Compute Sovereignty: Tension Between Open Models and Closed Fabs
Open models lower AI barriers, but training and large-scale deployment still rely on specific regions' power and manufacturing capacity.
Over the past two years, publicly weighted and open inference frameworks have significantly lowered the barrier to "using large AI models," but training and large-scale deployment still highly rely on specific regional resources.
Software Side Progress
Open source model progress:
| Dimension | 2025 | 2027 | |-----------|------|------| | Mainstream open model parameters | 70B | 7B (quantized) | | Inference cost (/token) | $0.1 | $0.003 | | Deployment barrier | A100×8 | RTX 4090 |
The software-side open-source movement has indeed lowered AI usage barriers.
Hardware-Side Concentration
However, reverse concentration trends on hardware side:
| Resource | Concentrated Region | Risk | |----------|-------------------|------| | Advanced processes (below 3nm) | Taiwan (~60%), Korea (~30%) | Geopolitical | | GPU capacity | NVIDIA (~80%) | Supply chain | | Energy costs | North America, Middle East | Energy security |
本文为虚构内容，不构成投资建议。
Disclaimer
This article is demo content on the site, consistent with the notice at the top: it may be fictional or synthetic. Do not use it as a basis for real decisions. Do not cite it as factual reporting.