Hard Fork

Ezra Klein × Jensen Huang × Kevin Roose × Casey Newton

Jensen Huang calls AI safety a solvable engineering problem: labs should not ship systems they cannot contain, with liability rules and third-party audits still relevant.He argues NVIDIA compute could become a fungible, durable asset class, with one-gigawatt factories and annual rents reaching $50B.Open models have flipped to seventy-thirty, while supply-demand inversion and an uncertain digestion period remain risks.

Tekedra Mawakana

The US exchanged $8.9 billion of conditional CHIPS funding for 10% of Intel, potentially securing domestic chip capacity but risking politicized demand for an execution-challenged incumbent.Waymo targets around a million weekly rides by the end of next year, supported by safety data and expanding markets, while tariffs, regulation, labor opposition and fleet economics remain unresolved.

Kevin Roose × Casey Newton × Jordan Schneider

DeepSeek narrowed the perceived US-China model gap through organizational freedom and efficient engineering, but its 1.9 million iOS downloads, bans, and Microsoft-OpenAI distillation investigation also exposed security and data-use risks.Export controls still matter because China needs compute, while Operator frames OpenAI’s $200 price as digital labor; its 90% domain project contrasted with costly unreliability, privacy exposure, and unresolved agent-safety risks.

DeepSeek’s claimed $5.5 million V3 training cost and R1 release challenged frontier AI’s infrastructure moat, helping drive NVIDIA down about 18% and pushing its app to No. 1 in the US App Store.Cheaper inference could compress model pricing and margins, though falling costs may expand usage, existing chips remain useful for inference, and the hosts say DeepSeek’s figures deserve “a grain of salt.”