Scope of the prediction Chamath was claiming that frontier‑model training at the "hundreds of billions of dollars" scale would not be funded by traditional venture capital across multiple companies; instead that kind of capital would come from a small set of mega‑cap incumbents (and similar balance‑sheet players).

What actually happened (mid‑2024 to late‑2025)

1. Hundreds‑of‑billions scale spending is coming from mega‑caps and similar, not VC funds.

   • Analysts project that hyperscalers like Alphabet, Amazon, Meta, and Microsoft will collectively invest around $1.7 trillion in AI‑related infrastructure by 2035, with capex in 2024 already about $253 billion and heavily focused on AI data centers and compute. (barrons.com)
   • Meta’s Mark Zuckerberg has explicitly said Meta will spend “hundreds of billions of dollars” on AI over time, and Meta has guided $60–65 billion of capex in 2025 largely for AI servers and data centers—funded from its own cash flow and debt, not from VC. (theguardian.com)
   • Microsoft, Alphabet, Amazon and others are likewise budgeting tens of billions per year each for AI‑related capex (data centers, GPUs, networking), again from operating cash flow and corporate financing. (medium.com)

2. The flagship “$100B+” AI infra project is a corporate/sovereign joint venture, not a VC‑funded startup.

   • Stargate LLC—a joint venture among OpenAI, SoftBank, Oracle and MGX—was launched with $100 billion in initial capital and plans to invest up to $500 billion in AI infrastructure by 2029. Its financing structure is dominated by SoftBank, corporate owners, and debt, not conventional VC funds raising LP capital. (en.wikipedia.org)

3. Venture capital is large and growing, but still 1–2 orders of magnitude smaller than the "hundreds of billions per company" level.

   • Global VC investment into generative AI was about $45–56 billion in 2024 and $49.2 billion just in H1 2025. That’s spread across the entire sector, not concentrated as hundreds of billions into individual labs. (ey.com)
   • Even the very largest startup rounds—Databricks ($10B), OpenAI ($6.6B), xAI ($6B), Anthropic ($4B), Waymo ($5B), Mistral (~$2B)—are single‑digit billions each, not anywhere near hundreds of billions. (kpmg.com)
   • Anthropic’s big 2025 round was $3.5B Series E led by Lightspeed and other VCs, bringing total capital raised to about $18.2B—a very large VC‑backed company, but still far from the “hundreds of billions” scale. (anthropic.com)

4. Pattern: mega‑caps & quasi‑sovereign capital dominate the extreme scale; VC plays at much smaller though still multi‑billion levels.

   • Industry analyses emphasize that the multi‑trillion‑dollar AI infrastructure build‑out is being financed primarily by hyperscalers out of operating cash flow and corporate balance sheets, not “venture speculation.” (medium.com)
   • VC investors have become more concentrated, writing bigger checks to a few winners, but the entire global generative‑AI VC flow per year is still at the tens of billions level, not hundreds of billions into individual companies.

Assessment relative to the prediction As of November 30, 2025:

• The only actors planning or executing $100B+ AI infrastructure projects are mega‑cap tech companies and corporate/sovereign joint ventures (e.g., Stargate), not traditional VC funds.
• Venture capital is heavily involved in AI, but at the 1–10B per company scale, not “hundreds of billions of dollars” to multiple independent frontier‑model labs.

Given this, Chamath’s structural claim—that VCs would not be the ones funding hundreds of billions into multiple AI companies for frontier‑model training, leaving that role to a small set of mega‑cap incumbents and similar balance‑sheet players—matches how the market has actually evolved so far.

Therefore the prediction is right based on evidence up to late 2025.