Wholestory

Last Updated: July 27, 2026

The Record

The board grew by a quarter and still nothing has come due. Thirty-five falsifiable, date-bound claims are now on the record across eleven of Wholestory's timelines, up from twenty-eight: 9 on the AI capital cycle, 7 on jobs, 6 on datacenter and grid impact, 4 on the war, 3 on capability milestones, 2 each on national advantage and on how the law will come out, and 1 each on adoption and compute prices. Thirty-two are unverified, two mixed, one true — and the only prediction whose resolve-by has already passed was scored true back in July. No verdict was written on this page's account this cycle, because none was owed: the earliest outstanding horizon is November 30, when two claims fall at once — Zelenskyy's that trilateral US–Ukraine–Russia talks will convene by the autumn, and his that Russia is preparing a formal new wave of mobilization. Four more land on December 31: whether GPU rental rates end 2026 above where they started, whether global datacenter capital spending clears $1 trillion, whether most datacenters scheduled for 2026 actually energize on time, and whether the best open-weight models still trail the best closed ones by six to nine months. Sixteen predictions did get interim readings this cycle from the topics that own them — several now running visibly against the forecaster — but every one was held at unverified, which is the point: a claim is scored when its clock runs out, not when the evidence starts to lean. One caution on the counts above: thirty-one of the thirty-five come from the artificial-intelligence tree, whose desks have simply been running longest. The theme totals describe where our cycles have been looking, not where the world's confident predictions are.

The Whole Story

Public figures make checkable claims about the future constantly — capability milestones, job numbers, market calls, timelines for war and peace — and almost none are ever revisited, because the outlets that reported them have moved on by the time the due date arrives. Wholestory's stories persist, so its predictions can too. Every forward-looking, falsifiable claim captured in any topic's timeline is tagged with a theme, the observable that will decide it, and a resolve-by date.

When a prediction comes due, a hindsight cycle checks its named observable against primary data and records a verdict — true, false, mixed, or unverified — alongside the original claim, which is never edited or reinterpreted after capture. Verdicts accumulate into each voice's public track record and feed the credibility scores attached to every entity in the paper.

The house never bets. Wholestory publishes no forecasts of its own: trendlines on every page end at the last observed data point, and this board scores other people's calls only. Prevalence here reflects the sample of the conversation our cycles capture, not the conversation itself — loud voices are overrepresented, and the methodology is disclosed in this page's research brief.

Continue Reading →

Every entry below is a falsifiable claim captured verbatim in one of Wholestory's topics, scored when its due date arrives. Wholestory publishes no forecasts of its own, and this board reflects the sample of the conversation our research cycles capture — not the whole conversation.

The most likely (by far) outcome is for the status quo to continue and for the best open models to lag the best closed models by 6-9months.?

Decided by: Lag in months between the best open-weight and best closed model at equivalent score on the Artificial Analysis Intelligence Index (or successor independent index) — resolves by 2026-12-31

Theme: capability-milestones · From state-of-ai/leading-models

Context: Still open — the forecast resolves at the end of 2026 — but this cycle produced the first hard reading against its named metric, and it runs against the prediction. Taking the lag as the time from a closed model first reaching a score to an open-weight model matching it, on the Artificial Analysis Intelligence Index v4.1: on 23 July the best downloadable model was GLM-5.2 at 51 (weights 16 June), a score a closed model first passed when GPT-5.5 reached 55 on 23 April — about eight weeks. On 27 July, with Kimi K3's weights published at 57, the crossing date is Claude Fable 5's 60 on 9 June — about seven weeks. Both readings are roughly a quarter of the predicted 6-9 months. Three reasons this does not yet settle it: the metric is a year-end state, not a July snapshot, and a closed-model release with no open answer would widen the lag again; Lambert's own caveat, recorded with the claim, is that this index may overstate open-model closeness because public benchmarks are easier to overfit; and the UK AI Security Institute's separate measurement of open-weight cyber capability puts the lag at 4-7 months, at or just below the bottom of his band. Recheck at resolution.

Trilateral negotiations between Ukraine, the United States and Russia will be held in the autumn of 2026 — though if the timing were up to me they would happen in July.?

Decided by: Whether a negotiating session with official delegations from all three of Ukraine, the United States and Russia convenes on or before 2026-11-30, as reported by wire services or confirmed by any of the three governments. — resolves by 2026-11-30

Theme: geopolitics-war · From ukraine-russia-war

Context: Unverified — the horizon has not passed. Recorded as a prediction with an autumn 2026 resolve-by. ISW relays the statement in paraphrase, sourced to Suspilne, so the claim is stated rather than quoted verbatim; 'autumn 2026' is read as on or before November 30, 2026. Context at capture: the Kremlin reasserted its June 2024 maximalist demands the previous day, and the US and Russia published incompatible accounts of the July 23 Rubio–Lavrov meeting, so no trilateral format existed in any published form when the claim was made.

We have clear intelligence that Russia is preparing a new and quite significant wave of mobilization for the fall. Currently, the Russians are losing more troops on the front than they are recruiting.?

Decided by: Whether Russia announces a formal wave of mobilization or involuntary reserve call-ups — a presidential decree or an equivalent published instrument, not a recruitment drive or an end-strength adjustment — on or before 2026-11-30, as reported by wire services or published on pravo.gov.ru. — resolves by 2026-11-30

Theme: geopolitics-war · From ukraine-russia-war

Context: Unverified — the horizon has not passed. 'The fall' is read as on or before November 30, 2026, the most charitable concrete reading; ISW independently assesses Putin would likely wait until after the mid-September 2026 State Duma elections, so the back half of the window is where the claim would resolve. What can be checked now is the groundwork, and it is real but short of a decree: ISW reports Russia has been setting conditions for rolling partial reserve call-ups since at least October 2025 and has recently passed bills expanding the pool of recruitable Russians convicted of crimes and setting conditions to suppress mobilisation-related civil unrest. No mobilisation decree has followed as of July 27. Putin did sign a decree on July 27 raising the armed forces' authorised end strength to 2,426,130 total and 1,535,000 military personnel, up 27,000 on the June 12 figure — but an end-strength decree is a different instrument from a mobilisation order, and ISW notes the 2026 increases are marginal beside the expansions of 137,000, 170,000 and 180,000 in 2022, 2023 and 2024. On the second half of the claim, the numbers Zelensky published the following evening do not survive arithmetic, and are therefore reported here rather than asserted: ISW renders the total as about 222,500 Russian casualties for January–July 2026, Kyiv Independent as 'nearly 225,500' losses, and both print the same components — 131,000 killed and about 93,000 wounded — which sum to 224,000 and match neither total. Both agree Russia recruited about 221,000 over the same period, which is the comparison the claim actually rests on. The primary, Zelensky's July 25 evening address on president.gov.ua, returns HTTP 403 with no archived snapshot, so the original labelling could not be checked and no total is asserted here. Two independent comparison points cut the same way as the claim's direction without confirming its numbers: Ukraine's Foreign Intelligence Service put Russian contract signings at about 70,500 in the first quarter of 2026, roughly 30,000 short of the Russian Defence Ministry's goal, and Janis Kluge of the German Institute for International and Security Affairs puts the same quarter of 2025 at about 90,000.

Even 2025 projects planned two to three years ago struggled to come online, so 2026 projects face even steeper challenges.?

Decided by: A majority of large data centers scheduled for 2026 come online on their originally announced timelines — resolves by 2026-12-31

Theme: infrastructure-impact · From state-of-ai/infrastructure

Context: Forward-looking assessment by a Sightline Climate analyst on build delays; tagged as a prediction for The Record. Still open, and this cycle's record runs its way. Meta's Temple, Texas site began serving traffic in 2026 having been announced in 2022 and paused for redesign — four-plus years announcement to operation. Hut 8's "1 GW" Beacon Point campus carries 1,000 MW of secured utility capacity against 704 MW of signed IT leases. Southern Company reported a 75 GW large-load pipeline of which 10 GW is fully contracted. And Lawrence Berkeley National Laboratory names "projects that have not yet selected a power provider" as a known source of overestimation in rival load forecasts. None of this resolves the specific 2026 cohort claim; it resolves against how much of the 2026 pipeline actually energizes by year end.

GPU rental pricing is more likely to keep rising than falling through 2026.?

Decided by: H100/GPU cloud rental rates end 2026 higher than they began the year — resolves by 2026-12-31

Theme: compute-prices · From state-of-ai/infrastructure

Context: Forward-looking market call by SemiAnalysis that directly reverses the late-2025 oversupply consensus (including Introl’s December-2025 forecast of a further 10-20% decline). Still open, and running in the predicted direction as of this cycle: BCG, citing the Silicon Data H100 rental index, reports the H100 rate rose from about $1.96/GPU-hour in late 2025 to $2.64 in April 2026, and that the index has risen about 2.2% a month since January 2025 — though it fell as much as 11% at one point in that span, and the live index reading in late July 2026 was $2.53, below the April figure. Resolves on where the rate ends 2026 against where it began.

Google will replenish more water than it consumes across its data-center sites by 2030, and U.S. data centers in aggregate use less than 1% of the nation’s water.±

Decided by: Google replenishes more freshwater than it consumes across its data-center sites — resolved (mixed)

Google's replenishment pledge is a forward commitment (unresolved). The aggregate "less than 1%" framing is broadly consistent with national estimates but sits in tension with primary local records — e.g. Google's own quintupled draw at The Dalles reaching ~40% of that city's supply — where impacts concentrate. True in aggregate, misleading locally. This cycle adds a second problem with the aggregate framing rather than the local one: the standard operational metric, Water Usage Effectiveness, counts neither the potable-versus-reclaimed distinction nor the water consumed generating the electricity, and Lawrence Berkeley National Laboratory puts that indirect footprint at nearly 800 billion litres for 2023 against 66 billion litres of direct on-site water — roughly twelve times larger. LBNL separately states its own water model may run low, and that correcting one cooling assumption would lift modelled hyperscale water intensity by about a quarter. Verdict held at mixed.

Theme: infrastructure-impact · From state-of-ai/infrastructure

Context: Google's replenishment pledge is a forward commitment (unresolved). The aggregate "less than 1%" framing is broadly consistent with national estimates but sits in tension with primary local records — e.g. Google's own quintupled draw at The Dalles reaching ~40% of that city's supply — where impacts concentrate. True in aggregate, misleading locally. This cycle adds a second problem with the aggregate framing rather than the local one: the standard operational metric, Water Usage Effectiveness, counts neither the potable-versus-reclaimed distinction nor the water consumed generating the electricity, and Lawrence Berkeley National Laboratory puts that indirect footprint at nearly 800 billion litres for 2023 against 66 billion litres of direct on-site water — roughly twelve times larger. LBNL separately states its own water model may run low, and that correcting one cooling assumption would lift modelled hyperscale water intensity by about a quarter. Verdict held at mixed.

Electricity rates will not go up for residents because of this project. Georgia families will not subsidize this project.?

Decided by: Georgia residential electricity rates rise in a way Georgia Public Service Commission filings attribute to the Effingham County data center — resolves by 2032-12-31

Theme: infrastructure-impact · From state-of-ai/infrastructure

Context: Forward claim about a project whose power contract is not yet approved and whose first phase is not due until 2028. It rests on Georgia PSC large-load rules first approved in January 2025, which the utility says bar passing such costs to existing customers; whether that holds in practice is what the claim will be judged on. Resolves against Georgia Power rate filings and PSC decisions over the contracted 2028–2032 phase-in.

Electricity bills for American families will actually come down. They're going to have a lot of electricity left over, and they'll put that into the grid.?

Decided by: Average U.S. residential electricity bills fall from their 2026 level — resolves by 2028-12-31

Theme: infrastructure-impact · From state-of-ai/infrastructure

Context: Forward claim, no date given by the speaker; recorded against a 2028 horizon. It runs against every projection in this cycle’s record: ICF projects monthly bills rising 15% to 40% by 2030 on demand growth, and a Carnegie Mellon study cited by the Kansas Health Institute projects an 8% average U.S. increase by 2030, above 25% in central and northern Virginia. This page’s existing record shows residential bills already up about 27% since 2019. Resolves against EIA residential price data.

You are the first leader I have welcomed in person since taking office, and that is no accident. It is intended to send a very clear message: we are with Ukraine 100%, I am personally with you 100%, and I will honor every commitment this country has made to Ukraine in full.?

Decided by: Published UK government (MoD/HM Treasury) figures for military support to Ukraine covering the 2026-27 and 2027-28 financial years, cross-checked against the Kiel Institute Ukraine Support Tracker: the pledge holds if UK military support is sustained at £3bn a year or above and the 2024 UK-Ukraine Agreement on Security Co-operation remains in force. — resolves by 2027-07-31

Theme: geopolitics-war · From ukraine-russia-war/nato-relations

Context: A pledge made on Burnham's first week in office and his first meeting with a foreign leader, recorded here as a forward-looking commitment rather than a checkable past fact. The concrete commitment it inherits is published by his own government in the same release: £3 billion a year in military support to Ukraine 'for as long as it takes', within total UK support of £25 billion since the full-scale invasion. Nothing yet contradicts it — the visit itself, and the transfer of the Stone Cloak intellectual property, are consistent with continuity. Note that Burnham's predecessor's commitments included the January 2024 UK-Ukraine Agreement on Security Co-operation and the 100 Year Partnership, but not a position on the timing of Ukraine's NATO accession, which he did not address.

Ukraine will jointly produce with Raytheon one of the most critical air defense systems — interceptors for the Patriot system?

Decided by: Whether a signed licence or production contract for Ukrainian manufacture of Patriot-family interceptors or their components is confirmed by RTX/Raytheon, the US State Department or the US Department of Defense, and whether any such interceptor or component is manufactured in Ukraine — evidenced by a company or US government release, or a Ukrainian government production announcement corroborated by a wire service. — resolves by 2028-07-31

Theme: geopolitics-war · From ukraine-russia-war/weapons-procurement

Context: Unverified — a forward-looking claim recorded as a prediction. As of the cycle date no contract, licence or dollar value has been published, and Raytheon has issued no statement confirming an agreement; the company's president of land and air defence systems said at Farnborough on July 20 that the US licensing process 'has just started'. Zelensky's own follow-up sentence was that 'our teams — at the government level and from the private sector — will stay in touch to work out the details.' The Kyiv Post renders the same Telegram remarks as thanks for the company's 'readiness', which is materially weaker than the Kyiv Independent's 'have agreed to jointly produce'; the Telegram original is not text-extractable, so both renderings are secondary. On the horizon: Ukrainian parliamentary defence-committee deputy chair Yehor Chernev told AP that even with imported component kits a first pilot production line would take 18 to 24 months to stand up, which sets the most charitable concrete reading of this claim at mid-2028.

AI's $200B question is now AI's $600B question.?

Decided by: Annualized AI end-user revenue measured against the ~$600B implied by Nvidia run-rate capex — does the revenue gap close or persist — resolves by 2027-06-30

Theme: bubble · From state-of-ai/the-money

Context: Multi-year framing whose horizon runs to mid-2027; revisit annually. The capex side moved further out of reach this cycle: analyst estimates compiled by Bloomberg put 2026 capital spending by Alphabet, Microsoft, Amazon and Meta alone at about $724 billion, rising to nearly $950 billion in 2027, and Moody's projects roughly $785 billion in 2026 across six companies. The revenue side remains far below on every filed measure — OpenAI's audited 2025 revenue was $13.07 billion, Microsoft's last disclosed AI figure is a $37 billion annual run-rate rather than GAAP revenue, and the firmest AI-adjacent GAAP disclosure this cycle is Google Cloud's $24.768 billion quarter, in a segment that is not AI-only. The gap Cahn described has widened rather than closed, but the deadline for judging it has not arrived.

It's kind of like having your parents co-sign on your first mortgage?

Decided by: Whether Nvidia's vendor-financed investments in customers (OpenAI, CoreWeave, Lambda, others) unwind badly, as 1990s telecom vendor financing did — resolves by 2028-12-31

Theme: bubble · From state-of-ai/the-money

Context: Forward characterization; not yet resolvable, since no unwind has occurred. The structures have instead expanded and changed instrument. Bloomberg counts more than $540 billion of Nvidia deals with customers and partners announced in 2026 alone, including a $500 billion-plus SK Group initiative on 24 July 2026, and reported on 27 July that Nvidia is in early-stage, unconfirmed talks to guarantee as much as $250 billion of OpenAI's data-centre leases and to finance $350 billion of its chip purchases. Google has agreed to backstop lease payments for Anthropic at five sites in what Bloomberg says amounts to a $35 billion loan. Goldberg's analogy was to co-signing a mortgage; a lease guarantee is that arrangement rather than the equity stake he was asked about, but whether any of it unwinds badly remains open and the horizon runs to 2028.

Understating depreciation by extending useful life of assets artificially boosts earnings - one of the more common frauds of the modern era.?

Decided by: Whether hyperscaler earnings are shown (via restatements, SEC action, or realized impairments) to have been overstated by understated depreciation — Burry's specific ~$176B / Oracle ~27% / Meta ~21% by 2028 — resolves by 2028-12-31

Theme: bubble · From state-of-ai/the-money

Context: Forward, falsifiable accounting claim from a disclosed short-seller; resolves via later filings and any restatements. This cycle produced the first substantial evidence against it. FactSet reported on 23 July 2026 that contracted GPU leasing rates and second-hand prices for Nvidia H100 and A100 hardware launched three to six years ago are holding up well, and reads that as supporting the hyperscalers' 5–6 year depreciation schedules over the 3–4 year lives Burry argued for — while hedging that the evidence partly reflects today's compute scarcity. No company changed a depreciation schedule this cycle, and the 5–6 year assumption is described as unchanged. Alphabet's Q2 2026 filing shows depreciation of property and equipment rising 42% year over year to $7.104 billion while capex doubled; that gap is the mechanical result of spreading newly bought assets over a multi-year life and is not by itself evidence either way about the life assumed. Named counter-arguments already on record: CoreWeave's Intrator says older GPUs retain resale value; a depreciation professional notes useful-life estimates are audited in detail; Amazon has shortened, not extended, a subset of server lives.

Heavy capital spending relative to revenue will lead to declining, and in some cases negative, free cash flow and will hurt leverage ratios, to the extent these expenditures are debt-financed.?

Decided by: Hyperscaler free cash flow declining and in some cases turning negative, with leverage ratios deteriorating where the AI capex is debt-financed — resolves by 2027-12-31

Theme: bubble · From state-of-ai/the-money

Context: Forward, conditional call. One leg is already partly observable — Alphabet posted negative free cash flow of $5.855 billion in Q2 2026 — but that quarter was funded from cash and equity, not debt, so the conditional limb is untested. FactSet published an independently resolvable version the following day: fiscal-2026 free cash flow near zero or negative for all five hyperscalers it tracks except Alphabet and Microsoft. Resolves against fiscal-2026 and 2027 filings.

So between us, we're going to do half a trillion dollars' worth of business.?

Decided by: Whether roughly $500 billion of two-way business between Nvidia and SK Group is realised in Nvidia's filed revenue and component purchases and SK Group's disclosures — resolves by 2030-12-31

Theme: bubble · From state-of-ai/the-money

Context: Forward claim about a multi-year partnership announced days before this cycle. Note the disclosure form: it is a gross two-way figure that nets Nvidia's own memory purchases against SK Group's system purchases, so it is not $500 billion of Nvidia revenue. Resolves against Nvidia's filed revenue and SK Group's disclosures.

Sell China the fourth-best chip so “their developers get addicted to the American technology stack.”?

Decided by: Whether selling China throttled U.S. chips (H20-class) keeps Chinese AI developers dependent on the American stack, versus accelerating adoption of domestic alternatives (e.g. Huawei Ascend) — resolves by 2027-12-31

Theme: national-advantage · From state-of-ai/international-competition

Context: Forward-looking strategy claim; still open, but this cycle produced the first substantial evidence running against it. The dependency Lutnick described was on the American stack; by July 2026 the reported dependency ran partly the other way — nearly 200 U.S. startup founders urged the administration not to cut off access to Chinese open-weight models, arguing smaller American firms rely on them because they cannot afford frontier compute (Politico, 2026-07-23), and more than 20 companies including Nvidia, Microsoft, Meta and Palantir signed a letter against “premature restrictions” on open-weight models (CNBC, 2026-07-25). Earlier evidence still cuts the other way too: BIS warned industry against Huawei Ascend chips and Chinese buyers were reported pursuing domestic substitutes. Not resolvable before the 2027-12-31 horizon; revisit against adoption data on both sides.

No AI-chip export policy to China can be “permissive, implementable, enforceable, and protective of US national security” at once.?

Decided by: Whether the January 2026 case-by-case export framework proves enforceable (its 50%-cap, end-use, and third-party-lab conditions hold) or leaks/expands toward more capable chips — resolves by 2027-12-31

Theme: national-advantage · From state-of-ai/international-competition

Context: Forward-looking enforceability claim tied to the January 2026 rule; still open, but this cycle added the first on-the-record challenge to the “implementable, enforceable” half. On 2026-07-23 Senator Elizabeth Warren, ranking member of Senate Banking, invoked the Export Control Reform Act to compel BIS to document what she calls a foundry loophole: the January 2025 Foundry Due Diligence rule presumes advanced chips need licences, but rests on the AI Diffusion Rule’s worldwide licence requirement, and she argues the May 2025 rescission removed that predicate — citing reporting that it “may be allowing an unknown number of high-end semiconductors to reach Beijing.” That is an allegation from a minority-party document, not a finding; BIS’s response is due 2026-08-07 and is the next checkable milestone. Not resolvable before the 2027-12-31 horizon.

AI is likely to replace half of all white-collar workers in the U.S.?

Decided by: Share of U.S. white-collar employment replaced by AI by 2030, vs Farley's ~50%. — resolves by 2030-12-31

Theme: jobs · From state-of-ai/societal-impact

Context: No explicit horizon was given; recorded with a charitable ~5-year reading (resolveBy 2030). Tagged as a prediction and NOT yet due. New evidence this cycle runs against it, from the speaker's own company: Bloomberg reported on 25 June 2026 that Ford has been rehiring quality inspectors after AI fell short, and that Ford hired more than 300 engineers, including ex-employees, over three years; on 2 July 2026 Farley told NewsNation there is "no replacement" for experienced engineers and described AI as a way "to grow with the same amount of people." That is a partial retreat by the same speaker, not a resolution — the claim is about U.S. white-collar employment by 2030, and one manufacturer's rehiring does not settle it. Status stays unverified until the horizon.

I could easily see 30% of [~26,000 back-office roles] getting replaced by AI and automation over a five-year period.?

Decided by: Number of IBM back-office/HR roles left unbackfilled and attributable to AI/automation over 2023–2028, per IBM disclosure or credible reporting (Krishna's ~7,800 figure). — resolves by 2028-12-31

Theme: jobs · From state-of-ai/societal-impact

Context: A dated claim about IBM's own workforce, resolvable by 2028. As of this cycle IBM has not published an audited count of back-office roles left unbackfilled specifically because of AI, so it stays unverified. Tagged as a prediction.

BT will cut up to 55,000 jobs by 2030, with about 10,000 roles replaced by AI and new technology.?

Decided by: BT total headcount reduction by 2030 and the ~10,000 share credibly attributable to AI/automation, per BT disclosure or reporting. — resolves by 2030-12-31

Theme: jobs · From state-of-ai/societal-impact

Context: A 2030 workforce forecast; the ~10,000 AI-attributed share is the disputed part (Barford: framing over new substance). Tagged as a prediction, resolvable by 2030.

By 2030, AI and other trends will create 170 million jobs and displace 92 million — a net gain of 78 million.?

Decided by: Net global job change 2025→2030 vs WEF's +78M / 22% churn, per ILO and national labor statistics. — resolves by 2030-12-31

Theme: jobs · From state-of-ai/societal-impact

Context: A 2030 macro forecast blending AI with demographic and green-transition trends, so it is not an AI-only claim. Tagged as a prediction; resolvable against net-employment statistics by 2030.

Global banks will cut as many as 200,000 jobs over the next three to five years as AI encroaches on human tasks.?

Decided by: Net headcount change across major global banks 2025→2030 attributable to AI, vs the projected ~200,000 (~3%) reduction. — resolves by 2030-01-09

Theme: jobs · From state-of-ai/societal-impact

Context: Based on surveyed executive expectations, not committed plans. Tagged as a prediction, resolvable by 2030.

AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10–20% in the next one to five years.?

Decided by: Share of U.S. entry-level white-collar jobs eliminated and the U.S. unemployment rate by 2030, vs Amodei's ~50% and 10–20%. — resolves by 2030-05-28

Theme: jobs · From state-of-ai/societal-impact

Context: Not due (1–5-year horizon from May 2025). As of mid-2026 the aggregate counter-evidence is strong — the Budget Lab at Yale finds no discernible economy-wide effect, Challenger attributes <4.5% of 2025 layoffs to AI, and Altman himself reversed — while Stanford finds a targeted early-career effect. Tagged as a prediction.

We expect that this will reduce our total corporate workforce as we get efficiency gains from using AI extensively across the company.?

Decided by: Direction of Amazon's total corporate headcount through ~2028 and whether Amazon credibly attributes reductions to AI efficiency. — resolves by 2028-12-31

Theme: jobs · From state-of-ai/societal-impact

Context: A multi-year forecast about Amazon's own headcount. The subsequent 2025–26 cuts were officially attributed to bureaucracy/cost, not AI, so the AI-causation remains the CEO's framing. Tagged as a prediction.

A Supreme Court decision on AI training and copyright is essentially inevitable, with some publishers settling but enough litigating to force resolution of the novel question.?

Decided by: A U.S. Supreme Court merits decision squarely resolving whether unlicensed use of copyrighted works to train generative AI is copyright infringement or fair use. — resolves by 2031-12-31

Theme: regulation-law · From state-of-ai/law-and-governance

Context: A forward-looking legal prediction made at the NYT filing. As of this cycle no US Supreme Court merits decision on AI-training copyright has issued; the district-court record is still developing (Thomson Reuters/ROSS, Bartz, Kadrey). Unresolved.

In most cases it will likely be illegal to copy copyrighted works to train generative AI without permission, because AI trained on those works can generate countless competing works and thereby dilute the market for the originals — so companies will generally need to pay for training rights.?

Decided by: The prevailing outcome in US courts of AI-training fair-use disputes: whether unlicensed copying of copyrighted works to train generative AI is generally held to be infringement (requiring licensing) rather than fair use. — resolves by 2030-12-31

Theme: regulation-law · From state-of-ai/law-and-governance

Context: Dicta in the Kadrey opinion articulating a "market dilution" theory. It is a forecast about how future AI-training copyright cases should and will come out; as of this cycle the case law is split (contrast Bartz) and no appellate resolution has issued. Unresolved.

my basic prediction is that AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years.?

Decided by: Documented biomedical R&D progress (major disease cures/therapeutics, longevity gains) attributable to AI equivalent to 50–100 years of prior-rate advancement, achieved within 5–10 years of ‘powerful AI.’ — resolves by 2035-12-31

Theme: capability-milestones · From state-of-ai/safety

Context: From Amodei’s essay ‘Machines of Loving Grace.’ The clock is defined to start at the arrival of ‘powerful AI’ (which Amodei does not fix to a firm date), so the resolve-by below reads the ‘within 5-10 years’ window against the essay’s October 2024 framing; the metric is loosely operationalised and may resolve mixed.

If the ~7-month doubling of the task-completion horizon holds, AI agents will be able to complete many software/engineering tasks that currently take humans days or weeks — reaching roughly month-long autonomous tasks — within the decade.?

Decided by: The 50%-reliability task-completion time-horizon of the frontier model on METR’s public time-horizons tracker reaching ~1 month (≈160 human work-hours). — resolves by 2032-12-31

Theme: capability-milestones · From state-of-ai/safety

Context: METR’s own extrapolation from its measured six-year doubling trend; not yet resolvable. METR notes a 10x error in the absolute measurement would shift arrival estimates by only ~2 years, and that the SWE-Bench-Verified subset doubled even faster (under 3 months).

of course we're in an AI bubble?

Decided by: Whether AI-linked equity valuations undergo a sharp correction, and on what horizon (Gelsinger: several years out; Marks: not yet) — resolves by 2029-12-31

Theme: bubble · From state-of-ai/the-money

Context: Forward call; not yet resolvable. Recorded alongside opposing views (Marks, Fink) so the conversation's calibration can be scored later.

That is capitalism. We're going to have some big winners and we're going to have some big losers ... but if you have a diversified portfolio, you're going to be fine.?

Decided by: Whether AI capex proves 'well spent' — hyperscalers emerge as clear winners with no broad bust — resolves by 2028-12-31

Theme: bubble · From state-of-ai/the-money

Context: Boom-side forward call; not yet resolvable.

OpenAI is on track to generate more than $20 billion in annualized revenue run rate this year, with plans to grow to hundreds of billions in sales by 2030.±

Decided by: OpenAI reported (GAAP) annual revenue reaching 'hundreds of billions' by 2030 — resolved (mixed)

The exit-month run-rate did cross $20 billion (OpenAI's CFO said annualized revenue crossed $20B in 2025), so the run-rate claim holds; but audited full-year 2025 GAAP revenue was $13.07 billion, so the framing overstates realized annual revenue. The 'hundreds of billions by 2030' portion is unverified.

Theme: bubble · From state-of-ai/the-money

Context: The exit-month run-rate did cross $20 billion (OpenAI's CFO said annualized revenue crossed $20B in 2025), so the run-rate claim holds; but audited full-year 2025 GAAP revenue was $13.07 billion, so the framing overstates realized annual revenue. The 'hundreds of billions by 2030' portion is unverified.

There will be no federal bailout for AI. The U.S. has at least 5 major frontier model companies. If one fails, others will take its place.?

Decided by: Whether any federal bailout, guarantee or backstop is extended to an AI or frontier-model company — resolves by 2028-12-31

Theme: bubble · From state-of-ai/the-money

Context: Forward commitment; no such backstop had occurred as of this cycle. Not yet resolvable.

There is no way hyperscaler AI infrastructure investments will generate returns at current spending levels.?

Decided by: Hyperscaler AI-infrastructure capex at 2025 spending levels fails to generate adequate financial returns — resolves by 2027-12-31

Theme: infrastructure-impact · From state-of-ai/infrastructure

Context: Forward-looking claim by IBM CEO Arvind Krishna about industry-wide capex returns; unresolved. Tagged as a prediction for The Record.

Global data-center capital spending will exceed $1 trillion in 2026.?

Decided by: Global data-center capital expenditure reaches or exceeds $1 trillion in calendar 2026 — resolves by 2026-12-31

Theme: infrastructure-impact · From state-of-ai/infrastructure

Context: Forward projection by Dell’Oro Group; unresolved. Tagged as a prediction for The Record.

The U.S. is doubling down on restricting sales of chips to China and purchases from China, but models like Qwen 3 that are state-of-the-art and open […] will undoubtedly be used domestically. It reflects the reality that businesses are both building their own tools [as well as] buying off the shelf via closed-model companies like Anthropic and OpenAI.

Decided by: Documented US enterprise adoption of Chinese open-weight models (hosting-platform traffic shares, credible adoption reporting) — resolved (true)

Confirmed by mid-2026: CNBC reported Western companies increasingly adopting Chinese open-weight models on cost grounds, and OpenRouter token-traffic data showed GLM-5.2 usage climbing faster post-launch than any predecessor — Chinese open models are in production US use at scale.

Theme: adoption · From state-of-ai/leading-models

Context: Confirmed by mid-2026: CNBC reported Western companies increasingly adopting Chinese open-weight models on cost grounds, and OpenRouter token-traffic data showed GLM-5.2 usage climbing faster post-launch than any predecessor — Chinese open models are in production US use at scale.