Loop intact · workloads reach CUDA/GPU
Diverted · closed labs + ASIC in-sourcing
Flywheel up: open weights multiply workloads, workloads land on Nvidia CUDA/GPU, that demand funds the next open-ecosystem bet.
Open-weight model flywheel. Open models remove licensing and API gates, so adoption multiplies AI workloads, which disproportionately land on Nvidia CUDA and GPU compute, generating the demand and cash that fund further open-ecosystem investment. A control flips between the loop spinning and stalling as workload growth is diverted to closed frontier or API models and to hyperscaler in-house ASICs.
Substrate
Nvidia CUDA / GPU
workloads land here
closed / API
brake a · frontier labs
in-house ASIC
brake b · hyperscaler
Supply
open weights · free to run
Demand
more workloads · Jevons
Reinvestment
NeMo · Nemotron · Mercor
reported interest
1
2
3
4
5 · feedback loop · cash + willingness