The prediction is explicitly about what will happen “over the next several years”, but we are less than half a year past the July 4, 2025 podcast date, so the forecast horizon has not elapsed.

Regulatory structure is still unsettled:

• States are rapidly enacting their own AI laws: in 2025, state lawmakers introduced 1,134 AI‑related bills and enacted 131 laws across 40 states, creating a patchwork of state rules. (ccianet.org)
• At the same time, there is an active federal push to preempt state AI laws (e.g., House bills including 10‑year preemption, and a White House draft executive order to challenge state AI regulations), while state AGs and groups like NCSL are lobbying hard against preemption. (nfcommission.gov)
  Because this federal–state fight is unresolved, we don’t yet know whether the medium‑term U.S. regime will truly be “predominantly state‑level, non‑preempted” as the prediction assumes.

Impact on startups vs incumbents is also not yet empirically clear:

• California’s SB 53 (the first major “frontier AI” safety law) and related AI bills are widely seen as positioning California as a lead U.S. regulator of AI, but SB 53 is narrowly scoped: it applies only to “large frontier developers” with models above 10^26 FLOPs and >$500M in annual revenue, deliberately excluding smaller developers. (lw.com)
• Critics from venture capital and industry (e.g., Andreessen Horowitz) argue that a 50‑state patchwork could make compliance costs unmanageable for startups and entrench big players, explicitly warning that SB 53 and similar state‑by‑state rules may “squeeze out startups” and “entrench the biggest players.” (forbes.com)
• However, supporters of SB 53 and academic/policy analysis emphasize that its obligations fall almost entirely on a small set of frontier labs and are not designed to burden smaller startups, and there is not yet systematic evidence that AI startup formation, funding, or growth has materially slowed because of these new state laws. (sb53.info)

Since:

1. the regulatory regime Chamath is describing is still being contested (federal preemption vs state autonomy), and
2. the concrete, measurable long‑term impact on AI startups and incumbents has not yet had time to materialize or be robustly measured,

it is too early to say whether his multi‑year prediction is right or wrong. Hence the result is inconclusive at this time.