The prediction hinged on the idea that the New York Times–Amazon AI licensing deal in mid‑2025—reported at $20–25M per year—was the peak of this type of training‑data/content‑licensing deal, and that comparable deals signed in future years would see per‑year dollar values “only go down from here.” (wsj.com)

What we can see as of 30 Nov 2025:

1. Size of the NYT–Amazon deal
   Multiple reports (summarizing the original WSJ scoop) say Amazon will pay The New York Times $20–25M annually under a multi‑year AI content‑licensing agreement for training models and powering Alexa and related products. (wsj.com) This is the specific deal Chamath was reacting to.

2. Other large AI–publisher deals already in market
   Before this NYT–Amazon deal, there were already larger AI training‑licensing arrangements on a per‑year basis:

   • News Corp–OpenAI: widely reported at >$250M over 5 years (≈$50M/year). (allmo.ai)
   • Reddit–Google & Reddit–OpenAI: coverage of Reddit’s filings and press reports imply ~$60–70M per year from OpenAI and ~$60M/year from Google for data licensing. (allmo.ai)
   • Dotdash Meredith–OpenAI: estimated fixed component around $16M/year. (allmo.ai)
     So even at the time of the podcast, $20–25M/year was not the highest per‑year figure in the ecosystem, but the user’s normalized prediction focuses on the future trend from that point.

3. Deals signed after Aug 1, 2025
   Since the podcast date, there have been new AI content‑licensing deals, but most have undisclosed financial terms:

   • Perplexity–Gannett / USA TODAY Network (Perplexity Publisher Program) – terms not disclosed. (businesswire.com)
   • Gannett–Microsoft AI licensing agreement via Microsoft’s Publisher Content Marketplace – announced in Gannett’s Q3 2025 results, but again no dollar figures given. (businesswire.com)
   • Perplexity–Getty Images multi‑year visual‑content licensing deal – no public pricing. (futureweek.com)
     Survey pieces that catalog the training‑data market as of October 2025 list the major known deals (News Corp, Reddit, Dotdash Meredith, NYT–Amazon, etc.) and explicitly note that only a few dollar figures are public and most contracts are confidential. (allmo.ai)

4. No post‑August 2025 deal with a clearly higher disclosed per‑year value
   In the public record through November 2025, no newly announced publisher/AI training‑data deal after Aug 1, 2025 has a disclosed annual payment that clearly exceeds the NYT–Amazon $20–25M/year range. However, because most newer deals do not publish financial terms at all, this absence of evidence is not strong evidence that the underlying economics are actually lower.

5. Time horizon and data limitations
   Chamath’s statement was about “from here” going forward and “future years”. Only about four months of dealflow have elapsed since the podcast, and the small number of post‑August deals with confidential pricing makes it impossible to robustly determine a trend in per‑year license values. Industry overviews emphasize that the market is still evolving, with a wide spread from single‑digit millions per year to nine‑figure multi‑year packages, and that public numbers should be treated as directional only. (allmo.ai)

Why the result is "inconclusive" rather than right or wrong:

• To decisively call the prediction wrong, we would need at least one post‑Aug‑2025 training‑data licensing deal, clearly comparable in scope, with a disclosed annual value above ~$25M/year. No such publicly documented example exists yet.
• To call it right, we would need evidence that subsequent comparable deals are consistently priced below that level (or that $20–25M/year is clearly a ceiling). Because most newer contracts do not report financials, and the observation window is short, we also lack that evidence.

Given the short time since the prediction and the opacity of deal terms, the direction of per‑year pricing for future AI training‑licensing deals cannot yet be determined from available data, so the outcome is best classified as inconclusive (too early to tell).