Pulse RevOpsPulseValue Added
AI Infrastructure · Top 10

The 10 Fastest Public Supercomputers in the US and Europe in 2027

By Kory White · Updated Nov 2026 · 5 min read
El Capitan supercomputer racks at Lawrence Livermore National Laboratory
Direct answer

El Capitan at Lawrence Livermore National Laboratory is the fastest public supercomputer in the US and Europe, with Frontier at Oak Ridge second and Aurora at Argonne third. The order follows each machine's measured HPL speed (Rmax) on the June 2026 TOP500 list.

At a glance

#PickHPL speed (Rmax)Best for
1El Capitan1,809.00 PFlop/sThe fastest machine any Western government runs
2Frontier1,353.00 PFlop/sThe longest-serving exascale machine
3Aurora1,012.00 PFlop/sHuge peak capacity built on Intel CPUs and GPUs
4JUPITER Booster1,000.00 PFlop/sEurope's only exascale system
5Alps434.90 PFlop/sNVIDIA Grace Hopper at national scale
6LUMI379.70 PFlop/sA EuroHPC flagship built on the same AMD design as Frontier
7Leonardo241.20 PFlop/sItaly's EuroHPC system on NVIDIA A100 GPUs
8Isambard-AI phase 2216.50 PFlop/sThe UK's national AI research machine
9Tuolumne208.10 PFlop/sEl Capitan hardware at the lowest listed power draw here
10MareNostrum 5 ACC175.30 PFlop/sSpain's EuroHPC GPU partition on NVIDIA H100

Pick by what you want

Fastest overall
El Capitan
Longest-serving exascale
Frontier
Fastest in Europe
JUPITER Booster
Built for AI research
Isambard-AI phase 2
Lowest power draw
Tuolumne

The ranking

1

El Capitan

El Capitan supercomputer racks at Lawrence Livermore National Laboratory
Photo: Lawrence Livermore National Laboratory
Pulse VerdictThe fastest public machine outside China.
HPL speed (Rmax): 1,809.00 PFlop/sPeak (Rpeak): 2,821.10 PFlop/sTOP500 rank: #2 overall (June 2026)Cores: 11,340,000Power: 29,685 kWInstalled: 2024

El Capitan runs at Lawrence Livermore National Laboratory for the NNSA and posts 1,809.00 PFlop/s on HPL from 11,340,000 cores. It pairs AMD 4th Gen EPYC processors with AMD Instinct MI300A accelerators on HPE's Slingshot-11 network. TOP500 reports an energy efficiency of 60.94 Gigaflops/watt, and its 17.41 Petaflop/s on the HPCG benchmark is No. 2 in the world.

Go ifThe fastest machine any Western government runs
Skip ifYou want a DOE Office of Science machine: it belongs to the NNSA
Why #1 over #2: El Capitan measures 1,809.00 PFlop/s against Frontier's 1,353.00 PFlop/s on the same benchmark.
2

Frontier

Frontier supercomputer cabinets at Oak Ridge National Laboratory
Photo: Oak Ridge National Laboratory / CC BY 2.0 via Wikimedia Commons
Pulse VerdictThe first exascale system is still the No. 2 public machine.
HPL speed (Rmax): 1,353.00 PFlop/sPeak (Rpeak): 2,055.72 PFlop/sTOP500 rank: #3 overall (June 2026)Cores: 9,066,176Power: 24,607 kWInstalled: 2021

Frontier at Oak Ridge National Laboratory was installed in 2021 and still holds 1,353.00 PFlop/s on HPL. It uses AMD 3rd Generation EPYC 64-core processors and AMD Instinct MI250X GPUs in HPE Cray EX235a cabinets. It is run for the DOE Office of Science.

Go ifThe longest-serving exascale machine
Skip ifYou need the newest chips: it is the oldest machine in the top four
Why #2 over #3: Frontier is faster on HPL than Aurora (1,353.00 vs 1,012.00 PFlop/s) while drawing 24,607 kW against Aurora's 38,698 kW.
3

Aurora

Team members in front of Aurora supercomputer racks at Argonne
Photo: Argonne National Laboratory (via University of Chicago News)
Pulse VerdictExascale on Intel silicon, with the biggest power bill here.
HPL speed (Rmax): 1,012.00 PFlop/sPeak (Rpeak): 1,980.01 PFlop/sTOP500 rank: #4 overall (June 2026)Cores: 9,264,128Power: 38,698 kWInstalled: 2023

Aurora at Argonne National Laboratory was built by Intel with HPE, using Intel Xeon CPU Max 9470 processors and Intel Data Center GPU Max accelerators. It measures 1,012.00 PFlop/s on HPL, and its theoretical peak of 1,980.01 PFlop/s is the third highest here. Argonne finished installing its 10,624th and final blade in July 2023.

Go ifHuge peak capacity built on Intel CPUs and GPUs
Skip ifEnergy use matters most: it draws the most power of the ten
Why #3 over #4: Aurora edges JUPITER Booster on HPL, 1,012.00 vs 1,000.00 PFlop/s, with a much higher peak.
4

JUPITER Booster

JUPITER Booster racks at Forschungszentrum Jülich
Photo: Forschungszentrum Jülich / Sascha Kreklau / CC BY 4.0 via Wikimedia Commons
Pulse VerdictEurope's first exascale machine.
HPL speed (Rmax): 1,000.00 PFlop/sPeak (Rpeak): 1,226.28 PFlop/sTOP500 rank: #5 overall (June 2026)Cores: 4,801,344Power: 15,794 kWInstalled: 2025

JUPITER Booster at Forschungszentrum Jülich in Germany is measured at exactly 1,000.00 PFlop/s, which TOP500 calls the first European exascale result. It runs on NVIDIA GH200 Grace Hopper Superchips in Eviden's direct liquid-cooled BullSequana XH3000 design. It draws 15,794 kW, far less than the three US exascale machines.

Go ifEurope's only exascale system
Skip ifYou want the top raw speed: three US machines on this list are faster
Why #4 over #5: JUPITER Booster reaches 1,000.00 PFlop/s; Alps reaches 434.90 PFlop/s.
5

Alps

Alps supercomputer at the Swiss National Supercomputing Centre
Photo: Marco Abram, CSCS / CC BY-SA 3.0 via Wikimedia Commons
Pulse VerdictThe fastest Grace Hopper system after JUPITER.
HPL speed (Rmax): 434.90 PFlop/sPeak (Rpeak): 574.84 PFlop/sTOP500 rank: #10 overall (June 2026)Cores: 2,121,600Power: 7,124 kWInstalled: 2024

Alps at the Swiss National Supercomputing Centre (CSCS) measures 434.90 PFlop/s on HPL. It is an HPE Cray EX254n system on NVIDIA Grace 72-core CPUs and GH200 Superchips, installed in 2024. It draws 7,124 kW.

Go ifNVIDIA Grace Hopper at national scale
Skip ifYou need exascale: it runs below half that line
Why #5 over #6: Alps beats LUMI on HPL, 434.90 vs 379.70 PFlop/s, at almost the same power (7,124 vs 7,107 kW).
6

LUMI

Visitors in front of the LUMI supercomputer in Finland
Photo: European Commission audiovisual service, Alessandro Rampazzo / CC BY 4.0 via Wikimedia Commons
Pulse VerdictFrontier-style AMD hardware at a European scale.
HPL speed (Rmax): 379.70 PFlop/sPeak (Rpeak): 531.51 PFlop/sTOP500 rank: #11 overall (June 2026)Cores: 2,752,704Power: 7,107 kWInstalled: 2023

LUMI is the EuroHPC system hosted by CSC in Finland and measures 379.70 PFlop/s on HPL. Its design matches Frontier: HPE Cray EX235a cabinets, AMD 3rd Generation EPYC 64-core processors and AMD Instinct MI250X GPUs. It came online in 2023 and draws 7,107 kW.

Go ifA EuroHPC flagship built on the same AMD design as Frontier
Skip ifYou want the newest GPUs: it uses MI250X, not MI300A
Why #6 over #7: LUMI is faster than Leonardo, 379.70 vs 241.20 PFlop/s, while using less power.
7

Leonardo

Leonardo supercomputer aisle at CINECA in Bologna
Photo: National Institute of Geophysics and Volcanology (INGV) / CC BY 4.0 via Wikimedia Commons
Pulse VerdictA strong A100-era machine, now mid-pack.
HPL speed (Rmax): 241.20 PFlop/sPeak (Rpeak): 306.31 PFlop/sTOP500 rank: #12 overall (June 2026)Cores: 1,824,768Power: 7,494 kWInstalled: 2023

Leonardo is the EuroHPC system at CINECA in Bologna, Italy, at 241.20 PFlop/s on HPL. It is a BullSequana XH2000 with Intel Xeon Platinum 8358 processors and NVIDIA A100 SXM4 64 GB GPUs. It draws 7,494 kW, the most of the European machines below JUPITER.

Go ifItaly's EuroHPC system on NVIDIA A100 GPUs
Skip ifEfficiency matters: it uses more power than Alps for less speed
Why #7 over #8: Leonardo measures 241.20 PFlop/s against 216.50 PFlop/s for Isambard-AI phase 2.
8

Isambard-AI phase 2

Isambard-AI cabinets in the Bristol data centre
Photo: University of Bristol
Pulse VerdictBritain's fastest public machine, built for AI.
HPL speed (Rmax): 216.50 PFlop/sPeak (Rpeak): 278.58 PFlop/sTOP500 rank: #13 overall (June 2026)Cores: 1,028,160Power: Not listedInstalled: 2025

Isambard-AI phase 2 at the University of Bristol measures 216.50 PFlop/s on HPL. The University says it is built on HPE Cray EX with 5,400 NVIDIA GH200 Grace Hopper Superchips and is a key part of the UK Government's AI Research Resource. TOP500 lists no power figure for this entry.

Go ifThe UK's national AI research machine
Skip ifYou need a published power figure: TOP500 lists none for it
Why #8 over #9: Isambard-AI phase 2 is faster on HPL than Tuolumne, 216.50 vs 208.10 PFlop/s.
9

Tuolumne

Tuolumne supercomputer at Lawrence Livermore National Laboratory
Photo: Lawrence Livermore National Laboratory
Pulse VerdictThe most frugal machine on this list.
HPL speed (Rmax): 208.10 PFlop/sPeak (Rpeak): 288.88 PFlop/sTOP500 rank: #14 overall (June 2026)Cores: 1,161,216Power: 3,387 kWInstalled: 2024

Tuolumne at Lawrence Livermore uses the same HPE Cray EX255a design, AMD 4th Gen EPYC processors and AMD Instinct MI300A accelerators as El Capitan, at a smaller size. It measures 208.10 PFlop/s on HPL from 1,161,216 cores. At 3,387 kW it has the lowest listed power draw of the ten.

Go ifEl Capitan hardware at the lowest listed power draw here
Skip ifYou want scale: it is a small sibling of El Capitan
Why #9 over #10: Tuolumne measures 208.10 PFlop/s; MareNostrum 5 ACC measures 175.30 PFlop/s.
10

MareNostrum 5 ACC

MareNostrum 5 racks at the Barcelona Supercomputing Center
Photo: Steve Jurvetson / CC BY 4.0 via Wikimedia Commons
Pulse VerdictSpain's fastest public machine, on NVIDIA H100.
HPL speed (Rmax): 175.30 PFlop/sPeak (Rpeak): 249.44 PFlop/sTOP500 rank: #16 overall (June 2026)Cores: 663,040Power: 4,159 kWInstalled: 2023

MareNostrum 5 ACC is the GPU partition of the EuroHPC system at the Barcelona Supercomputing Center. It is a BullSequana XH3000 with Intel Xeon Platinum 8460Y+ processors and NVIDIA H100 64GB GPUs, measuring 175.30 PFlop/s on HPL. It sits at No. 16 on the June 2026 list.

Go ifSpain's EuroHPC GPU partition on NVIDIA H100
Skip ifYou want a top-15 machine: it sits at No. 16

Faster machines left off this list

Five systems in the June 2026 TOP500 top 10 are not public machines in the US or Europe. They are shown here so the ranking above is not mistaken for the full world list.

TOP500 rankSystemSiteHPL (PFlop/s)Why it is left off
1LineShineNational Supercomputing Centre in Shenzhen, China2,198.40Outside the US and Europe
6HPC7Eni S.p.A., Italy571.50Company-owned (energy)
7EagleMicrosoft Azure, United States561.20Company-owned (cloud)
8HPC6Eni S.p.A., Italy477.90Company-owned (energy)
9Supercomputer FugakuRIKEN Center for Computational Science, Japan442.01Outside the US and Europe

How we ranked these

The order is each system's measured HPL speed (Rmax) on the June 2026 TOP500 list, the newest official ranking. Only machines run by national labs, national supercomputing centres, EuroHPC and universities in the US and Europe count; company-owned systems and systems in China and Japan are listed below for reference. Specs are copied from each system's TOP500 entry and the operators' own pages; check the official sites for the latest.

Questions people ask

What is the fastest supercomputer in the world right now?

LineShine at the National Supercomputing Centre in Shenzhen, China. It debuted at No. 1 on the June 2026 TOP500 list with 2,198.40 PFlop/s on the HPL benchmark, ending El Capitan's run at the top. It is left off this list because it is outside the US and Europe.

What is the fastest supercomputer in the United States?

El Capitan at Lawrence Livermore National Laboratory, at 1,809.00 PFlop/s on HPL. It is No. 2 in the world on the June 2026 TOP500 list.

What is the fastest supercomputer in Europe?

JUPITER Booster at Forschungszentrum Jülich in Germany, measured at exactly 1,000.00 PFlop/s (one exaflop). TOP500 calls it the first European exascale system.

Why are Eni HPC7, HPC6 and Microsoft Eagle not on this list?

They are company-owned. Eni's HPC7 (No. 6) and HPC6 (No. 8) serve Eni's own energy work, and Eagle (No. 7) runs inside Microsoft Azure. This list covers machines run by public labs, national centres, EuroHPC and universities.

How often does the TOP500 list change?

Twice a year, in June and November. This page uses the June 2026 list and will be updated when the November 2026 list is published.

HPL speed (PFlop/s) at a glance

MareNostrum 5 ACC175.3 PFlop/s
Tuolumne208.1 PFlop/s
Isambard-AI phase 2216.5 PFlop/s
Leonardo241.2 PFlop/s
LUMI379.7 PFlop/s
Alps434.9 PFlop/s
JUPITER Booster1,000 PFlop/s
Aurora1,012 PFlop/s
Frontier1,353 PFlop/s
El Capitan1,809 PFlop/s
Measured HPL speed (Rmax) in petaflop/s, from the June 2026 TOP500 list.

Official sites

Sources

Photos: llnl.gov · commons.wikimedia.org · news.uchicago.edu · bristol.ac.uk · hpc.llnl.gov (credits under each photo).

TOP500 publishes a new list every June and November; this page follows the June 2026 list.

More from Pulse

AI InfrastructureHow do you set up a GPU cluster for distributed training in 2027?AI InfrastructureHow much does a single NVIDIA H100 cluster node cost to lease in 2027?AI InfrastructureThe 10 Best AI Tools for GPU Cluster Orchestration in 2027AI InfrastructureHow many infrastructure engineers should you budget for per data scientist when scaling AI infrastructure in 2027?AI InfrastructureHow much can committed-use or reserved GPU cloud contracts save versus on-demand pricing for AI training in 2027?AI InfrastructureThe 10 Best GPU Spot Instance Marketplaces for AI Training in 2027
Share this page