Here's an unpopular take： What's really choking U.S. AI isn't Chinese models-it's the choices made by its own policymakers. Chamath's thread lays this out perfectly. AI has already forked into two non-overlapping paths： Path A： U.S. closed models， $26-$56 per million tokens. Path B： Chinese open-weight models， $0.50-$1 per million tokens. That's a 50-100x price gap-not a future projection， but today's reality. The OpenRouter data says it all： Chinese models' share of token usage among U.S. enterprises has surged from under 10% at the start of 2025 to over 58% now. Companies have already voted with their feet. If policy forces U.S. firms onto Path A exclusively， Chamath's analogy is spot-on： It's like mandating the use of $800-per-barrel oil when the market price is $80. Short-term， it may look like protecting domestic labs. Long-term， it torches the AI costs of the entire U.S. economy. SMEs， developers， startups-they'll be the first casualties. Deeper logic： AI is no longer a luxury-it's infrastructure， like electricity or oil. Force a 50x markup on "electricity，" and the efficiency of the whole economy collapses. The U.S. AI sector is a core valuation driver for the stock market. Once lab finances deteriorate， the contagion to tech equities and growth sectors will be faster than most expect. This is Chamath's real point： It's not about picking sides between Chinese or American models. The technological paradigm has shifted. The cost structure is irreversible. Policy that swims against this tide won't hurt rivals-it'll hurt itself. Here's my deeper cut： With Kimi K3 set to release its open weights on July 27， the path of open models plus extreme cost-efficiency will only accelerate. Any attempt to block it with policy will be steamrolled by the trend. The winning move over the next decade isn't who trains the strongest model first. It's who delivers sufficiently powerful capabilities at the lowest total cost of ownership. Who builds the most thriving developer ecosystem. Who finds a sustainable balance between safety and openness. History keeps proving this： Smoot-Hawley， 1980s semiconductor curbs against Japan-short-term political wins， long-term economic self-harm. We're back at that crossroads. The second half of the AI race won't be about parameters. It'll be about cost structures and open ecosystems. Mark my words on that， lol.
AYi@AYi_AInotes · X美国AI已分裂为两条路径：美国闭源模型每百万token收费26-56美元，中国开源模型仅0.50-1美元，价差达50-100倍。OpenRouter数据显示，中国企业模型在美国企业的token使用占比已从2025年初的不足10%飙升至58%以上。若政策强制美国企业走闭源高价路线，将如同强制使用每桶800美元的石油，最终拖垮整个美国经济的AI成本结构。
Here's an unpopular take： What's really choking U.S. AI isn't Chinese models-it's the choices made by its own policymakers. Chamath's thread lays this out perfectly. AI has already forked into two non-overlapping paths： Path A： U.S. closed models， $26-$56 per million tokens. Path B： Chinese open-weight models， $0.50-$1 per million tokens. That's a 50-100x price gap-not a future projection， but today's reality. The OpenRouter data says it all： Chinese models' share of token usage among U.S. enterprises has surged from under 10% at the start of 2025 to over 58% now. Companies have already voted with their feet. If policy forces U.S. firms onto Path A exclusively， Chamath's analogy is spot-on： It's like mandating the use of $800-per-barrel oil when the market price is $80. Short-term， it may look like protecting domestic labs. Long-term， it torches the AI costs of the entire U.S. economy. SMEs， developers， startups-they'll be the first casualties. Deeper logic： AI is no longer a luxury-it's infrastructure， like electricity or oil. Force a 50x markup on "electricity，" and the efficiency of the whole economy collapses. The U.S. AI sector is a core valuation driver for the stock market. Once lab finances deteriorate， the contagion to tech equities and growth sectors will be faster than most expect. This is Chamath's real point： It's not about picking sides between Chinese or American models. The technological paradigm has shifted. The cost structure is irreversible. Policy that swims against this tide won't hurt rivals-it'll hurt itself. Here's my deeper cut： With Kimi K3 set to release its open weights on July 27， the path of open models plus extreme cost-efficiency will only accelerate. Any attempt to block it with policy will be steamrolled by the trend. The winning move over the next decade isn't who trains the strongest model first. It's who delivers sufficiently powerful capabilities at the lowest total cost of ownership. Who builds the most thriving developer ecosystem. Who finds a sustainable balance between safety and openness. History keeps proving this： Smoot-Hawley， 1980s semiconductor curbs against Japan-short-term political wins， long-term economic self-harm. We're back at that crossroads. The second half of the AI race won't be about parameters. It'll be about cost structures and open ecosystems. Mark my words on that， lol.