AI startups may find their last moat in training data, not models
chrisgrayson · x · 2026-07-22
Christopher Grayson argues that the real moat for AI startups will increasingly be training data, not just algorithms or GPUs.
In the attached quote/image, he points to the more than $50 billion raised by the AI industry in 2024 and argues that some of that capital should go toward data rights licensing and acquisition. He also says that beyond mainstream books, there is a vast long tail of out-of-print periodicals, niche magazines, trade publications, and journals that could be salvaged, digitized, and fed into AI training.
His broader point is that AI companies should treat content as a strategic asset, and that preserving deteriorating print archives matters both for digital preservation and for model training.
Related event: Training Data Emerges as the Ultimate AI Moat(2 posts)→
More from Companies & People
- AI debate in Nigerian universities centers on outsourced final-year projects — Nasereliver · 2026-07-22
- Cell-therapy company acquires STEM-PD to test its biology foundation model in clinic — arjunrajlab · 2026-07-22
- Mark Cuban returns to All-In to talk AI bubble and enterprise limits — 2C_ornot2C · 2026-07-22
- Google is filling out capability, cost, and security models for agent builders — eyishazyer · 2026-07-22
- Google DeepMind opens 3-month equity-free AI for the Planet accelerator in Asia Pacific — m4rkmc · 2026-07-22
- Pretrained LLMs may fail at company work because they lack organization-specific memory — thebvg · 2026-07-22