Exponential Tracker
A graph-first atlas of exponential curves: transistor counts, the cost of a genome, the price of orbit.
55 curves in 9 categories. Sources & Methodology · Guided Tours · Changelog
Computing Moore’s Law — Fifty-five years of doubling, and the lead has changed hands twice.HDD Cost Collapse — The price of keeping bits around, down ~8 billion-fold in real terms.Cost per GFLOPS — The invisible curve behind simulation, graphics, science, and AI.TOP500 #1 Systems — A public scoreboard for the fastest machines humans can assemble.Server Work per Watt — More work from the same watt: 21.3× since 2009, in two long climbs and two stalls.Consumer GPU FP32 Performance — Single-precision throughput, the metric that tracks real architectural improvement, not just transistor count.Datacenter GPU AI Compute — Vendor headline training throughput per GPU, quoted in each generation’s own number format.Quantum Processor Qubits — From 2-qubit lab demos in 1998 to 6,100-qubit arrays today: physical qubits compound, but useful logical qubits are still double-digit.Energy Solar PV Cost — Solar modules fell about 500-fold in real terms, and reversed more than once on the way down.Battery Cell Cost — Lithium-ion cells fell 99% in real terms, the engine under the EV.Global Solar Installed Capacity — From backyard experiment to the fastest-growing energy source in history.Haitz’s Law: LED Lumens per Dollar — The Moore’s Law of lighting: ~10× per decade in lumens per dollar, half that after 2014.UK Cost of Light — The fuel cost of a fixed quantity of light fell 13,800-fold from 1300; the last half-century is almost flat.Global EV Sales — From a niche curiosity to over 20 million per year: the transport revolution in numbers.Grid Battery Storage — The missing piece of the renewable grid: cumulative deployed capacity going near-vertical.Global Wind Power Capacity — The other half of the renewable transition: onshore and offshore, scaling 76× since 2000.Fusion Energy Gain — The best energy gain any fusion experiment has achieved: past scientific breakeven in 2022, and setting records almost yearly since.AI AI Training Compute — The biggest training run ever done, doubling every five and a half months.Cost of o1-Level Reasoning — Frontier reasoning cost 150× less within nineteen months of arriving.AI Math Reasoning (Mock AIME) — Competition mathematics: from under 1% to 100% in three years.FrontierMath — Unpublished problems, vetted by mathematicians: eighteen of the 285 are left.AI Autonomy Horizon (METR) — How long a human-expert task can an AI complete alone? Doubling every ~6 months for 7 years.SWE-bench Verified — Real GitHub issues, attempted with no human in the loop; the score is the share fixed.HuggingFace Models — Public ML models hosted on HuggingFace: the "GitHub for AI" curve no closed lab can capture.Open-Weight AI Frontier — The best model anyone can download, 5.3 points behind the closed frontier when it landed.ChatGPT Weekly Users — From 1 million users in five days to a billion a week: among the fastest consumer-technology adoptions ever measured.Knowledge arXiv Preprints — From 306 papers in 1991 to 284,000 a year: the speed of open scientific publishing.Total GitHub Repositories — From a private beta in 2008 to over 630 million repositories: the shared codebase of civilisation.Annual Biomedical Papers (PubMed) — Three papers per minute, every minute, year-round. The publishing curve underneath every clinical decision.English Wikipedia Articles — An exponential that is visibly running out of new things to write about.Data Created per Year — Everything created, captured, copied, and consumed in a year: 2 zettabytes in 2010, 64 by 2020, a forecast 181 for 2025.Economy World GDP — Two thousand years of economic output on one axis: near-flat for eighteen centuries, then vertical.GDP per Capita — Eighteen hundred years stuck near $1,100 per person, then a breakout: 14× since 1820.
The atlas itself — the charts, the overlays, the guided tours — is drawn in your browser and needs JavaScript. Every curve above is a page of its own with its full data table.