Jensen Huang tells Ezra Klein that AI is just software whose existential risks are overstated, yet his standards would shut down OpenAI and 10x safety spending

If you are building AI, expect regulators and established players to demand extensive, costly safety testing before deployment.

Nvidia CEO Jensen Huang, in a podcast appearance, downplayed existential risks associated with AI, likening it to any other software advancement. He stated that AI would remain a "new abstraction level" and not fundamentally change reality. However, Huang also asserted that companies developing AI must invest heavily in safety, verification, and evaluation, suggesting that any AI labs failing to do so responsibly should be shut down. This stance, while seemingly emphasizing caution, was noted as potentially contradictory given his dismissal of existential threats. Huang also touched upon job market impacts, suggesting AI will create new industries and jobs despite potential displacement, attributing past manufacturing job losses primarily to outsourcing rather than technology.

Huang's insistence on extreme safety measures and the potential shutdown of labs that fail to meet unspecified standards introduces a significant new barrier to entry. This could disproportionately affect smaller AI startups and reinforce the dominance of larger, well-resourced companies capable of absorbing such rigorous and expensive compliance. His framework, if adopted, would significantly slow down the pace of AI development and deployment, prioritizing absolute safety over rapid innovation.

The contradiction between Huang's dismissal of AI's existential risks and his stringent safety demands signals a pragmatic, perhaps even protectionist, approach from a major hardware provider. By setting an impossibly high bar for software safety, Nvidia could indirectly favor its own hardware ecosystem, where it has greater control over development and deployment processes. This also positions Nvidia as a key arbiter of AI safety standards, influencing industry direction and potentially creating dependencies.