Does the export-control / geopolitical regime constrain the trajectory, or get routed around?

◇ Neutral question

A live variable, often mistreated as background.

◷ Share view, as of June 2026: confidence, rationales and priors are hidden. The questions, stances, claims, tests and evidence stay visible.

Stances · candidate answers

Q6-S6A

Export controls meaningfully slow Chinese frontier capability and bifurcate the market.

Held by US policy hawks, some semi analysts

Controls on leading-edge chips / tools create a real, compounding capability and cost gap.

would move me offChinese labs repeatedly reach frontier parity via domestic / smuggled hardware and efficiency gains.

Strongest opposing: Q6-S6B, Q6-S6C

Evidence behind its claims: 291 support · 52 contradict · 189 neutral

Q6-S6B

Controls are leaky and accelerate Chinese self-sufficiency; net harm to US vendors.

Held by some industry execs, China-watchers

Smuggling, domestic fab progress, and algorithmic efficiency route around controls while cutting US firms out of a huge market.

would move me offA durable, widening Chinese capability lag despite these efforts.

Strongest opposing: Q6-S6A, Q6-S6C

Evidence behind its claims: 291 support · 52 contradict · 189 neutral

Q6-S6C

Geopolitics is second-order; economics dominate either way.

Held by market analysts who think the commercial trajectory swamps policy

Real but bounded.

would move me offAn export-control or Taiwan event causes a structural supply-chain break that reprices the whole sector.

Strongest opposing: Q6-S6A, Q6-S6B

Evidence behind its claims: 144 support · 3 contradict · 67 neutral

The map: how claims bear on stances

supports contradictshover a node to highlight its links

Claims

6.1

Chinese frontier models lag US frontier by a widening margin.

Q6-S6AQ6-S6B

6.2

Algorithmic / training-efficiency gains reduce the compute (and chip-access) needed for frontier capability.

Q6-S6BQ6-S6A

6.3

Export controls cut US vendor addressable market more than they slow rivals.

Q6-S6BQ6-S6A

6.4

Government intervention authority (export controls, model-access suspensions, revenue-sharing mandates) is now exercised with sufficient frequency and unpredictability that US AI vendor China-addressable revenue must be modelled as structurally volatile, not cyclically recoverable.

Q6-S6AQ6-S6BQ6-S6C

Stances

Q6-S6A

Export controls meaningfully slow Chinese frontier capability and bifurcate the market.

held by US policy hawks, some semi analysts

Q6-S6B

Controls are leaky and accelerate Chinese self-sufficiency; net harm to US vendors.

held by some industry execs, China-watchers

Q6-S6C

Geopolitics is second-order; economics dominate either way.

held by market analysts who think the commercial trajectory swamps policy

Claims, each with its falsifiable test

6.1 · Chinese frontier models lag US frontier by a widening margin.

would falsifyChinese models reach / hold parity on trusted evals.

CapabilitySlow to resolve

6.2 · Algorithmic / training-efficiency gains reduce the compute (and chip-access) needed for frontier capability.

would falsifyFrontier capability stays tightly coupled to raw leading-edge compute access.

CapabilitySlow to resolve

6.3 · Export controls cut US vendor addressable market more than they slow rivals.

would falsifyUS vendor China revenue recovers despite controls.

MarketSlow to resolve

6.4 · Government intervention authority (export controls, model-access suspensions, revenue-sharing mandates) is now exercised with sufficient frequency and unpredictability that US AI vendor China-addressable revenue must be modelled as structurally volatile, not cyclically recoverable.

would falsifyNvidia China revenue variance across four consecutive quarters post-2025 exceeds variance in any other regional segment by a statistically significant margin, and analyst consensus China-revenue estimates are revised more than twice in a 12-month window.

MarketSlow to resolve

Bridge-claims connecting out

⤧ B3Training-efficiency gains will reduce aggregate compute demand (threatening "more compute = more value").Capability → Build-out

Q6 / 4 claims / 3 stances / 1 bridge·confidence moves are made manually; AI proposes only