{ "schema_version": "1.0.0", "topic": "Is AI a Bubble?", "question_revision": 1, "question": "Is AI a Bubble?", "report_id": "first-baseline", "evidence_checked_on": "2026-09-11", "check_timezone": "Asia/Bangkok", "source_dates_are_not_publication_dates": true, "observations": [ { "id": "evidence:r01-msft", "classification": "reported_finding", "statement": "Microsoft reports positive quarterly revenue and GAAP operating income.", "source_ids": [ "source:msft-fy26q4" ], "period_or_event_date": "three months ended 2026-06-30", "scope": "Consolidated company, nominal USD millions", "limitations": "Not AI-only profit, cash flow or return on newly invested capital.", "confidence": "high_as_attributed", "values": { "revenue": 90007, "gaap_operating_income": 40603, "currency": "USD", "unit": "million" } }, { "id": "evidence:r01-nvidia", "classification": "reported_finding", "statement": "NVIDIA reports positive quarterly revenue and GAAP operating income.", "source_ids": [ "source:nvidia-fy27q2" ], "period_or_event_date": "quarter ended 2026-07-26", "scope": "Consolidated company, nominal USD millions", "limitations": "Supplier earnings do not establish the return earned by compute buyers.", "confidence": "high_as_attributed", "values": { "revenue": 96221, "gaap_operating_income": 63734, "currency": "USD", "unit": "million" } }, { "id": "evidence:r01-amazon-investment", "classification": "announced_terms", "statement": "Amazon investment announcement splits an initial amount and a later conditional amount.", "source_ids": [ "source:openai-amazon" ], "period_or_event_date": "2026-02-27", "scope": "Investment leg, nominal USD billions", "limitations": "Initial and conditional are announcement descriptions, not confirmation of each cash transfer.", "confidence": "high_as_attributed", "values": { "total": 50, "initial": 15, "conditional": 35, "cash_paid_verified": null, "currency": "USD", "unit": "billion" } }, { "id": "evidence:r01-aws-procurement", "classification": "announced_terms", "statement": "OpenAI announces an incremental eight-year AWS purchase expansion, distinct from the investment leg.", "source_ids": [ "source:openai-amazon" ], "period_or_event_date": "2026-02-27", "scope": "Incremental services commitment, nominal USD billions", "limitations": "Not annual revenue, cash paid, delivered compute or traced use of Amazon investment proceeds.", "confidence": "high_as_attributed", "values": { "incremental_amount": 100, "term_years": 8, "cash_paid": null, "recognized_revenue": null, "currency": "USD", "unit": "billion" } }, { "id": "evidence:r01-round-close", "classification": "issuer_reported_close", "statement": "OpenAI reports a March 31 round close with $122 billion in committed capital and an $852 billion post-money valuation.", "source_ids": [ "source:openai-march-round" ], "period_or_event_date": "2026-03-31", "scope": "Aggregate round; nominal USD billions", "limitations": "Replaces the February headline for this round, not another $122 billion to add; individual cash receipt schedule not verified. Valuation is a financing term, not intrinsic value.", "confidence": "high_as_attributed", "values": { "committed_capital": 122, "post_money_valuation": 852, "cash_received_verified": null, "currency": "USD", "unit": "billion" } }, { "id": "evidence:r01-revolver", "classification": "issuer_reported_facility", "statement": "OpenAI describes an approximately $4.7 billion revolving facility as undrawn at the March close.", "source_ids": [ "source:openai-march-round" ], "period_or_event_date": "2026-03-31", "scope": "Facility limit versus drawings; nominal USD billions", "limitations": "Zero drawings is specific to the announced closing date, not a claim about current drawings; covenants, security and maturity remain unverified.", "confidence": "high_as_attributed", "values": { "facility_limit_approx": 4.7, "drawn_at_reported_close": 0, "current_drawn": null, "currency": "USD", "unit": "billion" } }, { "id": "evidence:r01-financing-target", "classification": "announced_target", "statement": "NVIDIA and six financial institutions announce memorandums aiming to mobilize more than $500 billion over time; final agreements remain required in that release.", "source_ids": [ "source:nvidia-financing" ], "period_or_event_date": "2026-08-10", "scope": "Third-party compute financing target, nominal USD billions", "limitations": "Not a committed fund size, borrower debt balance, deployed capital or final lender terms.", "confidence": "high_as_attributed", "values": { "target_lower_bound_exclusive": 500, "committed": null, "funded": null, "currency": "USD", "unit": "billion" } }, { "id": "evidence:r01-support", "classification": "research_finding", "statement": "A staggered support-tool rollout reports 15% more issues resolved per hour on average in a 5,172-agent study.", "source_ids": [ "source:ai-at-work-v2", "source:ai-at-work-html-v2" ], "period_or_event_date": "main rollout fall 2020/winter 2021; paper version 2024-11-06", "scope": "One enterprise software support setting; main estimate difference-in-differences", "limitations": "Not a universal productivity multiplier or directly observed monetary return; causal interpretation relies on rollout assumptions. Small randomized pilot is distinct from main sample.", "confidence": "high_as_attributed", "values": { "sample_agents": 5172, "reported_mean_change_percent": 15, "outcome": "issues resolved per hour", "design": "staggered observational rollout / difference-in-differences" } }, { "id": "evidence:r01-developer", "classification": "research_finding", "statement": "A randomized study of experienced open-source developers using early-2025 tools found 19% longer task completion time with AI access.", "source_ids": [ "source:metr-2025" ], "period_or_event_date": "early 2025; published 2025-07-10", "scope": "16 developers, 246 real tasks in familiar repositories", "limitations": "Task duration is the denominator; not a 19% economy-wide productivity decline or a current-tool conclusion.", "confidence": "high_as_attributed", "values": { "sample_developers": 16, "tasks": 246, "time_change_percent": 19, "outcome": "task completion time", "design": "randomized field experiment" } }, { "id": "evidence:r01-developer-follow-up", "classification": "methodological_finding", "statement": "METR's February update says selection and time-measurement problems make its newer speedup estimate unreliable.", "source_ids": [ "source:metr-2026" ], "period_or_event_date": "published 2026-02-24; late-2025 follow-up", "scope": "Task/participant selection and concurrent-agent measurement", "limitations": "Weakens extrapolation from either old or newer raw estimates; not a precise causal replacement estimate.", "confidence": "high_as_attributed", "values": {} }, { "id": "evidence:r01-survey", "classification": "self_reported_finding", "statement": "A convenience survey of 349 technical workers reports medians of 1.4–2 times perceived work value across three questions.", "source_ids": [ "source:metr-survey-2026" ], "period_or_event_date": "survey February–April 2026; published 2026-05-11", "scope": "Self-reported counterfactual value of work, largely around March 2026", "limitations": "The range is across question wordings, not a confidence interval; not a randomized output effect, measured cash savings or representative workforce estimate.", "confidence": "high_as_attributed", "values": { "analyzed_respondents": 349, "raw_unique_respondents": 359, "excluded_respondents": 10, "median_value_multiplier_across_questions_min": 1.4, "median_value_multiplier_across_questions_max": 2, "range_type": "different question formulations", "design": "convenience self-report survey" } }, { "id": "evidence:r01-power", "classification": "modeled_scenario", "statement": "IEA's April 2025 base case projects about 945 TWh of global data-centre electricity demand in 2030.", "source_ids": [ "source:iea-energy", "source:iea-energy-landing" ], "period_or_event_date": "forecast vintage 2025-04-10; target year 2030", "scope": "All data centres globally, AI and non-AI", "limitations": "Forecast is not observed consumption or energized capacity; no claim this is the newest available forecast.", "confidence": "high_as_attributed", "values": { "forecast_about": 945, "unit": "TWh per year", "target_year": 2030, "vintage": "2025-04-10" } }, { "id": "evidence:r01-assessment", "classification": "inference", "statement": "The inspected record supports real activity and scoped usefulness, but does not settle whole-chain payback or valuations; localized overinvestment and a productive buildout remain compatible.", "source_ids": [ "source:msft-fy26q4", "source:nvidia-fy27q2", "source:ai-at-work-html-v2", "source:openai-amazon", "source:nvidia-financing" ], "period_or_event_date": "research baseline 2026-09-11 Asia/Bangkok", "scope": "Selected companies, transactions and studies—not a sector-wide investment judgment", "limitations": "No current-price model, causal systemic-risk estimate, crash date or probability produced.", "confidence": "bounded_inference", "values": {} } ], "calculations": [ { "id": "calculation:r01-msft-margin", "label": "Microsoft GAAP operating margin", "formula": "GAAP operating income / revenue * 100", "inputs": { "gaap_operating_income": 40603, "revenue": 90007, "currency": "USD", "unit": "million" }, "evidence_ids": [ "evidence:r01-msft" ], "unrounded_percent": "45.11093581610319197395758108", "display_percent": "45.11", "rounding": "Decimal; half-up to 2 decimal percentage points", "interpretation": "Company-wide operating margin for its own reported quarter, not free cash flow or return on invested capital. No combined sector total." }, { "id": "calculation:r01-nvidia-margin", "label": "NVIDIA GAAP operating margin", "formula": "GAAP operating income / revenue * 100", "inputs": { "gaap_operating_income": 63734, "revenue": 96221, "currency": "USD", "unit": "million" }, "evidence_ids": [ "evidence:r01-nvidia" ], "unrounded_percent": "66.23710000935346753827127134", "display_percent": "66.24", "rounding": "Decimal; half-up to 2 decimal percentage points", "interpretation": "Company-wide operating margin for its own reported quarter, not free cash flow or return on invested capital. No combined sector total." }, { "id": "calculation:r01-amazon-split", "label": "Check investment announcement split", "formula": "15 + 35", "inputs": { "initial_announced": 15, "conditional_announced": 35, "currency": "USD", "unit": "billion" }, "result": 50, "evidence_ids": [ "evidence:r01-amazon-investment" ], "interpretation": "Reconciles announced components only; does not establish cash settlement." }, { "id": "calculation:r01-survey-count", "label": "Check analyzed survey count", "formula": "359 - 10", "inputs": { "unique_responses": 359, "excluded": 10, "unit": "respondents" }, "result": 349, "evidence_ids": [ "evidence:r01-survey" ], "interpretation": "Arithmetic consistency with researcher report, not a replication of filtering choices." } ], "sources": [ { "id": "source:msft-fy26q4", "title": "Microsoft FY2026 Q4 results", "url": "https://www.microsoft.com/en-us/Investor/earnings/FY-2026-Q4/press-release-webcast", "published_at": "2026-07-29", "publication_date_precision": "day", "inspection_scope": "Quarterly GAAP rows and income statements; selected operating commentary", "locator": "GAAP quarterly reconciliation and unaudited income statements, quarter ended 30 June 2026", "limitations": "Company-reported, unaudited release; not an audit or an AI-only income statement. No customer cash-flow reconciliation." }, { "id": "source:nvidia-fy27q2", "title": "NVIDIA FY2027 Q2 results", "url": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027", "published_at": "2026-08-26", "publication_date_precision": "day", "inspection_scope": "GAAP summary and consolidated income statement; financing highlight", "locator": "Q2 FY27 GAAP summary; quarter ended 26 July 2026", "limitations": "Issuer disclosure, not independent confirmation of customers' returns; GAAP rather than adjusted operating income used." }, { "id": "source:openai-amazon", "title": "OpenAI and Amazon announce strategic partnership", "url": "https://openai.com/index/amazon-partnership/", "published_at": "2026-02-27", "publication_date_precision": "day", "inspection_scope": "Investment split and incremental AWS procurement terms", "locator": "Opening investment paragraph; Expanding infrastructure commitments", "limitations": "Original announcement, not proof of individual payments, current conditions or delivered services; group names not a legal-entity tree." }, { "id": "source:openai-feb-round", "title": "Scaling AI for everyone", "url": "https://openai.com/index/scaling-ai-for-everyone/", "published_at": "2026-02-27", "publication_date_precision": "day", "inspection_scope": "Original February funding headline compared with March close", "locator": "Opening funding paragraph", "limitations": "Historical $110 billion announcement, not an additional round to add to March; not a current funding total." }, { "id": "source:openai-march-round", "title": "OpenAI raises $122 billion to accelerate the next phase of AI", "url": "https://openai.com/index/accelerating-the-next-phase-ai/", "published_at": "2026-03-31", "publication_date_precision": "day", "inspection_scope": "Opening closing statement, named participants, credit-facility paragraph", "locator": "Opening; Deep conviction across global capital", "limitations": "Closed round is stated in terms of committed capital; no individual draw/payment schedule, full investment agreement, debt covenants or audit inspected. Not verified as the latest subsequent financing event." }, { "id": "source:nvidia-financing", "title": "NVIDIA partnerships for compute financing platforms", "url": "https://nvidianews.nvidia.com/news/nvidia-partners-with-apollo-blackrock-blackstone-brookfield-goldman-sachs-and-kkr-to-establish-ai-compute-infrastructure-financing-platforms-to-mobilize-over-500-billion-of-third-party-capital", "published_at": "2026-08-10", "publication_date_precision": "day", "inspection_scope": "Opening target, signed memorandums, final-agreement condition", "locator": "Opening paragraphs and sentence before About NVIDIA", "limitations": "Target to mobilize third-party capital over time, not executed facilities or funds deployed. No current settlement or vehicles verified." }, { "id": "source:metr-2025", "title": "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity", "url": "https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/", "published_at": "2025-07-10", "publication_date_precision": "day", "inspection_scope": "Study summary, design, sample, treatment and stated generalization limits", "locator": "Summary; study design; scope limitations", "limitations": "Small randomized field sample of experienced developers in familiar repositories with early-2025 tools; not economy-wide or current-tool estimate. Raw data not reanalyzed." }, { "id": "source:metr-2026", "title": "We are Changing our Developer Productivity Experiment Design", "url": "https://metr.org/blog/2026-02-24-uplift-update/", "published_at": "2026-02-24", "publication_date_precision": "day", "inspection_scope": "Selection, task refusal, concurrent-agent timing and conclusions", "locator": "Why the estimate is unreliable; measurement discussion", "limitations": "Follow-up changes confidence in present-tense application; does not erase the earlier randomized result or establish a precise current speedup." }, { "id": "source:metr-survey-2026", "title": "Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker Productivity", "url": "https://metr.org/blog/2026-05-11-ai-usage-survey/", "published_at": "2026-05-11", "publication_date_precision": "day", "inspection_scope": "Summary, convenience sampling, question definitions and result limitations", "locator": "Summary; Methodology; Productivity gains around March 2026", "limitations": "Convenience survey; low email response rate, selection and counterfactual self-report. Value measures are not measured output, wages or realized financial returns; no respondent-level reanalysis." }, { "id": "source:ai-at-work-v2", "title": "Generative AI at Work, version 2", "url": "https://arxiv.org/abs/2304.11771", "published_at": "2024-11-06", "publication_date_precision": "day", "inspection_scope": "Version and abstract checked; expanded through separately registered HTML full text", "locator": "Version 2 abstract", "limitations": "Version date differs from deployment dates; full statistical replication not performed." }, { "id": "source:ai-at-work-html-v2", "title": "Generative AI at Work, version 2 HTML", "url": "https://arxiv.org/html/2304.11771v2", "published_at": "2024-11-06", "publication_date_precision": "day", "inspection_scope": "Abstract and sections 2–3 on setting, rollout, sample and difference-in-differences; generalization limits", "locator": "Sections 2.2, 2.3, 3.1–3.3 and abstract", "limitations": "Main rollout is observational, with identifying assumptions, not a randomized assignment of all 5,172 workers; one firm/client setting; no macro or wage inference. Underlying data not replicated." }, { "id": "source:iea-energy", "title": "IEA Energy and AI: executive summary", "url": "https://www.iea.org/reports/energy-and-ai/executive-summary", "published_at": "2025-04-10", "publication_date_precision": "day", "inspection_scope": "2025-vintage global data-centre estimate and base-case forecast, grid constraint discussion", "locator": "Electricity demand for data centres more than doubles by 2030", "limitations": "All data centres, not AI alone; modeled 2030 demand, not delivered capacity. Reinspection does not make this the latest forecast vintage." }, { "id": "source:iea-energy-landing", "title": "IEA Energy and AI report metadata", "url": "https://www.iea.org/reports/energy-and-ai", "published_at": "2025-04-10", "publication_date_precision": "day", "inspection_scope": "Publication date and links to online executive summary", "locator": "Published 10 April 2025", "limitations": "Metadata only; PDF not needed or opened; unrelated embedded agent instructions ignored." } ], "aggregation_rules": [ "Do not add company sales along a supply chain into sector demand.", "Do not add equity, purchase commitments, facility limits and platform targets.", "Do not add the February and March versions of the same funding round.", "Do not average heterogeneous task outcomes or self-reported value." ], "unknown_means": "not established by this record; not zero", "unresolved": [ "Current participant payments and debt drawings", "Collateral, maturity and covenants", "Customer retention and unsubsidized demand", "Comparable cash investment and returns on capital", "Current valuation expectations", "Latest forecast vintage and delivered physical capacity", "Non-US competition and lower-cost substitutes" ] }