Trend and distribution views for the current scope — set the provider above with the scope chips.
Severity breakdown (incidents by severity · counted by start date)
Incidents per month
Time to resolve (log scale · columns = severity · points = provider)
criticalmajorminornone
Components by average time to resolve (top 15 · colour = provider · click a bar to filter)
No resolved incidents with components in this range.
Average incident time by week (provider × impact)
Smoothing
No incidents with a duration in scope.
Each line is one provider × impact. "Weekly" is each week's average incident duration; "4-wk running avg" smooths those weekly averages into a running monthly trend.
Provider
Date
Karma
Comments
Title
Author
Link
Downdetector comments per day
User-reported problem counts. Spikes correlate with outage days on the status page.
Provider
Time
Author
Comment
Provider postmortems — Anthropic engineering-blog write-ups and OpenAI status-page incident write-ups. Filtered by the scope chip group above. "Incidents covered" counts status-page incidents linked to each postmortem: direct matches share the root cause; likely-related are dated within the same window with overlapping components.
Provider
Published
Title
Scope
Incidents covered
Source
No postmortems for the current scope.
Provider
Date
Category
Title
Models
Source
Cumulative Anthropic funding raised
Each step shows a funding round closed; line tracks cumulative dollars committed (USD billions). Strategic rounds from Amazon and Google included.
Models — what cards disclose + Epoch AI training-compute estimates
"Disclosed" columns come from each lab's system/model cards (Anthropic + OpenAI). "Estimate (FLOPs)" comes from Epoch AI's database; bounds shown where available. Neither lab publishes training compute, parameter counts, or hardware for recent models, so those cells are intentionally blank — OpenAI discloses even less than Anthropic. Epoch's estimates exist mainly for older, well-studied models (e.g. GPT-4, GPT-3).
Provider
Model
Released
Cutoff
Hardware mentions (card)
Epoch est. FLOPs
Est. cost (2023 USD)
Est. method
Card
Compute capacity & partnership events
Date
Partner
Type
Headline
Capacity
$ committed
Source
Funding rounds
Date
Round
Amount ($B)
Post-money valuation ($B)
Lead investors
Source
Concurrent AWS & GCP outages — for correlation only
AWS = curated list of major outages affecting us-east-1 / us-east-2 / us-west-2 / global. GCP = full status.cloud.google.com/incidents.json archive. Time-window overlap with a frontier-lab incident is correlation, not causation.
Date
Provider
Severity
Regions
Services
Summary
Source
Infrastructure dependency graph
Frontier labs and the infrastructure they depend on — clouds, chip and memory vendors, optics, fiber, power, cooling, colocation, and investors. Edges are hand-curated from research drops (provider_relationships), not scraped. A solid edge is a documented relationship; a dashed edge is inferred from secondary reporting. Edge thickness tracks risk weight — how concentrated the dependency is. This view ships infrastructure dependencies only; model and data-center nodes are on the roadmap. Drag nodes to reshape the graph; click any node or edge for detail.
No relationship graph data in this build.
Incident pressure by hour of week (trailing 90 days · cell intensity = total active-incident time; overlaps intensify)
Incidents active per day (trailing 12 months · multi-day incidents count on every day they span)
Status-update words per day (trailing 12 months · words counted on the day each update was posted)
Component-impacts per day (trailing 12 months · affected components summed across incidents active that day)