By November 30, 2025, the key conditions in Chamath’s prediction have not yet been reached, and the target year (~2027) is still in the future, so it’s too early to judge.

1. Time horizon and cost condition
Chamath’s claim is conditional and time‑bound: if frontier model training costs rise toward ~$100B by ~2027, then only mega‑cap firms (Google, Microsoft, Meta, Amazon) will be able to fund that level, and startups like Ilya Sutskever’s Safe Superintelligence Inc. (SSI) won’t be able to raise comparable capital. As of 2025, available analyses of frontier training costs put even the largest projected single‑model training runs in the $1B+ range by around 2027, not $100B; today’s actual runs are far below that. (arxiv.org) There are speculative forecasts of a possible “$100B model” by ~2027, but these are forward‑looking projections, not realized costs. (getcoai.com) Since 2027 has not arrived and no $100B‑scale training run has occurred, the antecedent of his conditional has not clearly been met.

2. Capital available to non‑mega‑cap AI labs so far
Sutskever’s SSI has raised on the order of $3B across rounds (about $1B in Sept 2024 and ~$2B by early/mid‑2025), at valuations up to roughly $30–32B, with strategic backing from major investors and Alphabet/Nvidia for compute. (app.dealroom.co) That is significant but nowhere near $100B in deployable capital for a single model‑training effort.

Other leading AI startups not inside the big‑four cloud companies—most notably Anthropic—have raised much larger sums than SSI (e.g., a $13B Series F in 2025, bringing its valuation to about $183B, on top of earlier multibillion equity rounds and large cloud/credit facilities from Amazon and others). (cnbc.com) Still, even Anthropic’s total capital base is far below the kind of $100B single‑run spend Chamath is talking about, and much of its effective compute budget is intertwined with mega‑cap partners’ infrastructure (AWS, etc.), which somewhat supports his broader point that true frontier‑scale capex is dominated by the hyperscalers.

3. Structural trend vs. empirical verdict
Research and industry reporting do indicate that frontier AI development costs and required capex are rising extremely fast, with multiple analyses arguing that only a handful of very well‑funded organizations will be able to finance the very largest training runs over the next several years. (arxiv.org) However, whether this dynamic fully excludes independent labs like SSI at the $100B‑per‑model level by 2027 is still speculative—no one has yet attempted or financed such a run.

Because:

• The forecast year (~2027) has not arrived,
• No ~$100B frontier training run has actually occurred, and
• We cannot yet observe whether startups categorically fail to access that scale of capital when/if it’s needed,

the prediction cannot presently be called right or wrong. It remains unresolved, hence the classification "inconclusive" (too early).