Get your free personalized podcast brief
We scan new podcasts and send you the top 5 insights daily.
Sebastian Malabai argues that U.S. chip export bans are ineffective because China circumvents them by renting GPU capacity in other countries and using "distillation" to reverse-engineer and copycat advanced U.S. models. This suggests a need for a new strategy focused on collaborative safety.
Related Insights
The performance gap between US and Chinese AI models may be widening due to second-order effects of chip controls. By limiting inference at scale, the controls reduce the volume of customer interactions and feedback Chinese firms receive. This starves them of the data needed to identify and patch model weaknesses on diverse, real-world tasks.
Despite impressive models from companies like DeepSeek, China's AI ecosystem is heavily reliant on "distilling"—essentially copying and refining—open-source models from the US. This dependency on an external innovation engine is a major weakness in their national strategy to achieve genuine AI leadership and self-sufficiency.
Even if Chinese firms use "distillation" to steal capabilities from US models, the process is computationally intensive. Restricting access to advanced chips thus limits direct training *and* makes large-scale IP theft more difficult.
US officials and AI labs allege Chinese firms are engaged in industrial-scale IP theft. They reportedly use fraudulent accounts to extract capabilities from US models like Claude to train their own, creating a facade of domestic innovation.
Contrary to their intent, U.S. export controls on AI chips have backfired. Instead of crippling China's AI development, the restrictions provided the necessary incentive for China to aggressively invest in and accelerate its own semiconductor industry, potentially eroding the U.S.'s long-term competitive advantage.
Faced with limited access to top-tier hardware, Chinese AI companies have been forced to innovate on model architecture to compete. They've developed superior techniques in memory management and multi-token prediction, making their models highly efficient and formidable competitors despite hardware constraints.
China's superior ability to rapidly build energy infrastructure and data centers means it could have outpaced US firms in building massive AI training facilities. Export controls are the primary reason Chinese hyperscalers haven't matched the massive capital spending of their US counterparts.
Chinese firms are closing the AI capability gap by using "distillation" to replicate the intelligence of leading US models. This creates a strategic vulnerability, as copying software models is easier than replicating China's hardware manufacturing prowess.
The effectiveness of US export controls on advanced AI chips stems from a deep technological gap. According to China's own projections, it won't be able to domestically produce chips as powerful as those the US is restricting until 2028, creating a significant and lasting strategic advantage for democracies.
U.S. export controls on advanced semiconductors, intended to slow China, have instead galvanized its domestic industry. The restrictions accelerated China's existing push for self-sufficiency, forcing local companies to innovate with less advanced chips and develop their own GPU and manufacturing capabilities, diminishing the policy's long-term effectiveness.
