supercomputer (HPC)
A supercomputer is a room-sized machine consisting of thousands of interconnected processors, performing calculations in hours that would take an average computer years.
In plain terms
Imagine you have to assemble a huge puzzle. Alone, it would take months, but if ten thousand people work on it simultaneously, each on their own section, and constantly coordinate, it is done in a few hours. A supercomputer does the same: it is not a single miracle processor, but countless computing units working in coordination on the same task.
It is used for tasks too large for a traditional computer: weather and climate forecasting, drug molecule testing, and fracture and flow simulations for cars and aircraft. In recent years, increasingly for training large artificial intelligence models.
What sets it apart from your laptop is not its operating principle but its size and coordination. It consists of a room full of server racks, has its own cooling system, and the largest ones consume as much electricity as a small town. It has no screen or keyboard: researchers submit their jobs remotely and queue for machine time.
How it works
- The machine consists of hundreds, thousands, or even tens of thousands of separate computers, called nodes, in which processors and increasingly graphics accelerators (GPUs) work.
- The nodes are connected by a very fast internal network so that partial results can be passed to each other without delay.
- The program is written so that the task can be broken down into many small parts, and these parts run simultaneously and in parallel on different nodes.
- A scheduler software assigns which researcher's task runs when and on how many nodes, because the machine is used by many simultaneously.
- The immense heat generated during operation is usually dissipated in large machines using liquid cooling (typically water). Performance is measured in flops: this shows how many floating-point operations (operations with decimal numbers) the machine performs per second.
An everyday example
When you check your phone in the morning to see if it will rain in the afternoon, you see a forecast calculated by a supercomputer using data from millions of points in the atmosphere. Similarly, the large language models you chat with in chatbots were trained on giant machines consisting of tens of thousands of GPUs.
Who is affected
A common misconception
Limits and risks
- Strong only in parallelizable tasks: what cannot be broken down is not significantly accelerated.
- It is very expensive to build and operate, and its electricity and cooling requirements pose a significant environmental burden.
- It becomes obsolete quickly: a cutting-edge machine today slips down the rankings within a few years.
- Access is scarce: computing time must be bid for or paid for, and the most powerful machines are in the hands of a few countries and large corporations.
Similar, but not the same
- quantum computer – A quantum computer operates on a completely different physical principle using qubits and remains largely an experimental tool, whereas a supercomputer is built from traditional chips and has been in daily use for decades.
- data center – The data center is the building and infrastructure in which the machines are housed. The machines in an average data center serve many independent tasks (websites, storage, applications). The supercomputer is also housed in a data center, but there many machines work simultaneously on a single large task, closely coordinated.
- cloud service – Cloud is rentable, flexibly expandable general computing capacity, while a supercomputer is a specific machine built for peak performance, to which research or industrial projects typically gain access time.
Related concepts
computing capacity, AI infrastructure, high-performance computing
exascale computer, AI factory, GPU cluster
data center, GPU, cloud service, quantum computer, model training
A brief history
The first supercomputer is generally considered to be the CDC 6600, designed by Seymour Cray and introduced in 1964. Since 1993, the world's fastest machines have been ranked by the TOP500 list, updated every six months.
Where things stand
By the end of 2026, the most powerful machines already operate in the exascale category, performing at least one quintillion (10^18) operations per second. Examples include the American El Capitan, Frontier and Aurora, as well as Europe's first exascale machine, JUPITER in Germany. Most new investments are aimed specifically at training AI models. In Europe, the EuroHPC Joint Undertaking is building a network of supercomputers and so-called AI factories. Hungary's best-known supercomputer is Komondor, handed over in 2023.
In the news
- UK's largest AI supercomputer may wait until the 2030s for sufficient powerSep 29, 2026 · Ipar
- Power supply problems delay launch of UK's largest AI supercomputerSep 24, 2026 · Ipar
- EU earmarks billions for AI gigafactories and tightens oversight of ChatGPTAug 31, 2026 · Agent Age
- UC San Diego tests $8.48 million system for grid-responsive AI computeAug 24, 2026 · Ipar
- DCAI and Corti launch sovereign AI control layer on Denmark's Gefion supercomputerAug 20, 2026 · Health
- Szeged conference presents health-industry applications of supercomputing and AIAug 17, 2026 · Health
- AI projects aim to strengthen US power grids, from rural Alaska to AWS–PNNL workAug 10, 2026 · Ipar
- OTP Bank selects SambaNova systems to build Europe's fastest AI supercomputerAug 4, 2026 · Money
Also known as
supercomputer · szuperkomputer · supercomputer · supercomputer · HPC · high-performance computing · high-performance computing · exascale computer
Sources
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.