How open model ecosystems compound
Interconnects · Nathan Lambert · 2026-05-12
Nathan Lambert argues that open AI model ecosystems gain compounding competitive advantages because roughly 80% of frontier model development compute goes to R&D rather than final training runs, a dynamic China's open ecosystem exploits more effectively than Western labs through systematic cross-lab knowledge sharing.
Appears in
Extraction
Topics: open-source-aillm-developmentchina-aiai-ecosystemcompute-economics
Claims
- Approximately 80% of compute spent building frontier models goes to R&D rather than the final training run, based on Ai2 and Epoch AI research.
- China's open AI ecosystem functions similarly to open-source software by enabling labs to learn from each other's public technical reports, reducing redundant research compute.
- Open-source AI lacks the self-reinforcing feedback loops of traditional open-source software because almost all development costs fall on the model creator rather than the user community.
- Open AI models reduce future development costs across the ecosystem but do not provide deployment cost advantages over hosted closed solutions for typical users.
- An open model consortium may become the only financially viable path to competing at frontier scale with open models.
Key quotes
Most of the compute to build a leading frontier model comes from R&D costs, rather than the compute to train the final, big model end-to-end.
The Chinese system is designed around quickly learning from your peers and avoiding double-spending research compute — or infra effort.
This shared resource is far more efficient and may become the only financially viable way to compete at the future frontier scale with open models.