Dylan Patel: Sources & History

Hopes for slower AI progress to preserve a balance of power, while opposing lab collusion and stressing that China is not slowing. [1]

Initial statement. Initial reviewed placement, preserving the scope and qualifications of the attributed recommendation.

Source [1] · Interview transcript

Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028

Attributed to: Dylan Patel · Dwarkesh Podcast

Published 25 Aug 2026

“…or we have governments slow everything down and people slow everything down, and you have a slowdown of progress somehow hopefully, and there is more of a balance of power.”

01:07:52 section, Dylan Patel’s response about centralization and balance of power · View original source ↗

Source context. Dylan contrasts extreme concentration of resources with governments and people slowing progress, expressing hope for the latter and a greater balance of power. The surrounding dialogue discusses slow takeoff and centralization. His preference is distinct from his forecasts. The interviewer’s suggestion that internal-use restrictions would be ideal belongs to Dwarkesh Patel, not Dylan. Publisher transcript inspected; recording date not established and audio not independently checked.

Scope: Progress toward AGI, ASI and recursive self-improvement in the context of frontier labs concentrating compute and power.
Conditions: No detailed policy instrument or numerical rate specified; opposes lab collusion and stresses that China is not slowing.
Stated rationale: Avoid extreme concentration of compute and power in a single company.

Source details

primary · Original publisher text inspected during research on October 20, 2026. Cached original transcript retained; audio not independently checked.

Discovered 20 Oct 2026.

Accessed 20 Oct 2026.

Editorial assessment. A stated hope for governments/people to slow progress is an endorsed desired direction, even without a detailed policy instrument. Does not justify Pause or claim general anticompetitive cooperation acceptable. Later China analysis limits feasibility, not plainly retracting personal hope. His August 2 post (https://x.com/dylan522p/status/2084051253278208138) objects to lab coordination to slow AI as anticompetitive collusion. This is his policy/legal opinion, not an independently certified legal finding; it limits acceptable mechanisms without erasing the later slower-progress preference. His October 8 coauthored analysis, Beijing Will Not Pace the Frontier (https://newsletter.semianalysis.com/p/beijing-will-not-pace-the-frontier), argues China prioritizes rapid development and lacks binding frontier constraints. That analysis limits coordinated-slowdown feasibility; neither its title nor quoted Chinese officials establish Dylan’s personal acceleration endorsement. A purported pause/efficiency reply found through a mirror remains unverified and is excluded.

Dylan Patel

PACE