Trackers
Numbers that change every day, kept current: each tracker re-runs its model on the latest data and shows how its forecasts have moved.
Cyberattack and outage tracker
Major incidents, past 14 days143.3 times the weekly pace before themOf $1B or more, past 12 months1the 2023–26 record is 1.8 a yearMajor incidents a monthEvery major cyberattack and outage reported since January 2026, one row each with its source, counted by week and by month, with a separate count for the world’s largest companies. It watches the number the cyberattack calculator starts from, how often large attacks come, before the dollar estimates arrive.
Anthropic & OpenAI ARR Forecast 2026–2027
Anthropic$86Btoday · $96B end-2026 · $205B end-2027OpenAI$62Btoday · $73B end-2026 · $200B end-2027Both labs’ annualized revenue, estimated daily from every reported figure and, for Anthropic, spend through OpenRouter, with forecasts for the end of 2026 and 2027.
OpenRouter tokens and spend
Tokens a day40Tpast week · 46× a year earlierSpend, a year$5.0Bpast week · 23× a year earlierChinese models69%of tokens, past four weeks · 34% of dollarsUS modelsChinese modelsTokens a day and estimated spend through OpenRouter, split between models from US and Chinese labs, updated daily with forecasts for the end of 2026 and 2027.
GPU spot prices
B300, past 7 days$6.24a GPU-hour · 66% of on-demandH100, past 7 days$2.03a GPU-hour · 45% of on-demandB300H100What it costs to rent B300, B200, H200, H100 and RTX PRO 6000 GPUs on Nebius’s spot market, by region, with seven-day averages. Prices have floated since October 8; updated twice a day.
Frontier launch tracker
Fable 5.26.4%of OpenRouter spend, day 10GPT-6 Astra, best week22.8%of OpenRouter spend · 5.1% by week 6Fable 5.1, best week8.9%of OpenRouter spend · 30% of Anthropic’sFable modelsGPT-6 AstraHow fast the newest, most expensive AI models get adopted, lined up on launch day: Claude Fable 5.2 from its first day, against Fable 5 and 5.1, GPT-6 Astra and the flagships around them.
GPU Compute Price Index
H100, on-demand$3.82a GPU-hour · 23 providersH100, spot$2.49a GPU-hour · 8 providersOn-demandSpotWhat an hour of a data-centre GPU costs to rent across 31 providers, from spot to five-year commitments, for 9 GPUs, by region and type of provider. Collected six times a day, with a daily fixing at 16:00 UTC.
GPU Crack Spreads
B200 on V4.1-Flash+$15.50crack per GPU-hour: $23.37 of tokens less $7.87 rentH200 on V4.1-Flash−$2.88crack per GPU-hour: $2.68 of tokens less $5.56 rentB200 token valueB200 rentWhat an hour of a GPU earns serving open models at the prices OpenRouter’s hosts are paid, less what it costs to rent: a crack spread, as refiners measure theirs, for seven GPUs and three open models, with yields from SemiAnalysis InferenceX. Priced through the day, with a daily fixing at 16:00 UTC.