Is AI a Bubble? · History
History
Earlier assessments, explanatory pages and claim-specific context—preserved at their original addresses.
These pages preserve the earlier assessments and explanatory material at their original addresses. They are dated context, not prerequisites for the current answer. A historical label does not invalidate every finding inside a report; consequential corrections and later-period differences are identified here.
Historical pages
Historical research reports
The complete Research catalogue lists these four and both current investigations together, with descriptions and data links.
Amazon: the missed disclosure was already corrected
The earliest baseline left Amazon’s announced OpenAI funding unverified. Cash, commitments and capacity subsequently inspected the June-quarter filing and reported $15bn invested in the first quarter, $13.7bn in the second and $21.3bn after June. The filing had already been available when the earlier baseline was researched; this was an avoidable evidence-coverage omission, not money newly paid at the date of the correction. Amazon investment note.
The current assessment retains that earlier correction. It narrows payment uncertainty but does not establish every other investor’s settlement, recipient-side bank verification or sustainable end-customer cash.
CoreWeave: preserve the March schedule, use June with care
| Snapshot | Total principal | Scheduled through 2027 | Share |
|---|---|---|---|
| 31 March 2026 | 25.149 | 11.718 | 46.6% |
| 30 June 2026 | 35.551 | 10.597 | 29.8% |
The June share is smaller, but that does not establish comparable risk reduction. The remaining calendar-2026 window shrank, borrowing and scheduled obligations changed, and the denominator grew. Neither schedule is a complete September balance. The older report’s contract comparisons remain optional dated context; the current assessment supplies the June table and bounded later-event discussion. Preserved March timetable; June filing; current explanation.
NVIDIA: historical caution remains appropriate
The earlier financing report treated the prepaid forward as a contractual obligation without claiming a verified receipt. The stronger characterization in the underlying whole-system research was later withdrawn; it was not carried as verified receipt in the reader account. The current account is:
NVIDIA entered into a $1.5bn prepaid-forward contract with Energy Global, LP. The agreement requires payment within three business days after its date. The documents inspected for this investigation do not verify receipt.
That is not a finding of nonpayment. The forward is separate from the offering-linked share subscription and contingent guarantee. Documentary correction and source details.
A forecast retains its vintage
The earlier baseline retained the IEA’s April 2025 base case of approximately 945 TWh of global data-center electricity consumption in 2030. The current whole-system account uses the April 2026 update: an estimated 485 TWh for 2025 and a projected 950 TWh for 2030. These are global data-center boundaries, not AI-only current consumption or proof of profitable capacity. Earlier baseline and source; April 2026 outlook.
Read the dated report without inheriting its coverage limits
The old Method page’s statement that no security valuation model had been completed describes its own evidence boundary. The current assessment includes conditional reverse-valuation and investment-recovery exercises, with important limits, rather than a comprehensive security appraisal. Current return models.
Likewise, the current trailing-year cash comparison does not invalidate the older half-year comparison: the periods overlap. Company results and productivity studies were already present in the earlier baseline. Current explanations connect them more fully without pretending that usefulness or supplier profits were newly discovered.