Free tools Windows power users keep installed
One-click scans. No signup required.
There is no official worldwide ranking of supercomputers by price. The most defensible answer is a ranking of the 10 most expensive publicly documented or independently estimated supercomputer projects, ordered approximately by reported acquisition, development, or project cost.
That distinction is essential. A published figure may cover only the machine, a vendor contract, software and technology development, a building and cooling plant, application development, or even several years of operations. The list below identifies what each number includes rather than presenting unlike figures as though they were comparable invoices.
Short answer: Japan’s K Computer and Fugaku occupy the top tier of publicly documented national supercomputing projects at roughly ¥110 billion each. Frontier and El Capitan are the clearest modern U.S. system contracts at more than $600 million and $600 million, respectively. If private AI clusters and broader capital-cost estimates are included, xAI’s Colossus may be several times more expensive—but that estimate is not an official disclosure and is not directly comparable with traditional scientific supercomputers.
How this ranking measures cost
Supercomputer prices are reported in at least four different ways:
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →#1 Best Overall
- ✅Unframed Wall Art Poster Print - ALPINE INK CO. wall art posters are unframed and printed in standard sizes from 8x10 to 24x32 inches. This allows you to your wall decor with a frame that matches your home, kitchen, or office style.
- ✅Professionally Printed in the USA - Each print is professionally produced in the USA on high-quality fine art matte paper, ensuring crisp detail, vibrant color, and a smooth, non-glare finish. Designed for long-lasting modern wall art and fashion art prints.
- ✅Modern & Trendy Wall Art Decor - These unframed matte posters wall art prints are ideal for living rooms, bedrooms, kitchens, dining areas, offices, and workspaces. A stylish way to elevate your space with trendy art, fashion wall art, and contemporary wall decor.
- ✅Designed to Gift & Inspire - Our poster prints make a thoughtful gift for housewarmings, birthdays, holidays, or special occasions. A great choice for anyone who loves modern art, trendy wall decor, and stylish fashion art.
- ✅Original Artwork - Features an original design created for decorative wall use. This print is intended for personal display and interior styling.
- Hardware price: compute nodes, processors, accelerators, memory, storage, and networking.
- System contract: the machine plus integration, vendor engineering, support, software, and related technology development.
- Project cost: the system plus facilities, power, cooling, installation, application development, and sometimes research and development.
- Total cost of ownership: acquisition, electricity, staffing, maintenance, upgrades, and decommissioning over the system’s life.
The main ranking uses the first three categories where a credible figure exists. It excludes lifetime electricity and staffing costs, whole national programs, speculative future budgets, classified systems with no public price, and general-purpose data-center construction that cannot be attributed to the computer. The figures are shown in their original currencies and historical terms. Historical dollar equivalents are included only when the cited source reports them; they are not inflation-adjusted to 2026.
Because the cost definitions differ, the positions are approximate tiers rather than a mathematically precise global league table. TOP500 is useful for performance, but it ranks systems by benchmark results—not cost—and therefore cannot answer the price question by itself.
The 10 most expensive publicly documented supercomputers
| Approximate position | System | Original reported cost and date | Cost basis | Evidence |
|---|---|---|---|---|
| 1–2 | K Computer | ¥111.1 billion, fiscal years 2006–2012 | Broad project cost, including system, software, facilities, and applications | High for the reported project total |
| 1–2 | Fugaku | Approximately ¥110 billion, development-era estimate through deployment | National development cost, including application development | High for the reported development figure |
| 3–4 | Frontier | More than $600 million, 2019-era contract | System and technology development contract | High for the contract value |
| 3–4 | El Capitan | $600 million, 2019-era contract | Cray contract to build the system | High for the contract value |
| 5–7 | Earth Simulator | Approximately ¥60 billion, 2002-era project | Reported original project cost | Medium |
| 5–7 | Aurora | More than $500 million, initial contract | Initial system contract; final configuration changed during development | High for the original contract |
| 5–7, conditional | JUPITER Booster | Approximately €273 million for hardware, software, and services; approximately €500 million broader budget | Broader figure includes infrastructure and operating categories | Conditional |
| 8 | Tianhe-2 | Approximately RMB2.4 billion, 2013-era report; about $390 million at the time | Reported system or project cost | Medium-low |
| 9 | Sunway TaihuLight | RMB1.8 billion, 2016; about $270 million at the time | Building and system project cost | Medium-high |
| 10 | Summit | Approximately $200 million, estimated share of 2014 joint procurement | Estimated allocation from a $325 million Summit–Sierra contract | Medium |
The years identify the project, contract, launch, or estimate period—not a standardized accounting date. JUPITER’s position changes substantially depending on whether the broader operating budget or only hardware and services are counted.
1–2. K Computer: ¥111.1 billion
RIKEN reports that the K Computer project cost ¥111.1 billion from fiscal 2006 through fiscal 2012. This was not a hardware-only invoice. The total included system development and manufacturing, software development and evaluation, facility construction, and Grand Challenge application development, according to RIKEN’s project funding summary.
The K Computer became the first system to exceed 10 petaflops on the LINPACK benchmark, reaching 10.51 petaflops in 2011. RIKEN’s announcement described the milestone as the first time a supercomputer had passed the 10-petaflop mark.
Its mission was broad national scientific computing, including climate, materials, life-science, and engineering research. The machine operated from 2012 until 2019, when it was succeeded by newer systems at RIKEN’s computing centers; RIKEN’s retrospective documents its operating period.
Why it ranks here: ¥111.1 billion makes K Computer one of the costliest conventional supercomputing projects publicly documented. But the correct description is a broad national-project cost, not the price of the cabinets alone. Its number is most fairly compared with Fugaku’s similarly broad development figure.
1–2. Fugaku: approximately ¥110 billion
Fugaku, developed by RIKEN and Fujitsu as K Computer’s successor, required approximately ¥110 billion in public development funds, including application development. The figure comes from RIKEN’s Fugaku funding page and should be read as a national development estimate rather than a retail purchase price.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Fugaku became the world’s fastest system on the TOP500 list in June 2020. It was designed for large-scale scientific workloads such as climate and weather modeling, disaster prevention, drug and materials research, and industrial simulation. Its importance was not simply its peak speed: the project also involved co-designing processors, software, and applications so researchers could use the machine effectively.
Fugaku demonstrates why a price tag can include much more than compute hardware. Application development was part of the published ¥110 billion figure, just as facility and software work was included in K Computer’s total. Treating either number as a simple equipment purchase would exaggerate the precision of the comparison.
It also has substantial ongoing expenses. RIKEN reported a ¥13.84 billion 2022 operating budget for Fugaku-related operations. Of that, ¥12.56 billion was allocated to building costs, maintenance, utilities, communications networks, and related expenses, as shown in the 2022 Fugaku annual report. That annual figure is not part of the ¥110 billion development total; it illustrates the difference between building a supercomputer and running one.
3–4. Frontier: more than $600 million
The U.S. Department of Energy awarded a contract worth more than $600 million for Frontier’s system and technology development. The amount is documented in the DOE announcement.
Frontier was built by Cray with AMD processors and accelerators for Oak Ridge National Laboratory. It became the first officially recognized exascale system on TOP500 in 2022, marking the point at which a conventional scientific supercomputer exceeded one exaflop on the benchmark.
The wording of the DOE announcement matters: this was a system and technology development contract. It may include engineering and technology work in addition to hardware, so it should not be compared as though it were a bare equipment invoice. It also does not automatically represent every cost of the Oak Ridge facility, future upgrades, staffing, electricity, or scientific operations.
Frontier remained a leading machine years after its debut. In the June 2026 TOP500 list, it ranked third by the High Performance Linpack benchmark, behind LineShine and El Capitan.
3–4. El Capitan: $600 million
The U.S. Department of Energy and the National Nuclear Security Administration signed a $600 million contract with Cray to build El Capitan. That public contract value is documented in the DOE and NNSA announcement.
Installed at Lawrence Livermore National Laboratory, El Capitan supports NNSA’s nuclear-stockpile stewardship mission. Those workloads require extremely large simulations and modeling, and much of the surrounding mission environment is security-sensitive. The public $600 million figure therefore should not be interpreted as the complete cost of the national-security program, its classified software, or all related laboratory infrastructure.
El Capitan reached 1.742 exaflops on HPL and became the world’s fastest system in November 2024, according to NNSA’s announcement. As of June 2026, it ranked second on TOP500 behind China’s LineShine.
El Capitan is one of the cleaner entries in this comparison because the public price is tied to a single system contract. It is still not a full lifetime cost, but it is easier to attribute than a multi-system procurement or a broad national development budget.
5–7. Earth Simulator: approximately ¥60 billion
The original Earth Simulator project in Japan is widely reported to have cost approximately ¥60 billion, equivalent to roughly $500 million to $600 million at contemporary exchange rates. It was a national project focused on climate and Earth-science simulation, developed with participation from Japan’s marine, aerospace, and atomic-energy research organizations. JAMSTEC’s technical overview describes the system and its role.
Recommended Free Tools
The original machine opened in 2002 and was designed to model the atmosphere, oceans, climate, and other large Earth systems. These simulations are valuable because they allow researchers to explore long time periods and complex physical interactions that would be impractical to reproduce through physical experiments alone.
The cost evidence is weaker than RIKEN’s published K Computer and Fugaku figures. The ¥60 billion figure is a reported original project cost, not a clearly itemized hardware-only contract on the current English-language JAMSTEC pages. It should not be combined with the cost of later upgrades or successor Earth Simulator systems.
5–7. Aurora: more than $500 million
The DOE awarded Intel a contract worth more than $500 million to build Aurora, according to the original DOE announcement. The figure is the initial contract value, and Aurora’s configuration changed during development, so it should not be presented as a final all-in ownership cost.
Aurora is housed at Argonne National Laboratory and was developed for large-scale scientific research, including simulation, data analysis, and artificial-intelligence workloads. Its place in the cost ranking is therefore based on the original contract, not on a claim that every later upgrade, facility expense, or operating cost is included.
The system is also a reminder that contract price and final delivered capability can diverge over a long project. Hardware road maps, processor availability, software readiness, and system configuration can all change before a machine reaches production use. As of the June 2026 TOP500 list, Aurora ranked fourth by HPL.
5–7, conditional. JUPITER Booster: €273 million for hardware and services, or approximately €500 million on a broader budget basis
JUPITER Booster is the European exascale system associated with the EuroHPC initiative and the Jülich Supercomputing Centre. Widely reported figures distinguish between approximately €273 million for hardware, software, and services and approximately €500 million for a broader program budget that includes power, cooling, installation, maintenance, and operations. The figures are summarized in the JUPITER reference material.
This is the most important conditional entry in the list. If the ranking measures system hardware and services, JUPITER belongs below the large U.S. contracts. If it measures the broader acquisition and operating budget, it enters the same general tier as the $500-million-plus systems. Those are different comparisons, and neither should be silently substituted for the other.
JUPITER ranked fifth on the June 2026 TOP500 list, behind LineShine, El Capitan, Frontier, and Aurora. Its position in a price ranking is less certain because the commonly cited €500 million figure includes categories that are excluded from some other entries.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →8. Tianhe-2: approximately RMB2.4 billion
Tianhe-2, also called MilkyWay-2, is commonly reported to have cost approximately RMB2.4 billion, or about $390 million at the time. The figure is a reported system or project cost rather than a readily accessible single procurement document; the Tianhe-2 reference summary records the commonly cited amount.
Developed in China by a large research and engineering team, Tianhe-2 became the world’s fastest supercomputer in 2013 and remained at or near the top of TOP500 for several editions. It was used for scientific and engineering workloads requiring very large-scale parallel computation.
Two qualifications matter. First, the RMB2.4 billion number should be described as reported or estimated, not as an officially itemized hardware price. Second, the original Tianhe-2 and later Tianhe-2A configuration should not automatically be treated as the same machine with the same cost. Historical system upgrades can add capability without creating a clean new price tag.
9. Sunway TaihuLight: RMB1.8 billion
The technical report on Sunway TaihuLight gives a cost of RMB1.8 billion, approximately $270 million at the time. The report describes this as the cost of the building and system project, making it broader than a processor-and-node invoice. The figure appears in the technical report by Jack Dongarra and collaborators.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Sunway TaihuLight used Chinese-designed processors and achieved roughly 93 petaflops on LINPACK. It became the world’s fastest supercomputer in 2016, a milestone also reported by Tsinghua University.
Its cost is difficult to compare directly with U.S. dollar contracts. Domestic processor design, local manufacturing, construction, labor, procurement rules, and exchange rates all affect what a reported RMB total means. The number is still useful because it is tied to a published technical account, but it should not be treated as a universal cost per petaflop.
10. Summit: approximately $200 million, estimated
The DOE procurement behind Summit was a $325 million contract for two systems: Summit and Sierra. The official documentation does not assign a separate public price to each machine. Industry analysis estimated that roughly $200 million represented Summit and $125 million represented Sierra. The combined procurement is documented in the LLNL annual report, while the separate allocation is an estimate rather than a disclosed contract split.
Summit was installed at Oak Ridge National Laboratory and was announced with approximately 200 petaflops of peak performance in the DOE launch announcement. It supported climate modeling, materials science, astrophysics, biology, energy research, and other scientific workloads.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsThe correct wording is therefore: Summit is estimated to have represented roughly $200 million of a $325 million joint Summit–Sierra procurement. It is incorrect to say Summit itself cost $325 million. Because its individual number is estimated, it ranks below entries with directly disclosed single-system contracts.
A strong alternate: Blue Waters at approximately $188 million
Blue Waters is a useful alternate if the ranking gives priority to a machine-specific contract rather than historical leadership. Cray built the system for the National Center for Supercomputing Applications at the University of Illinois, and the reported contract value was approximately $188 million. The commonly cited figure is summarized in the Blue Waters reference material.
There is no single perfect choice between Blue Waters and Summit for a top-10 list. Blue Waters has a more directly attributable single-system figure. Summit was a much more prominent leadership system, but its price is an estimated share of a two-system contract. A transparent article should either include Blue Waters as an alternate or explain why Summit was selected.
Why private AI supercomputers change the answer
Traditional government and university supercomputers are usually built for scientific workloads and are often discussed using FP64 High Performance Linpack results. Modern private AI clusters are optimized mainly for training and serving machine-learning models, frequently using lower-precision arithmetic and different measures of throughput. Their cost and performance cannot be placed cleanly on the same table without changing the methodology.
xAI Colossus: an independent estimate of $7 billion to $12.9 billion
xAI describes Colossus as a 200,000-GPU AI training system on its official Colossus page. That scale makes it relevant to any modern discussion of the most expensive computing systems, even though it is not a conventional publicly procured scientific supercomputer.
Epoch AI estimated approximately $7 billion in hardware for Colossus as of March 2025. Its later data-center estimate put the broader capital cost at approximately $12.9 billion, based on hardware, power, and facility assumptions.
Those figures are not necessarily contradictory:
- Approximately $7 billion: estimated hardware cost.
- Approximately $12.9 billion: estimated broader capital cost, including facility and power assumptions.
Neither number is an official xAI cost disclosure. Colossus should therefore be discussed in a separate AI category rather than inserted into the public-HPC ranking as though its estimate were equivalent to a government contract. It may be the most expensive system in this article under a broad private-capital definition, but the evidence is less direct.
Meta Research SuperCluster
Meta publicly described its Research SuperCluster as a system with 16,000 NVIDIA A100 GPUs in the initial configuration and discussed a planned expansion. Meta’s original announcement and later engineering description explain the system’s scale and purpose, but do not publish a complete acquisition price.
That makes Meta’s system an important example of the disclosure problem. A large private cluster can be much more expensive than a university machine, yet its price may remain unknown because the owner buys hardware, leases capacity, constructs facilities, or expands in stages.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why supercomputers cost so much
A supercomputer is not simply a large collection of desktop processors. The major cost drivers usually include:
- Thousands of compute nodes: each node contains processors or accelerators, memory, local storage, and specialized boards.
- High-bandwidth memory: scientific simulations and AI models can be limited by how quickly data moves to the processor, not just by arithmetic speed.
- Custom interconnects: nodes must exchange data rapidly and predictably, requiring switches, cables, network adapters, and software.
- Parallel storage: simulations can generate enormous data sets that need high-throughput storage and backup systems.
- Liquid cooling: dense modern systems produce too much heat for ordinary office-style cooling. Pumps, heat exchangers, plumbing, and cooling towers add capital and operating costs.
- Electrical infrastructure: high-voltage equipment, transformers, power distribution, backup systems, and monitoring are part of the facility.
- Buildings and installation: some projects require a new data hall or a substantial modification to an existing laboratory.
- Vendor co-design: a first-of-its-kind exascale machine can require years of engineering before production hardware is ready.
- Software and applications: researchers need compilers, libraries, scheduling systems, ported codes, and application development to use the hardware efficiently.
- Reliability and security: defense and national-laboratory systems may require redundancy, specialized controls, and secure operating environments.
- Maintenance and upgrades: a machine must be supported throughout its useful life, often while components and software are changing.
This is why a project such as K Computer can have a reported cost far above the likely price of its raw processors. The project paid for a usable national research capability, not merely for compute cabinets.
Does the most expensive supercomputer tend to be the fastest?
No. Cost and benchmark performance are related, but they are not the same ranking.
Free tools Windows power users keep installed
One-click scans. No signup required.
A system may be expensive because its budget includes processor development, a new building, application research, or years of installation and support. Another system may achieve impressive benchmark performance by using commodity components, an existing facility, or a more narrowly defined contract. A private AI cluster may have enormous hardware spending but be optimized for mixed-precision model training rather than the double-precision HPL benchmark used by TOP500.
The current performance snapshot makes the distinction clear. In the June 2026 TOP500 announcement, China’s LineShine ranked first, followed by El Capitan, Frontier, Aurora, and JUPITER Booster. LineShine is not assigned a price in this article because its cost was not publicly established in the sources reviewed. It would be misleading to turn its first-place performance into an invented price ranking.
HPL is also only one measure. Scientific users may care about energy efficiency, memory capacity, communication speed, application performance, reliability, or how well a system handles AI and data-intensive workloads. A machine can be expensive for reasons that HPL does not capture.
How much does it cost to operate a supercomputer?
There is no universal annual operating cost. Electricity prices, utilization, cooling efficiency, staffing, maintenance contracts, network charges, building ownership, and upgrade schedules can change the answer dramatically.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsFugaku provides a documented example: RIKEN reported ¥13.84 billion in its 2022 budget, including ¥12.56 billion for building costs, maintenance, utilities, communications networks, and related expenses. This is not a pure electricity bill, and it is not a standard percentage that can be applied to every machine. It does show why the purchase or development figure is only one part of the financial picture.
For a privately owned AI cluster, the operating budget may also include leased power capacity, data-center rent or construction, networking, cooling water, hardware replacement, and the engineering staff needed to train models. When an independent estimate includes those facilities, it can be several times higher than the estimated GPU purchase price.
How to interpret the price tags
When comparing any two entries, ask these questions:
- Is this one system or several? Summit’s $325 million contract covered Summit and Sierra, so it cannot be assigned entirely to Summit.
- Is it a contract, an estimate, or a project total? Frontier and El Capitan have public contract values; Tianhe-2 and Earth Simulator rely more heavily on reported secondary figures.
- Are facilities included? K Computer, Sunway TaihuLight, and JUPITER demonstrate how buildings, power, cooling, and installation can materially change the headline number.
- Are software and applications included? Fugaku’s ¥110 billion development figure includes application development.
- Does the figure cover operations? JUPITER’s broader €500 million number does; many other entries do not.
- Is the dollar conversion historical? A dollar figure reported at the time of a project is not the same as its 2026 inflation-adjusted value or its value at today’s exchange rate.
- Is the machine an original system or an upgrade? Tianhe-2 and Earth Simulator should not have later configurations or successor systems silently added to their original costs.
What this list cannot prove
Public figures do not reveal the full cost of every advanced system. Classified systems may disclose a contract while withholding the surrounding security environment, software, facility, or mission expenses. El Capitan is a good example: its $600 million build contract is public, but that does not expose the complete cost of NNSA’s nuclear-stockpile stewardship computing infrastructure.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallPrivate companies create a different information gap. They may not publish purchase prices, may build in stages, and may combine AI training hardware with general-purpose data-center infrastructure. Independent estimates can be useful, but they should remain labeled as estimates.
For these reasons, the most accurate headline is not the 10 most expensive supercomputers in the world without qualification. It is the 10 most expensive publicly reported or independently estimated supercomputer projects, ranked approximately by acquisition or development cost.
Frequently Asked Questions
Why is xAI Colossus not ranked number one in the main list?
Colossus may be more expensive than any system in the conventional list if private AI clusters and broader capital costs are included. However, its approximately $7 billion hardware estimate and approximately $12.9 billion broader estimate come from independent analysis, not an official xAI disclosure. Its AI throughput is also measured differently from TOP500’s traditional FP64 HPL performance, so it is treated separately.
Why are K Computer and Fugaku tied?
Both have publicly reported Japanese national-project figures of roughly ¥110 billion. Neither figure is a simple hardware invoice, and each includes a different mix of development, software, facilities, and applications. A precise order would imply more certainty than the published evidence supports.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why does Summit not have a $325 million price tag?
The $325 million contract covered two systems: Summit and Sierra. Summit’s often-cited approximately $200 million price is an estimated allocation of that joint procurement, not a separately disclosed contract value.
What is the difference between a supercomputer’s purchase price and its total cost?
A purchase or system-contract figure may cover hardware, integration, support, and vendor engineering. A project total can also include buildings, cooling, power, software, and application development. Total cost of ownership adds electricity, staffing, maintenance, upgrades, and decommissioning over the machine’s lifetime.
The Bottom Line
Bottom line: K Computer and Fugaku are among the largest publicly documented national supercomputing projects, at approximately ¥110 billion each when development, software, facilities, and applications are included. Frontier and El Capitan are the clearest modern $600-million-class system contracts. Earth Simulator, Aurora, JUPITER, Tianhe-2, Sunway TaihuLight, and Summit belong in the same conversation, but their figures use different definitions and levels of certainty.
Private AI systems such as xAI Colossus may now require multibillion-dollar investments and could surpass the traditional leaders under a broad capital-cost definition. There is still no definitive global cost ranking until governments and companies report comparable figures for hardware, facilities, software, and operations.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
