As of November 30, 2025, there is not enough evidence to say Chamath’s long‑run structural prediction has clearly come true or is clearly false.

Why it’s too early to call:

• The claim is explicitly long term (“over time”) and strong in form (“crush the value of models to zero economically”). Less than two years have passed since the February 2024 podcast, which is short for judging industry‑structure outcomes of this kind.
• In that time, proprietary general‑purpose model providers have not seen their economic value approach zero. OpenAI’s annualized revenue hit about $10B by mid‑2025 with projections above $12B for 2025, and internal projections plus secondary sales talk value it in the hundreds of billions of dollars. (finance.yahoo.com) Anthropic likewise has reached multi‑billion‑dollar run‑rate revenue and raised at valuations around $60B and then ~$180B in 2025. (sacra.com) That is the opposite of “near zero” economic value so far, but it doesn’t rule out later commoditization.

Evidence on open source vs proprietary models:

• Open‑source models have advanced rapidly. A 2025 benchmark finds the best open‑source LLMs are now only single‑digit points behind the top proprietary models on quality, while being ~7–8x cheaper per token and often faster—clear support for the technical and cost side of commoditization. (whatllm.org)
• However, a recent study cited by the Linux Foundation reports that open models account for roughly 20% of usage but only about 4% of AI‑model revenue; enterprises still overwhelmingly pay for closed‑source APIs due to trust, compliance, and switching‑cost advantages. (itpro.com) That means open source has not yet eroded the bulk of model‑provider economics, even if it is exerting price pressure at the margin.

Evidence on “picks and shovels” and proprietary data owners:

• The “picks and shovels” part of his thesis is strongly supported so far. Nvidia’s market cap has exploded into the multi‑trillion range on the back of AI‑data‑center GPUs, which now constitute the vast majority of its revenue, and hyperscalers are driving unprecedented AI capex. (markets.financialcontent.com) OpenAI’s own long‑term infrastructure deals (e.g., massive, multi‑hundred‑billion‑dollar cloud and data‑center commitments with Oracle and partners under the Stargate project) underline how much value is accruing to compute and data‑center providers. (group.softbank)
• Data‑rich incumbents are indeed monetizing proprietary content with AI (for example, Thomson Reuters’ AI‑enhanced legal and tax products contributing to solid organic revenue growth in its core segments), but the scale of value captured here is still much smaller than that at Nvidia, cloud hyperscalers, or the leading model labs. (reuters.com) It’s directionally consistent with Chamath’s view but not yet clearly “most” of the sustainable gains.

Net assessment:

• Central strong claim (“open source will drive the economic value of general‑purpose model providers to near zero”) is not borne out so far: model providers are currently among the most valuable and fastest‑growing companies in the sector.
• Supporting sub‑claims (infra/picks‑and‑shovels win big; proprietary data is valuable; open source compresses prices and utility goes up) are partially supported by current evidence.
• Because the prediction concerns the eventual industry structure and gives no explicit time horizon, the present data can’t definitively validate or falsify the end‑state he describes. Hence, the fairest classification today is **“inconclusive (too early)” rather than clearly right or wrong.