Josef Chen

The space-heater ledger

opened 13 Aug 2026 · updated 16 Aug 2026 · 0 prints since publication

Public scorekeeping for the five predictions in The AI Data-Centre Bust Will Look Like a Boom. Disclosures only, no positions. Entries that cut against the call are logged the same way as entries that support it, and the grading conditions were fixed when the essay was published. Where a call needs a measure nobody yet publishes, the measuring source is named here before it is used; a call with no public measure by 2030 grades unresolved, not true.

The five calls

  1. 1The six-year GPU book will not survive the decadeopen

    Grades true when: A hyperscaler or a major neocloud discloses a useful life of four years or fewer for accelerators before 2030.

  2. 2AI will get its hashpriceopen

    Grades true when: A continuously quoted rate for a named hardware generation is cited in a credit agreement before 2030.

  3. 3Custom silicon takes the payroll and the GPU keeps the frontieropen

    Grades true when: By 2030 the majority of the largest platforms' own stable, high-volume inference runs on chips they control, while merchant GPUs keep the changing work.

  4. 4The secondary market prints the number everyone is avoidingopen

    Grades true when: A liquid secondary market quotes used accelerators, and the first large GPU impairment lands in the same quarter as record industry demand.

  5. 5Most neoclouds will not die as cloudsopen

    Grades true when: Surviving neoclouds re-describe themselves around power and buildings; the rest are absorbed for contracts and megawatts.

Nothing has printed since publication. NVIDIA reports in late August; the hyperscalers and CoreWeave file for the September quarter between October and November. Useful-life changes usually surface in annual filings in the new year.

Recorded before publication

The filings the essay itself cites, logged so the ledger starts honest rather than empty.

  • 2025Amazon shortened a server and networking subset from six years back to five, citing faster AI/ML change; the reversal added $1.4 billion to 2025 depreciation. Amazon 10-K
  • Feb 2026CoreWeave disclosed a six-year technology-equipment life beside an about-five-year weighted-average committed customer term; the earning life of the same fleet appears nowhere. CoreWeave 10-K
  • Aug 2026Nebius said pricing for older GPUs improved more than 30% from the prior quarter, with expected payback of about twenty-two months. Nebius Q2 shareholder letteragainst
  • Aug 2026CoreWeave's credit stack priced the tail: $8.5 billion at investment grade against contract-wrapped GPUs in March, then $2.6 billion below investment grade on a roughly five-year loan against contracts averaging about three years, with no residual-value guarantee in the filed agreement. CoreWeave 8-K
  • Mar 2026SemiAnalysis's H100 rental index, distributed on Bloomberg: one-year contract rates rose nearly 40% from their October 2025 low, and on-demand capacity has been sold out since February. SemiAnalysis H100 indexagainst
  • Aug 2026NVIDIA's financing post made residual value the official pitch: a support mechanism for up to 25% of an opportunity, the factory described as flexible and fungible, the A100's economic life stretched toward a decade, and the one-year H100 rental series quoted by the vendor itself. NVIDIA financing post

Disclosed useful lives

Every disclosed change the record carries, oldest first within each group. Management estimates for unlike asset classes; none isolates a GPU.

CompanyEquipmentyearseffective
MARAmining rigs5 to 32023
Riotmining rigs2 to 32024
CleanSparkmining rigs5 to 32024
MicrosoftAI infrastructure4 to 62022
AlphabetAI infrastructure4 to 62023
CoreWeaveAI infrastructure5 to 62023
Amazon serversAI infrastructure5 to 62024
Amazon subsetAI infrastructure6 to 52025
MetaAI infrastructure5 to 5.52025

The ledger updates when the footnotes do. Back to the essay.