4 quotes from 3 episodes on Dwarkesh Podcast, each with a timestamped link to the source.
4 quotes3 episodes
The short version
Dwarkesh Patel argues that physical hardware availability constrains AI compute scaling, even as agent swarms find ways to train models faster. ASML builds fewer than 100 extreme ultraviolet lithography tools each year, restricting global wafer output growth to just 2x annually.
Most interesting insights
A 50% API price increase at 200,000 tokens exposes memory bandwidth restrictions inside the physical cluster hardware.
“With longer context, Gemini 3.1 is 50% more expensive if you go over 200k tokens than if you're below 200k tokens.”
Labs will distribute daily training runs to agent swarms
Swarms of agents test smaller changes to bypass physical hardware constraints. By late 2027, top companies could provide enough daily compute for 10,000 individual agents to each run GPT-3 scale experiments.
“And if this trend continues, then by the end of 2027, OpenAI could almost give 10,000 agents, so the amount that were involved in solving the Millennium Prize problem, each enough compute to run a GPT3 size training run every single day.”
Dwarkesh Patel, Dwarkesh Podcast · November 2026 · Watch at 40:39 ↗
Dwarkesh Patel forecasts that automated AI researchers capable of matching the top human researchers will arrive by 2028, unlocking recursive self-improvement.
Long context windows spanning multi-month working memory remove the need for constant architectural weight updates to achieve continual learning.
Gemini 3.1's 50% price jump for context lengths over 200,000 tokens isn't arbitrary; it signals a hard constraint on memory bandwidth, not just compute, in underlying hardware.
Input (prefill) tokens costing up to 5 times less than output (decode) tokens indicates that the decode phase of LLM inference is heavily bottlenecked by memory bandwidth, not floating-point operations.
How we attribute quotes. Every quote was matched against the episode transcript, so the words and the timestamp are real (we trim filler words like "um", nothing else). The name comes from our written summary of the episode, and we use it only when a separate check of the captions finds that person on the episode. YouTube gives us no voice-by-voice transcript, so open the timestamp to hear who is talking. See a wrong name? Tell us and we fix or remove it.