Every meter is a modeled estimate, grouped by usage, compute, energy, capital, labor, and research. Every number on this page is a modeled estimate smeared from a public print — not a live API, not a hard claim.
AI tokens · modeled estimate · today
Tokens generated today
—
~4,398,148,148 / second
Modeled global inference across disclosed floors and triangulated channels. Tokens from different models are not a single physical unit — tokenizers and input/output/reasoning/multimodal accounting differ. Treat the token count as a directional index, not barrels of oil.
About 4.4 billion tokens every second — War and Peace, 7,800 times, every second.
Source: Tokens Per Day modeled total (~360T/day, Jul 2026), grown to a September run-rate.
Queries · modeled estimate · today
Queries modeled today
—
~97,222 / second
OpenAI publicly reported ChatGPT at more than 2.5 billion messages per day. Epoch later modeled ChatGPT closer to ~4 billion daily messages — an estimate, not the same kind of disclosed number. Our global query meter is a modeled stack on top of the disclosed floor, plus Gemini, Claude, Grok, and the open-weight fog.
A smear of modeled conversations, not a live query log.
Source: OpenAI disclosed >2.5B ChatGPT messages/day; Epoch ~4B estimate for ChatGPT; AiSignal models the rest of the stack. Not a utility meter.
Images · modeled estimate · today
Images generated today
—
~3,241 / second
Still images are the bulk of generative media. OpenAI users alone were producing on the order of 78 million images a day; the long tail of open models is larger still.
More than 3,200 pictures a second — a new photograph every blink.
Source: OpenAI image-volume reporting plus open-weight triangulation, 2026.
People using AI · modeled estimate · today
People who used generative AI today
—
~6,019 / second
Standalone tools passed a billion monthly users; embedded AI in search, mail, and office software lifts the wider net toward 1.5 billion. Daily reach is modeled from that base.
About one in sixteen humans, every day.
Source: DataReportal Digital 2026; Microsoft Global AI Diffusion; OpenAI MAU, Jun 2026.
AI accelerators · modeled estimate · stock
H100-equivalents installed
—
~2 / hour
Global operational AI compute, normalized to NVIDIA H100s. Nvidia still holds the majority; Google TPUs and Amazon Trainium make up much of the rest.
Growing by about one H100-equivalent every 1.5 seconds.
Source: Stanford AI Index 2026 — 17.1M H100-equivalents, 3.3×/year since 2022.
New accelerators · modeled estimate · today
New H100-equivalents today
—
~2 / hour
The derivative of the fleet. Clusters light up in Texas, Memphis, Nevada, Inner Mongolia, and on campuses that did not exist two years ago.
A new supercomputer's worth of accelerators, every morning.
Source: Implied by Stanford AI Index 2026 growth rate applied to mid-year stock.
Top fleets · modeled estimate · stock
Accelerators in the ten largest fleets
—
~2 / minute
Google, Meta, xAI Colossus, Microsoft, OpenAI/Stargate, Amazon, CoreWeave, ByteDance, Oracle, Tesla — a first-pass sum of public cluster estimates.
xAI's Memphis site alone is the largest AI cluster on Earth.
Source: gradually.ai cluster estimates, Jun 2026 (Colossus 555k at a single site).
Electricity · modeled estimate · today
Electricity used by AI today
—
~5,920 / second
AI data-center IT capacity around 29.6 GW. We meter a 72% average load — training clusters spike, inference idles, cooling is extra.
About 21 GW continuous — a mid-sized country's evening demand.
Source: Stanford AI Index 2026 (29.6 GW AI data-center power capacity).
CO₂ · modeled estimate · today
CO₂ from AI electricity today
—
~2,072 / second
Uses 0.35 kg CO₂ per kWh, a blended grid factor. Real footprints vary wildly by site: hydro in Quebec, gas in Memphis, coal in parts of Asia.
Grok 4's reported training emissions: 72,816 tons CO₂e — now a few hours of inference.
Source: Capacity from Stanford AI Index 2026; grid factor is a AiSignal estimate.
Home equivalent · modeled estimate · stock
Homes the AI grid could power
—
~5 / minute
AI IT load versus a 1.2 kW average U.S. household draw. Cooling and site overhead are not fully included, so this is a lower bound on disruption.
On the order of New York State at peak, if you include the cooling plant.
Source: Stanford AI Index 2026 capacity; EIA household electricity intensity.
AI spend · modeled estimate · year
Worldwide AI spend this year
—
~$82,072 / second
Gartner's May 2026 forecast for worldwide AI spending, including infrastructure. Software-only market tallies are 3–4× smaller. Scope is the whole fight.
About $82,000 every second of 2026.
Source: Gartner, Worldwide AI Spending Forecast, May 2026 (+47% YoY).
U.S. investment · modeled estimate · stock
U.S. private AI investment, 2025 vintage
—
Snapshot · not a live derivative
Stanford's 2025 figure, still the last full-year print. 2026 is running hotter — Crunchbase logged ~$300B of global VC in Q1 2026 alone.
More than 23× reported private investment in China.
Source: Stanford AI Index 2026. China private: $12.4B — an undercount of state funds.
Hyperscaler capex · modeled estimate · year
Hyperscaler capex this year (modeled)
—
~$13,309 / second
Amazon, Microsoft, Google, Meta, Oracle — the five that pour slabs. Modeled from 2025 run-rates plus disclosed 2026 guides, AI-heavy.
More than the GDP of Denmark, spent on rooms that hum.
Source: Company 10-K/earnings guides; Exponential View 2026 synthesis.
Wage premium · modeled estimate · stock
Wage premium for AI-skilled workers
—
Snapshot · not a live derivative
Not a ticker — a cliff. PwC's jobs barometer found the premium more than doubled in a year as employers priced a skill the schools could not mint fast enough.
25% in 2024. 56% in 2025. The market is running ahead of the workforce.
Source: PwC AI Jobs Barometer 2025, nearly one billion job ads.
Agent-equivalents · modeled estimate · stock
Fully autonomous agent-equivalents
—
~9 / hour
Not chatbots — the theoretical number of full-time knowledge workers the current token factory could replace if every token went to autonomous work. Tiny today. Not tiny in 2028.
~1,300 now; ~1.2 million by 2028 in the median scenario.
Source: The Neuron, AI Economics 101 (Mar 2026), token-to-salary conversion.
Knowledge workers · modeled estimate · year
Knowledge workers using AI this year
—
~13 / second
Enterprises reporting AI in at least one function hit 88% in late 2025. We meter individual knowledge workers who opened a copilot this year, not firms.
Most of them are still using it as a faster intern, not an agent.
Source: McKinsey Global Survey Nov 2025; Microsoft diffusion; AiSignal modeling.
AI papers · modeled estimate · year
AI papers posted this year
—
~10 / hour
cs.AI, cs.LG, cs.CL and cs.CV combined, modeled from 2025 daily rates. The archive is no longer readable by a human. It is barely readable by a model.
A paper every six minutes, including Christmas.
Source: arXiv category volumes, 2025–26 run-rate.
Frontier launches · modeled estimate · year
Frontier-class model launches this year
—
Models that land on the Artificial Analysis Intelligence Index above 40, or equivalent public evals. 2026 is the year six labs were on the 50-point line at once.
Four frontier launches in eight days this July.
Source: Artificial Analysis launch log, 2026; lab blogs.