Follow a score back to its evidence.
These are editorial, coverage-dependent indicators. They are not scientific measures of AI progress, population sentiment, investment merit, or price direction.
Reviewed intelligence snapshot · · Market data refreshes automatically
Heuristic · 7/11 components · 15 events · 16 sources / 10 publishers
24h: not established · 7d: not established · 30d: not established · 90d: not established →AI Doomer Pressure74/ 100Heuristic · 5/9 components · 11 events · 11 sources / 10 publishers
24h: not established · 7d: not established · 30d: not established · 90d: not established →Narrative divergenceUnavailablePressure / investment balanceThree distinct events and two sources per side required. No sentiment survey or price series.
Inspect evidence & limitations →How evidence becomes intelligence
- Official lab, policy, open-source and network sources join the existing News, Signals, project-record and source-review systems.
- Normalize the original URL, publisher, dates and entity. Social claims remain leads.
- Candidate ranking considers novelty, materiality, source quality, evidence completeness, narrative and exposure relevance. A separate factual review must confirm materiality, confidence and lineage before promotion (minimum 10 points). Ranking alone never publishes a catalyst.
- Classify narrative effects and potential exposure as research interpretation, separate from the factual summary.
- Deduplicate event keys and retire superseded versions from current scoring.
Primary documents can contain company claims. Verifying a release is not independently validating its benchmarks, promised capacity or adoption. Detected time is when this desk recorded the event, not when the world first learned it.
Conservative scoring
Window: 30 calendar days by event date, excluding events not yet detected, future events, repeated events and superseded versions. Each effect has direction ±1 and analyst-assigned strength 1–3. Evidence weights: primary record 1.0; reviewed secondary 0.6; company/lab claim 0.5; unverified lead 0.15. Confidence weights: high 1.0, medium 0.65, low 0.3.
Narrative = round(50 + 50 × Σ(direction × strength/3 × weight) / Σ(weight))
No evidence returns unavailable, not 50. A high score describes the direction of the selected evidence, not its representativeness. Source and event counts are shown beside it. Composite indices average covered components equally only with at least three distinct events and two covered components; missing components are excluded and coverage remains visible. Components are correlated: this is not independent statistical confirmation.
AI Acceleration Index
15 distinct events · 16 sources / 10 publishers · 7/11 components covered.
AI acceleration · 77 · 5 events
Google commits €13B to Finnish infrastructure and energy
Anthropic releases Fable 5.1 with a separate trusted-access configuration
NVIDIA reports $89B quarterly Data Center revenue
AMD and Cisco systems begin serving HUMAIN customers in Saudi Arabia
Oracle reports rapid infrastructure growth alongside negative free cash flow
Capex acceleration · 87 · 5 events
Google commits €13B to Finnish infrastructure and energy
NVIDIA reports $89B quarterly Data Center revenue
Oracle reports rapid infrastructure growth alongside negative free cash flow
NVIDIA and Australian partners announce up-to-2GW infrastructure buildout
Mistral announces EUR3 billion Series D for frontier research and sovereign AI
Frontier-model progress · 83 · 1 events
Anthropic releases Fable 5.1 with a separate trusted-access configuration
Inference growth · 83 · 2 events
Anthropic releases Fable 5.1 with a separate trusted-access configuration
d-Matrix selects NVIDIA NVLink Fusion for planned Raptor integration
Frontier-training growth · Unavailable · 0 events
No eligible evidence in this snapshot.
AI-agent adoption · 83 · 2 events
OpenAI launches Agents API public beta for managed cloud agents
OpenAI opens GPT-Live-1 full-duplex voice model to API developers
Decentralization · Unavailable · 0 events
No eligible evidence in this snapshot.
AI adoption pathways · 74 · 6 events
UK healthcare AI commission proposes staged authorizations
Mistral and Cloudera plan inference inside customer-controlled environments
AMD and Cisco systems begin serving HUMAIN customers in Saudi Arabia
Epoch/Ipsos survey finds increased frequent US adult AI use
NVIDIA and Palantir report sovereign AI deployment in NVIDIA's supply chain
Google, NVIDIA and Emerald AI launch a flexible-power coalition
Open-model progress · 67 · 1 events
Mistral and Cloudera plan inference inside customer-controlled environments
Lower inference cost · Unavailable · 0 events
No eligible evidence in this snapshot.
Power expansion · Unavailable · 0 events
No eligible evidence in this snapshot.
AI Doomer Pressure Index
11 distinct events · 11 sources / 10 publishers · 5/9 components covered. Pressure measures observed restriction/slowdown themes, not a judgment about safety policy.
AI doom / slowdown · 67 · 1 events
Anthropic revises its assessment of evaluation incidents and adds a fourth case
AI safety pressure · 83 · 3 events
Anthropic revises its assessment of evaluation incidents and adds a fourth case
Anthropic reports surveillance misuse and limits of account enforcement
Mistral discloses high-severity Vibe permission flaws and fixes
Regulatory restriction · 81 · 6 events
Clearfield pauses data-center development while writing land-use rules
UK healthcare AI commission proposes staged authorizations
OpenAI endorses four California AI bills and mandatory federal safety rules
House passes bill aimed at addressing data-center energy costs
Virginia unveils a data-center accountability framework
Canada launches responsible data-center development principles
Open-source restriction · Unavailable · 0 events
No eligible evidence in this snapshot.
AI labor displacement · Unavailable · 0 events
No eligible evidence in this snapshot.
Capex slowdown · Unavailable · 0 events
No eligible evidence in this snapshot.
Model / export access restrictions · 67 · 1 events
Anthropic reports surveillance misuse and limits of account enforcement
AI incidents / public backlash · Unavailable · 0 events
No eligible evidence in this snapshot.
AI infrastructure bottlenecks · 74 · 6 events
Google commits €13B to Finnish infrastructure and energy
Clearfield pauses data-center development while writing land-use rules
House passes bill aimed at addressing data-center energy costs
Virginia unveils a data-center accountability framework
Google, NVIDIA and Emerald AI launch a flexible-power coalition
Canada launches responsible data-center development principles
Divergence and history
Pressure versus investment uses the same 30-day window. Each side requires at least three events not shared with the other side, two sources from two publishers and two covered components. The balance subtracts pressure direction from investment direction. It is not public sentiment, statistical significance or a trend; no “rising” label follows from a single snapshot.
Divergence unavailable — insufficient comparable evidence. Investment evidence: 5 events; pressure evidence: 11 events.
History starts with the first retained observation. We do not backdate today's knowledge or invent a weekly/monthly delta. Full catalyst state, sources, component coverage, compute and fundamentals are retained in immutable observations. 24h/7d/30d/90d index changes require the same method version and covered component set; changing methodology starts a new comparable series. Old scores remain as originally observed.
| Observed snapshot | Acceleration | Pressure |
|---|---|---|
| 2026-09-15T20:23:56.426Z | 83 | Unavailable |
| 2026-09-15T21:02:48.652Z | 77 | 72 |
| 2026-09-15T21:04:45.478Z | 77 | 72 |
| 2026-09-15T21:21:57.209Z | 78 | 72 |
| 2026-09-15T21:54:03.049Z | 79 | 72 |
| 2026-09-15T22:39:04.034Z | 79 | 73 |
| 2026-09-15T23:10:14.431Z | 79 | 73 |
| 2026-09-18T16:12:15.136Z | 79 | 71 |
Compute and fundamentals
Prices are USD per GPU-hour where explicitly normalized. A rate-card observation is not inventory confirmation. Discounts require a reviewed comparison group, both market classes, observed inventory and quotes no older than seven days. CPU/RAM, storage, networking, tenancy, region, support and contract terms must be comparable. A stale article cannot be a current quote.
Protocol-reported, independently verified, estimated and unavailable values remain distinct. Leased GPU allocation is not hardware utilization. Customer spend, protocol revenue, fees and emissions are different metrics. Cumulative counts are not current active users. No valuation ratios are generated without aligned definitions and observation periods.
Distribution and freshness
The public API exposes reviewed research, not trading instructions. Alerts require high confidence, primary evidence and an explicit distribution approval. No external channel is enabled by adding a catalyst. Existing posting controls and deduplication remain authoritative.
Distribution contract ↗