It is too early to evaluate this prediction.

Chamath is relaying Dario Amodei’s forecast that by around 2027 the cost to train a top‑tier, “good functional” frontier model could reach roughly $100B, up from “billions” in 2024.

As of the current date (November 30, 2025):

• Public analyses of training costs for leading 2023–2025 models (e.g., GPT‑4, Claude, Gemini) generally estimate single‑digit to low double‑digit billions of dollars at most for training plus associated compute build‑out, not ~$100B per model. These are rough outside estimates, as companies do not publish exact numbers.
• Dario Amodei and Anthropic have discussed rapidly rising compute and training costs and the need for massive capital (e.g., Anthropic’s multi‑billion‑dollar funding rounds and cloud commitments), but there is no credible public evidence that any single model training run has approached $100B yet.
• The prediction explicitly targets “by 2027”, and 2027 has not occurred; both hardware, algorithms, and business strategies could significantly change costs before then.

Because the time horizon of the forecast (around 2027) has not yet arrived, we cannot say whether the prediction was right or wrong. Therefore the correct status as of now is inconclusive (too early to tell).