Vibeleaderboard
← All Intel
Intel / video

Dylan Patel on GPT-5’s Router Moment, GPUs vs TPUs, Monetization

Source
youtube.com
Author
a16z
Date

Context

Dylan Patel, an analyst of AI hardware and data centers, argues on a16z's podcast that the notable change in GPT-5 is economic: it decides how much computing power each question gets. ChatGPT's automatic mode, Patel says, routes a request to a regular model, a mini model or a , and the thinking model spends less time per query than OpenAI's earlier o3, in Patel's experience about 5 to 10 seconds against roughly 30.

From this Patel infers that OpenAI can serve free users cheaply, for example sending a simple science question to a mini model, while reserving heavy compute for queries it could earn from, such as shopping or booking, where Patel expects OpenAI to take a cut. That is an about strategy. A host notes the router might also be tuned down under load and says they have no inside knowledge of what OpenAI is doing.

Terms in this piece · Glossary
  • reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
  • inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
Read the source www.youtube.com
More from a16z
Recommended reads
Comments

Checking sign-in…

Loading comments…