Macro theses — AI infrastructure forecast desk

Warnings

Where I think the consensus is too comfortable.

Enterprise on-prem AI at scale stays undisclosed

92%80% CI 87–95%

By 2027-06-30, fewer than three Fortune 500 companies outside the technology, semiconductor and cloud sectors will have disclosed, in an 8-K, annual report, or company-issued press release, an owned on-premises AI compute deployment of 50 MW or more.

warningby June 30, 2027Fortune 500

Why

The ledger's disclosure-type questions (a named primary document by a hard date) resolve at 46% (n=37) against a book average of 53% (n=74); the software lens runs at 45% (n=27). Enterprises consume frontier compute through hyperscalers and neoclouds and have weak incentives to disclose owned megawatts. The thesis is falsified by three qualifying disclosures, not by trend pieces or vendor case studies.

What would change my mind

A sovereign or regulated-industry buyer publishing a capacity figure in a filing.
A vendor 8-K naming a non-tech Fortune 500 customer with an MW-scale on-prem order.

Milestones on the live board

Council

94% grid-and-power-economist (openai/gpt-5.6-sol): A 50 MW owned on-premises AI deployment is exceptionally large for a non-technology enterprise and would require substantial power, cooling, interconnection, and equipment commitments. Most enterprises obtain frontier compute through cloud or hosted capacity, while public disclosures rarely specify both ownership and megawatts. With only eight months remaining, reaching three qualifying Fortune 500 disclosures—each in an enumerated primary document—requires several unusually explicit announcements. Ambiguous vendor case studies or planned facilities would not qualify.
91% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): A 50 MW owned deployment is exceptionally large for a non-technology enterprise and would require substantial power, facilities, accelerators, and cooling. Most Fortune 500 enterprises obtain frontier AI capacity through cloud or colocation providers, while hyperscaler capex growth reinforces that concentration. The resolution also requires three companies to document ownership, scale, and on-premises deployment in narrowly specified primary documents within eight months; vendor announcements, plans, and partnerships would not qualify.
95% capital-markets-credit-analyst (openai/gpt-5.6-sol): TRUE is highly likely. The test requires three separate non-tech Fortune 500 companies to disclose owned, on-premises deployments of at least 50 MW in specified primary documents. Enterprises generally obtain frontier compute through cloud providers, while 50 MW is data-center-scale infrastructure rarely described as owned enterprise AI capacity. With only eight months remaining and no qualifying disclosures identified in the supplied digest, reaching three would require an unusually concentrated disclosure wave.
86% superforecaster-statistician (openai/gpt-5.6-sol): Base rate: 0.46 for questions requiring a named primary disclosure by a hard date. I adjust upward because TRUE requires only that fewer than three counterexamples appear, while each counterexample must combine Fortune 500 status, excluded-sector compliance, ownership, on-premises deployment, a ≥50 MW threshold, and timely primary disclosure. Eight months is short, and hyperscaler capex growth does not establish qualifying enterprise-owned deployments. I temper confidence for uncertain sector and ownership classifications.
92% geopolitics-regulation-analyst (openai/gpt-5.6-sol): The threshold is highly conjunctive: Fortune 500 status, excluded-sector screen, enterprise ownership, on-premises AI use, at least 50 MW, and disclosure in a specified primary document. Fifty megawatts is hyperscale infrastructure, while nontechnology enterprises generally procure frontier compute from cloud providers and rarely report facility power. With only eight months remaining, reaching three would likely require multiple projects already near commissioning and unusually explicit disclosures.

Public series used

huggingface-hub-models: n/a (as of 2026-10-19)
sec-xbrl-companyfacts: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

A listed neocloud has a credit event

52%80% CI 43–60%

By 2027-09-30, at least one publicly listed GPU-cloud operator (CoreWeave, Nebius, or another listed neocloud) discloses in an SEC filing a covenant waiver or amendment, a going-concern qualification, a missed debt service payment, or the loss of a customer representing 10% or more of remaining performance obligations.

warningby September 30, 2027CoreWeaveNebius

Why

Capital-lens questions resolve at 63% (n=14). High-yield OAS is 3.23% as of 2026-10-15, so credit markets are not signalling stress today; the thesis is a duration-mismatch call, not a spread call: GPU useful lives, contract tenors, and amortising debt are misaligned and the first disclosure of a tenor gap is already in the ledger. Twelve months is enough for one renewal cycle to miss. Falsified if no listed neocloud files any of the four triggers.

What would change my mind

Hyperscaler contract extensions that lengthen tenors past debt maturities.
Equity raises that retire GPU-backed debt at par.
High-yield OAS compressing further alongside neocloud refinancing at lower coupons.

Milestones on the live board

Council

46% grid-and-power-economist (openai/gpt-5.6-sol): The trigger set is broad, but covenant amendment is the only materially likely route; missed payments and going-concern language remain tail outcomes. CoreWeave’s $24.9B debt and extreme capex create refinancing and execution risk despite $103.7B RPO and rapid growth. A 5.1% Treasury yield and sharply wider single-B spreads increase pressure, while benign aggregate high-yield spreads and the small listed-neocloud universe limit the one-year probability. Exact SEC-filing language also makes customer-loss resolution difficult.
56% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): CoreWeave’s $24.9B long-term debt and continued capital intensity create meaningful refinancing and covenant risk, while the trigger is broad: even a covenant amendment qualifies. The issuer-level backdrop is less benign than headline high-yield spreads suggest, with single-B OAS at 12.78%. However, rapidly growing RPO, strong revenue growth, and access to expanding AI infrastructure demand reduce near-term payment and going-concern risk. Strict SEC-filing documentation and the small listed-neocloud universe limit the probability.
56% capital-markets-credit-analyst (openai/gpt-5.6-sol): The trigger is broader than outright default: a disclosed covenant amendment alone qualifies. CoreWeave’s $24.9B long-term debt, rapid capex and revenue growth, and concentrated contracted demand create meaningful refinancing and execution risk across roughly four filing cycles. The 12.78 single-B OAS signals substantial weak-credit stress despite benign aggregate HY spreads. However, large and growing RPO provides liquidity visibility, while immaterial amendments or customer changes may not receive qualifying SEC disclosure.
43% superforecaster-statistician (openai/gpt-5.6-sol): Base rate: 0.404 for >90-day questions. I adjust upward because multiple issuers and four alternative triggers create several paths, while CoreWeave’s $24.9B debt and sharply wider single-B spreads indicate refinancing risk. I adjust downward because broad high-yield spreads remain moderate, reported RPO is growing strongly, and resolution requires an explicit SEC filing by a hard date. The forecaster’s 0.5–0.7 calls have been materially overconfident.
57% geopolitics-regulation-analyst (openai/gpt-5.6-sol): The trigger is a broad OR condition across multiple issuers, and covenant amendments are materially more common than missed payments or going-concern opinions. CoreWeave’s $24.9B debt and extreme capex create refinancing and execution exposure despite rapidly growing RPO. Elevated rates and a 12.78 single-B OAS strengthen the downside case, but aggregate high-yield spreads, strong demand, and large backlogs argue against acute stress. A qualifying SEC disclosure within eleven months is slightly more likely than not.

Public series used

fred-BAMLH0A0HYM2: 3.23 (as of 2026-10-15)
fred-BAMLH0A3HYC: 12.78 (as of 2026-10-15)
sec-xbrl-companyfacts: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

Calls

Where I think the direction is set and the timing is knowable.

HBM stays sold out into 2027

65%80% CI 57–73%

By 2027-03-31, at least two of SK hynix, Samsung Electronics, and Micron state in an earnings release, earnings call, or company press release that their 2027 HBM supply is fully allocated, sold out, or substantially committed under long-term agreements.

callby March 31, 2027SK hynixSamsung ElectronicsMicronNVIDIA

Why

Hardware-lens questions in the ledger resolve at 59% (n=15), the book's strongest lens, and the memory-allocation questions in particular have hit or partially hit. Memory makers have used 'sold out' language for the forward year once qualification is complete; two of three doing so for 2027 requires the supply-demand gap to persist through the Vera Rubin ramp. Falsified if fewer than two make the statement by the horizon.

What would change my mind

A memory maker guiding to HBM price declines or capacity additions outrunning orders.
NVIDIA or a hyperscaler cutting 2027 accelerator unit guidance.

Milestones on the live board

Council

68% grid-and-power-economist (openai/gpt-5.6-sol): SK hynix and Micron are the likeliest pair: HBM capacity is typically contracted well ahead, qualification constrains fungibility, and the digest indicates continued AI-hardware demand. Both have multiple reporting opportunities before March 31. Samsung adds a third path but may avoid categorical allocation language or retain capacity while qualifications progress. The main risk is documentary: strong demand or customer agreements do not count unless two manufacturers explicitly characterize 2027 supply as sold out, fully allocated, or substantially committed.
65% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): HBM qualification and advanced-packaging constraints make SK hynix and Micron the likeliest pair: both have historically pre-allocated forward HBM output once customer ramps firm, while Samsung is less dependable because qualification and wording can lag. Strong AI-chain revenue and backlog signals support continued demand, but rising rates and capacity additions could loosen 2027 supply. Resolution requires explicit company language by March 31, not merely analyst reports or tight market conditions.
69% capital-markets-credit-analyst (openai/gpt-5.6-sol): The likeliest qualifying pair is SK hynix and Micron; both have historically disclosed forward HBM allocation during earnings, while Samsung is a less reliable source of explicit sold-out language. Five months allows at least one major reporting opportunity for each. Strong AI-related semiconductor demand supports continued commitments, but added HBM capacity, customer diversification, or wording limited to strong demand rather than 2027 allocation could prevent literal resolution.
59% superforecaster-statistician (openai/gpt-5.6-sol): Base rate: 0.594 for hardware questions, tempered by 0.404 for >90-day horizons. Prior memory cycles support forward-year allocation language, and Micron’s reported commitments indicate strong demand. However, this requires qualifying disclosures from two named manufacturers by a hard date; tight supply alone is insufficient. Rising rates add demand risk, while the ledger’s 0.5–0.6 forecasts have been materially overconfident, warranting restraint.
64% geopolitics-regulation-analyst (openai/gpt-5.6-sol): HBM demand and AI accelerator commitments remain strong, and two reporting cycles remain before the deadline. SK hynix and Micron are the likeliest pair to use qualifying 2027 allocation language; Samsung provides a third path. The broader phrase “substantially committed under long-term agreements” materially helps. Main risks are added capacity, customer diversification, or cautious wording that describes strong demand without explicitly characterizing 2027 supply as allocated or committed.

Public series used

sec-xbrl-companyfacts: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

Power-first hyperscale campuses keep coming

43%80% CI 35–52%

Between 2026-09-06 and 2027-06-30, hyperscalers (Microsoft, Amazon, Alphabet, Meta, Oracle) announce, each in a primary company disclosure, at least three distinct new AI campuses or generation blocks of 500 MW or more that are explicitly behind-the-meter, co-located with dedicated generation, or described as power-first.

callby June 30, 2027MicrosoftAmazonAlphabetMeta

Why

Power-lens questions are the book's most reliable: 79% (n=6). Behind-the-meter announcements have repeated because grid interconnection is the binding constraint: 13.9% of 2015-2020 interconnection requests had reached commercial operation by end-2025 (LBNL Queued Up, n=11622). Three announcements in ten months is a modest bar against the recent cadence; the risk is in the 500 MW and primary-disclosure qualifiers, not in the trend.

What would change my mind

A federal or state rule that makes behind-the-meter generation uneconomic for large loads.
Hyperscaler capex guidance cuts of 20% or more for 2027.

Milestones on the live board

Council

43% grid-and-power-economist (openai/gpt-5.6-sol): The underlying power-first trend is strong, but the resolution test is unusually strict: three distinct projects, each at least 500 MW, explicitly characterized as behind-the-meter, dedicated-generation co-location, or power-first, and disclosed by a named hyperscaler. Many plausible deals will instead appear in utility, developer, or government releases, or omit precise capacity and topology. With no qualifying disclosures established in the digest and roughly eight months remaining, three clean hits is below even odds.
47% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): Grid delays strongly motivate dedicated-generation designs, and hyperscaler AI spending makes several large-campus announcements plausible. But resolution requires three disclosures within ten months, each from a named hyperscaler and explicitly satisfying both the 500 MW threshold and power-first, co-location, or behind-the-meter language. Such details are often disclosed by utilities, developers, or governments rather than the hyperscaler itself. The favorable power-lens base rate is also based on only six cases.
38% capital-markets-credit-analyst (openai/gpt-5.6-sol): Grid delays strongly favor power-first designs: median request-to-COD reached 5.44 years in 2025, while generation and equipment backlogs are expanding. However, three qualifying disclosures is demanding. Hyperscalers often omit campus MW, announce ordinary grid-supplied sites, or leave generation details to utilities and developers. The requirement for each announcement to be primary, explicit, new, and at least 500 MW materially lowers the hit rate despite strong underlying demand.
42% superforecaster-statistician (openai/gpt-5.6-sol): I start from the 0.404 base rate for >90-day questions, not the 0.786 power-lens rate, which has only six cases. Severe queue delays make dedicated-power designs plausible, and five well-funded issuers create multiple shots. But resolution requires three separate ≥500 MW projects in primary issuer disclosures within ten months; utility, vendor, or media announcements do not count. That count-plus-documentation hurdle and visible midrange overconfidence outweigh the favorable theme.
47% geopolitics-regulation-analyst (openai/gpt-5.6-sol): Grid interconnection delays strongly favor power-first designs, and five hyperscalers provide multiple announcement opportunities. However, resolution requires three separate disclosures within ten months, each clearing 500 MW and using explicit behind-the-meter, dedicated-generation, co-location, or power-first language. Corporate announcements often discuss power procurement without satisfying those terms. No qualifying announcement since September 6 is established in the supplied digest, and the small power-lens sample should not outweigh the strict multi-event test.

Public series used

lbnl-queued-up: n/a (as of 2026-10-19)
sec-xbrl-companyfacts: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

State-level data-center policy friction spreads

59%80% CI 51–68%

By 2027-06-30, at least three additional US states (beyond those counted at 2026-09-06) adopt, through enacted legislation or a public utility commission order, a statewide data-center moratorium, a discretionary-permit pause, or a large-load tariff or grid-interconnection rule.

callby June 30, 2027FERCPJMERCOT

Why

State action clusters after visible ratepayer impact, and most 2027 legislative sessions convene in January, inside the horizon. The ledger's two live policy questions are single-leg threshold questions, a type that resolves at 65% (n=25). Three states in ten months is bold but measurable; county pauses do not count.

What would change my mind

Retail rate relief mechanisms that remove the ratepayer argument.
Federal preemption of state large-load rules.

Milestones on the live board

Council

58% grid-and-power-economist (openai/gpt-5.6-sol): The broad qualifying categories—especially commission-approved large-load tariffs or interconnection rules—create multiple paths across roughly eight months and most 2027 legislative sessions. Visible cost-allocation and reliability concerns should produce clustered state action. However, three additional states is demanding: proposals, utility filings, local moratoria, and preliminary commission decisions do not qualify, while contested tariffs can slip past June 30. The absence of a current state-by-state docket ledger warrants discounting the threshold-question base rate.
61% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): The broad inclusion of statewide large-load tariffs and interconnection rules makes three states plausible: commissions can act faster than legislatures, while the deadline captures most 2027 sessions. Rapid load growth and ratepayer-cost concerns create correlated policy pressure. However, three qualifying adoptions within eight months remains demanding, and utility-specific tariffs, proposed bills, county pauses, or unfinished proceedings may fail the literal statewide and adoption requirements.
62% capital-markets-credit-analyst (openai/gpt-5.6-sol): The broad inclusion of PUC-approved large-load tariffs and interconnection rules makes three states attainable, especially as 2027 legislative sessions respond to grid constraints and ratepayer concerns. Long interconnection timelines strengthen pressure for special large-load treatment. However, the deadline leaves only one legislative cycle, enactment can lag proposals, and utility-specific tariffs or local permitting pauses may fail the statewide requirement. The forecaster’s long-horizon record also supports discounting the threshold base rate.
52% superforecaster-statistician (openai/gpt-5.6-sol): Base rate: 0.549 for single-leg questions; the threshold class is higher (0.654), but >90-day questions are only 0.404. The broad “tariff or interconnection rule” pathway and 2027 legislative sessions support diffusion. However, TRUE requires three distinct formal statewide actions within eight months—not proposals or local pauses—and the digest provides no verified state-level pipeline. The forecaster’s 0.5–0.7 bins have been overconfident, supporting an estimate near even odds.
64% geopolitics-regulation-analyst (openai/gpt-5.6-sol): The horizon includes nearly a full 2027 legislative cycle, while PUC orders provide a second, often faster route. Rising large-load demand and ratepayer-allocation disputes should push multiple states toward dedicated tariffs or interconnection rules. However, three additional qualifying states is demanding: proceedings often slip, proposed or interim measures may not count, and county-level pauses are explicitly excluded. The broad tariff-or-grid-rule language keeps the outcome modestly more likely than not.

Public series used

lbnl-queued-up: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

Interconnection conversion stays below one in five

91%80% CI 87–93%

The next LBNL Queued Up edition (data through 2026), if published by 2027-09-30, reports that no more than 20% of the generation capacity requested in 2018-2020 had reached commercial operation by end-2026; if no edition is published by then the thesis resolves against the engine.

callby September 30, 2027Lawrence Berkeley National LaboratoryPJMMISOERCOT

Why

13.9% of 2015-2020 interconnection requests had reached commercial operation by end-2025 (LBNL Queued Up, n=11622). The 2018-2020 vintages sit at 2018: 8.2% of MW, 2019: 8.0% of MW, 2020: 4.5% of MW today, and median request-to-COD time has been lengthening. Twelve more months adds single-digit points of completion for those vintages, so 20% is a comfortable ceiling.

What would change my mind

FERC Order 2023 cluster studies clearing a large backlog of executed agreements into service.
A methodology change in the LBNL edition that redefines the vintage denominator.

Milestones on the live board

Council

93% grid-and-power-economist (openai/gpt-5.6-sol): Conditional on publication, the threshold is highly likely: the three vintages currently show only 4.5%-8.2% of requested MW operating, while median request-to-COD duration has risen to 5.44 years and cumulative COD incidence remains low. A one-year increase sufficient to exceed 20% would be historically exceptional, especially after substantial withdrawals. The principal failure risk is procedural: no qualifying data-through-2026 Queued Up edition by the hard deadline, or a material methodology revision.
91% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): Through 2025, the three vintage COD shares are only 8.2%, 8.0%, and 4.5%. Historical cumulative incidence rises just 2.6 points between years five and seven, while median request-to-COD reached 5.44 years. Crossing 20% by end-2026 would therefore require an unprecedented one-year conversion surge. The principal risk is procedural: the required data-through-2026 edition may be delayed beyond September 2027 or alter its vintage reporting.
92% capital-markets-credit-analyst (openai/gpt-5.6-sol): The 2018-2020 vintages currently show only 4.5%-8.2% of requested MW operating. Crossing 20% by end-2026 would require an unprecedented one-year conversion surge despite median request-to-COD time reaching 5.44 years and cumulative incidence remaining only 10% after seven years. The principal failure risk is procedural: no data-through-2026 edition by September 30, 2027, or a material retrospective denominator revision.
88% superforecaster-statistician (openai/gpt-5.6-sol): Base rate: 0.654 for threshold questions. Conditional on publication, TRUE is highly likely: the three vintages were only 4.5%-8.2% complete by end-2025, while all-vintage cumulative COD incidence is just 10.0% after seven years and completion times are lengthening. Exceeding 20% would require an unusually large one-year jump. I reduce for the required LBNL edition by 2027-09-30 and the ledger’s high-confidence overprediction.
89% geopolitics-regulation-analyst (openai/gpt-5.6-sol): The threshold leg is highly likely: 2018-2020 vintages were only 4.5%-8.2% operational through 2025, while the all-vintage cumulative COD incidence reaches just 10% after seven years and completion times continue lengthening. Higher rates and widening credit spreads are additional headwinds. The main failure risk is procedural: the resolution requires LBNL to publish the data-through-2026 edition by September 30, 2027. Annual publication precedent makes that likely, but not certain.

Public series used

lbnl-queued-up: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

Second-source 1.6T and co-packaged optics ship

49%80% CI 39–58%

By 2027-06-30, at least two vendors other than Broadcom (from Marvell, Credo, NVIDIA, Cisco, Ciena, Coherent, Lumentum) state in an earnings release or earnings call that 1.6T or co-packaged-optics products are shipping in production volume to at least one hyperscaler.

callby June 30, 2027BroadcomMarvellCredoNVIDIA

Why

The networking lens has a mixed record and the resolved optics questions have landed as partials or misses as often as hits: design wins are announced, production revenue lags. 'Production volume to a hyperscaler' in an earnings document is the strict version of the milestone questions. Two named vendors by mid-2027 aligns with the Vera Rubin ramp but is exposed to CPO reliability slips.

What would change my mind

A hyperscaler publicly delaying CPO adoption to a later switch generation.
Broadcom guidance implying it holds essentially all 1.6T volume through 2027.

Milestones on the live board

Council

53% grid-and-power-economist (openai/gpt-5.6-sol): Eight months leaves two to three earnings cycles. Strong growth at Credo, Marvell, and Ciena supports an active AI-networking ramp, while seven eligible vendors create multiple paths. Marvell, Credo, and optical-component suppliers are likelier near-term qualifiers than CPO systems, whose qualification schedules can slip. I discount heavily because two companies must explicitly establish both production volume and a hyperscaler in an earnings channel; design wins, samples, or generic cloud revenue do not qualify.
54% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): The seven-vendor pool and two to three remaining reporting cycles make two qualifiers plausible, particularly via 1.6T DSPs or transceivers from Marvell, Credo, or Coherent; CPO is less likely to contribute on time. Strong Credo and Marvell growth supports deployment activity but does not establish the required product mix. The main discount is documentary: sampling, qualification, initial shipments, or unnamed customer revenue would not satisfy an explicit earnings-document statement of production-volume hyperscaler shipments.
54% capital-markets-credit-analyst (openai/gpt-5.6-sol): Seven eligible vendors provide multiple paths, and strong networking revenue growth supports a 1.6T ramp through mid-2027. Coherent, Lumentum, Marvell, and Credo are the likeliest qualifiers. However, the rule requires two vendors to use earnings materials to establish both production-volume shipping and a hyperscaler destination. Design wins, sampling, qualification, or generic customer shipments fail; correlated qualification or CPO reliability delays could therefore prevent the second qualifying disclosure.
38% superforecaster-statistician (openai/gpt-5.6-sol): I use the 0.346 networking-question base rate; the >90-day reference class is 0.404. Strong CIEN and CRDO growth supports optical-networking demand, and seven eligible vendors provide multiple paths. However, this is a two-vendor conjunction with unusually strict wording: production volume, hyperscaler delivery, and disclosure specifically in an earnings release or call by June. Product announcements, sampling, design wins, or later confirmation fail. The calibration record also warrants discounting medium-confidence optimism.
44% geopolitics-regulation-analyst (openai/gpt-5.6-sol): The underlying 1.6T ramp is plausible: the digest shows strong revenue growth at Credo, Marvell, and Ciena, while Rubin-era networking should create demand before the deadline. But resolution requires two vendors to use earnings communications to confirm production-volume shipments specifically to a hyperscaler. Vendors often disclose sampling, design wins, or ramps without customer class or volume language. CPO reliability and qualification timelines add delay risk, leaving only about three earnings cycles.

Public series used

sec-xbrl-companyfacts: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

Contrarian

Where I am on the other side of the room.

Vendor-dated milestones slip more than the book expects

56%80% CI 47–64%

Of the ledger's live vendor-timing questions lq-p83, lq-p77, lq-p95, lq-p86, and lq-p33, at most two resolve as a hit by 2027-03-31 under their written resolution rules.

contrarianby March 31, 2027DeepSeekArtificial AnalysisSemiAnalysisOpen Compute Project

Why

Timing-type questions in the ledger resolve at 50% (n=10) and multi-leg questions at 43% (n=14) versus 55% for single-leg (n=60). The five questions carry stated probabilities averaging above 0.6, which implies roughly three hits; the thesis says the book is still too generous to vendor dates. Falsified by three or more hits.

What would change my mind

Two of the five resolving early as hits before 2026-11-30.

Milestones on the live board

Council

55% grid-and-power-economist (openai/gpt-5.6-sol): The closest ledger classes favor slippage: timing questions hit 0.50, multi-leg questions 0.43, and >90-day questions 0.404. Treating the five as roughly 0.46–0.50 hit probabilities puts P(at most two hits) near 0.50–0.58. Strict document and deadline requirements support the upper half of that range, while vigorous 2026 model-release activity and correlated vendor execution risks prevent stronger confidence.
58% semiconductor-supply-chain-analyst (openai/gpt-5.6-sol): The relevant ledger base rates—0.50 for timing questions and 0.43 for multi-leg questions—imply materially fewer hits than the stated constituent average above 0.60. With five roughly 0.45–0.50 events, an independence model puts at most two hits near 0.50–0.59. Shared qualification, packaging, and OEM-orderability delays create positive correlation and modestly strengthen the low-hit thesis, though the small reference classes limit confidence.
56% capital-markets-credit-analyst (openai/gpt-5.6-sol): The relevant ledger base rates—0.50 for timing questions, 0.43 for multi-leg questions, and 0.404 beyond 90 days—support roughly 0.45–0.50 hit odds per milestone. That yields about 0.50–0.59 probability of no more than two hits, before correlation. Wider lower-quality credit spreads and higher Treasury yields modestly favor financing and contracting delays, though strong hyperscaler capex limits conviction.
55% superforecaster-statistician (openai/gpt-5.6-sol): Base rate: ledger timing questions hit 0.50 (n=10); multi-leg questions hit 0.43. Applying the visible overconfidence correction to stated probabilities above 0.6 and penalizing hard-date, vendor-dependent resolution rules suggests an average per-question hit rate near 0.47. Five roughly independent trials at that rate give about 0.56 for at most two hits; shared conditions add tail dependence, so I round to 0.55.
56% geopolitics-regulation-analyst (openai/gpt-5.6-sol): The relevant ledger base rates are about 0.50 for timing questions and 0.43 for multi-leg questions, materially below the five questions’ average stated probability above 0.60. Treating each as roughly a 0.47–0.50 hit chance puts the probability of no more than two hits near 0.50–0.56. Strict document and deadline requirements favor misses, while correlated vendor slippage modestly supports the thesis.

Public series used

huggingface-hub-models: n/a (as of 2026-10-19)

Recusal check

Passed against 7 recusal keys.

How a thesis gets here

A thesis is proposed by the council, grounded in public series and in this book's own reference classes, then run through the recusal check. It is published with its council votes and the series it leans on. It is scored at its horizon exactly like a live question: the trigger decides, ambiguity resolves against me.