By November 30, 2025 we are ~16 months after the July 12, 2024 podcast, so only part‑way through Friedberg’s stated 24–36 month horizon for “extraordinary ROI” on the AI infrastructure buildout. The core of the prediction is thus still within its own time window.

On the evidence side, there are strong positive signals but not conclusive proof:

• Vendors supplying AI infrastructure, especially Nvidia, are generating enormous, highly profitable data‑center revenues driven by AI demand (tens of billions per quarter with very high margins, plus large backlogs for AI GPUs). These numbers show that at least some parts of the AI infrastructure stack are producing very strong financial returns.【2search1】【2search5】
• Hyperscalers (Microsoft, Amazon, Alphabet, Meta) have pushed capex heavily into AI data centers and GPUs—analysts estimate hundreds of billions in AI‑related capex, with cloud and AI services revenues growing rapidly. However, this spending is often framed as a long‑term strategic bet, with revenue growth still catching up to capex growth.【1search0】【1search5】

At the same time, multiple broad surveys and analyst reports indicate that AI ROI is not yet clearly or uniformly “proven” across the wider economy:

• Gartner and CIO‑focused coverage describe 2024–2025 as a "build‑out" or foundation‑laying phase where CIOs are increasing AI budgets substantially, but they emphasize that the ROI of many AI projects remains uncertain even as spending accelerates.【1search1】【1search7】
• CloudZero’s 2025 “State of AI Costs” report finds that only about half of organizations say they can confidently evaluate AI ROI, and that a large share of AI cloud spend is wasted on idle or poorly optimized resources.【1search6】 A separate analysis similarly notes AI costs rising sharply while ROI remains unclear for many firms.【1search4】
• Credit and bond‑market commentary on hyperscaler AI capex highlights that investors are effectively betting that future AI revenues will justify massive, debt‑funded build‑outs, but explicitly warns that a slowdown in AI adoption could leave companies overextended—i.e., the economic payoff is still a forward‑looking assumption, not yet a settled fact.【1news14】

Putting this together:

• The 2–3 year horizon has not yet elapsed, so we cannot fairly judge the ultimate claim that the AI infrastructure buildout will be “extraordinarily” ROI‑positive by mid‑2026 to mid‑2027.
• Within the nearer “next couple of quarters” sub‑claim, we do see strong profitability for chip vendors and some cloud segments, but broad, demonstrable proof that the overall AI infrastructure buildout is economically justified for hyperscalers and enterprises is not yet clearly established in the aggregate data and commentary.

Given that the primary window of the prediction is still open and the available evidence is mixed and evolving, the fairest assessment as of November 30, 2025 is “inconclusive (too early)” rather than clearly right or clearly wrong.