Supercomputers/Analysis
Derived views
Analysis
Fourteen views that cannot be produced from a ranked list as it exists, because a ranked list is system-centric and these are all questions about companies, countries, generations and time. Each will ship as a static page with a chart, a written read, a “how this was calculated” expander, and a downloadable CSV of our derived output: never of the source list.
All fourteen are live. They ship now because the dataset spans 1993 to 2026, which is enough for a time series to mean something rather than to look like one. The last to ship, the shadow list, waited on something no amount of chart work substitutes for: a published estimator, and the judgement that a number built from uncertain inputs is only worth publishing if every input is traceable. It uses no vendor peak figures, only measured rates from systems in this dataset, and it names the one substitution it makes.
Every live page carries a chart, a written read, a “how this was calculated” expander that states the attribution rules, a table view of the same numbers, and a CSV of our derived output: never of the source list.
Machine-count share is the metric everyone quotes, and it systematically understates whoever supplies the largest systems. One exascale machine and one departmental cluster are one system each by count and three orders of magnitude apart by capability.
2. Interconnect share over time
LiveInfiniBand, Slingshot, Omni-Path, Tofu and the proprietary Chinese fabrics, by installed FLOPS and by system count. The fabric layer consolidated hard when NVIDIA bought Mellanox and HPE bought Cray: two acquisitions that turned four independent suppliers into two.
3. Domestic substitution index
LiveFor each country, the share of installed FLOPS whose CPU, accelerator, interconnect and OS are all domestically supplied. The China line is the single most newsworthy series this dataset can produce, and the seed already contains its inflection points: Tianhe-1A in 2010 with American compute on a Chinese fabric, Sunway TaihuLight in 2016 with neither.
4. The shadow list
LiveTotal ranked FLOPS versus a bounded estimate of the compute sitting in large unranked commercial AI clusters. This is the gap the incumbent structurally cannot show, because a machine that never submitted a result cannot appear on a list of results.
5. Announcement-to-first-light lag
LiveBy vendor and by country, over time. Aurora took eight years from contract to first light through a cancelled architecture and a redesign; that is a vendor-execution story, and it is only visible when you have both dates for enough systems to compare.
6. Public-market exposure by ticker
LiveFor each ticker: the systems it appears in, aggregate FLOPS attributable, and whether that exposure is concentrated in one program. A supplier whose entire presence is a single national laboratory contract is a different risk from one spread across thirty installations.
How long each accelerator generation stays in the installed base before displacement. Summit ran V100s for six years; the seed records its retirement, which removes the largest remaining block of POWER installed base in one step.
The ranking stops at the rack. Cooling architecture, utility, substation capacity and site are where the constraint on the next generation actually lives, and lead times on medium-voltage switchgear now gate schedules more often than silicon does.
How concentrated the supplier base is for five component roles, as a Herfindahl-Hirschman Index over time, not just who leads. A ranking cannot show this at all: it is a question about the shape of a market, not the speed of a machine.
10. Sector mix: who operates compute
LiveGovernment labs, universities, big tech, AI labs, GPU clouds, industry, finance and life sciences, by installed FLOPS and by machine count. The two measures disagree on purpose: the sectors that grew fastest are the ones that never submit a benchmark, which is exactly what a ranking cannot show.
Megawatts that operators and developers say they are building, by operator, sector, country and expected window. A handful of campuses state more power than every operating system in the dataset that states a figure, and half of it has no stated first-power year.
12. How far to trust this data
LiveThe verified, reported and estimated tiers by sector, country and status, what re-reading every cited source found, the shortlist ready for promotion and the citations that need a second look. A dataset that shows its weak spots is one you can filter.
13. Performance per watt, 2000 to today
LiveMeasured HPL performance per watt for every system that reports both figures. Efficiency has doubled about every two and a half years, accelerated systems lead, and the frontier is a short list of named machines. It is the ranked world only: the AI campuses report neither a benchmark nor a measured draw.
14. How long supercomputers last
LiveYears in service of every retired system with both dates, by era and sector. The median has barely moved in four decades, which says the cadence is set by funding cycles, not by the technology, and the page is candid about the old rows nobody has checked.