Fact CheckThe Offshore Weight Gambit
When Chinese labs like DeepSeek trained their models on heavily restricted, older hardware, what was the actual result?
They fell 3 to 5 years behind the current U.S. frontier models.
They resorted to stealing U.S. model weights via corporate espionage.
They achieved parity by burning up to 4 times more electricity and writing hyper-efficient code.
Correct. A spectacular policy backfire.