Anthropic CEO Dario Amodei Says Open-Weight AI Won’t Decentralize Power

In a recent exchange on X, Anthropic CEO Dario Amodei dismissed the notion that open-weight AI could decentralize power, arguing that advanced chips and compute infrastructure remain concentrated. During a discussion on TechCrunch Disrupt 2023, Amodei rejected the idea that policymakers are faced with a choice between centralized regulation or widespread distribution of AI.

He contended that the debate is not a false choice between concentrated power and widespread distribution, but rather about establishing equitable frameworks. Amodei emphasized that AI inherently tends to concentrate power due to the computational demands of scaling laws. He warned that distributing model weights does not change the dominance held by entities controlling the underlying chips and infrastructure.

Anthropic's CEO defended his company's policy track record, highlighting their support for measures like California's SB 53 and tiered evaluation frameworks proposed to the White House and the Center for AI Safety. Amodei also denied that safety warnings have negatively impacted public sentiment, instead viewing skepticism as a long-standing trust deficit among AI companies.

To address this, he noted that Anthropic is accelerating research in biology and medicine to deliver tangible clinical advancements. The debate between open-weight advocates and safety-focused frontier labs exposes a fundamental reality: while open weights allow for greater software accessibility, training and operating the most capable systems still require substantial compute, advanced chips, and access to large-scale infrastructure. This limits the extent to which open weights can decentralize power.