitcoin — const_reborn · 2026-09-29
- PSA: tok/s comparisons across tokenizers are unit illusions — convert to chars/token first — challis88ocarina · 2026-09-29
- Modal Labs closing in on $750M round at $15.75B valuation — TechCrunch AI · 2026-09-29