- Supercomputing and AI make it possible to create digital twins of the planet, cities, and human organs to simulate and anticipate complex scenarios.
- Europe is promoting projects such as Destination Earth, LUMI and the BSC, combining supercomputers, networks of centers and the development of its own chips.
- The focus of AI is shifting from massive training to inference, with new servers, PCs, and even desktop supercomputers being prepared for AI.
- Spain is participating in this race with MareNostrum, the Spanish Supercomputing Network and systems like Picasso, providing services to science, industry and society.
Supercomputing and artificial intelligence have become the hottest pairing in today's technology. We're no longer just talking about huge data centers hidden in scientific bunkers, but about machines capable of creating digital twins of the planet, the human heart , or an entire city, and even desktop supercomputers that fit (more or less) under an office desk.
At the same time, we are experiencing a phase shift in AI: from the boom in training gigantic models, we have moved to an obsession with inference —that is, with using these models at full capacity in everyday tasks. Hardware manufacturers, research centers, and universities are now competing to offer everything from supercomputing centers like the Barcelona Supercomputing Center (BSC) or the European LUMI, to compact servers and AI-ready PCs that bring a kind of "mini supercomputer" to the desktop.
What is supercomputing really and how is its power measured?
When we talk about supercomputing, we're not referring to a souped-up PC, but rather to clusters of thousands of computers working in a coordinated manner as if they were a single machine. Each of these computers is a node, with its own CPUs, GPUs, RAM, and storage, all connected by ultrafast interconnection networks that minimize latency , the biggest enemy of performance.
The power of these machines is expressed in FLOPS (floating-point operations per second) . At home, a powerful computer can operate in the teraFLOPS (TFLOPS) range. In supercomputing, we're playing in a different league: it's common to talk about petaFLOPS ( 10¹⁵ operations per second) and, in the most advanced systems, exaFLOPS ( 10¹⁸ ). Frontier, in the United States, was the first to officially break the exascale barrier.
To give you an idea, a modern supercomputer can perform in one hour what a home computer would take years to calculate . This incredible computing power is what allows us to simulate everything from hurricanes to protein dynamics, or to train AI models with trillions of parameters.
What do supercomputers physically look like and why do they need so much cooling?
Visually, a supercomputer looks nothing like a typical desktop PC. It usually resembles a room full of metal cabinets , each containing hundreds or thousands of processors, GPUs, and hard drives. The power is so immense that energy consumption can reach several megawatts , and a significant portion of that energy is lost as heat.
That's why these systems need dedicated rooms with extreme cooling : industrial climate control, hot and cold aisles, direct liquid cooling to the chip, and even creative solutions to harness that heat. In Switzerland, for example, the heat from a supercomputer is reused to heat university buildings , turning a problem into an advantage.
In some cases, highly sophisticated security and protection systems are used, such as glass enclosures with special fire suppression systems that use micronized water capable of extinguishing fires without damaging the electronics. This is the case of the original MareNostrum in Barcelona, installed inside the chapel of the Polytechnic University of Catalonia: probably one of the supercomputers with the most unusual locations in the world.
The digital twin revolution: from Earth to the human heart
The combination of supercomputing and AI is driving a key concept: digital twins . These are not simply virtual models, but dynamic replicas that integrate real-world data in near real time to simulate, anticipate, and optimize what happens in the physical world.
In Europe, the European Commission is promoting the Destination Earth (DestinE) program , which aims to develop a highly accurate digital twin of the Earth within a few years. Thanks to supercomputers like LUMI , the most powerful in the European Union, very high-resolution, long-term climate simulations can be carried out, incorporating the atmosphere, oceans, and land surface with a level of detail that until recently was only achievable with very short-term weather models.
According to Utz-Uwe Haus, head of the HPE HPC/AI EMEA Research Lab, this capability allows for a better understanding of extreme events for disaster management, the study of climate change scenarios, and the assessment of the impact of glaciers, sea ice, vegetation, and aerosols on the global climate. But it also enables something very practical: predicting local effects with enormous precision, such as average rainfall, droughts, or floods at the scale of a region or city.
This has direct consequences for agricultural planning (which crops are viable in a given area and at what risk), for investment in renewables (forecasting hours of sunshine and wind over decades), and for infrastructure design. It is a clear example of how supercomputing ceases to be something abstract and begins to influence very concrete economic decisions.
Digital twins in cities, rivers and ports
Digital twins aren't limited to the global climate. The Barcelona metropolitan area has a digital twin of its 164 municipalities that allows for the simulation of urban, economic, mobility, housing, and knowledge scenarios for the coming decades. Policies and plans can be tested on this virtual replica before making decisions in the real world.
In the port and river transport sector, the Port of Seville is developing Guadaltwin, a digital twin of the Guadalquivir Waterway, as part of its digitalization plan. This system integrates AI and machine learning to improve predictions and decisions regarding river traffic, draft management, tides, infrastructure, and safety.
Even in seemingly disparate fields like high-energy physics and fashion, digital twins have begun to make their way into the industry. CERN is researching how to use these models in its particle physics experiments, robotics, and cooling systems, while companies like H&M have created digital replicas of human models for advertising campaigns, sparking debates about image rights and the future of creative work.
The human body as the next great digital twin
One of the most ambitious challenges lies in healthcare. Teams like Steven Niederer 's at Imperial College London are working on digital twins of individual hearts, with their specific shape, size, and function. These models allow for the simulation of surgeries and treatments without risk to the patient and are already being used in clinical trials and intervention planning.
Researchers like Andreu Climent and María de la Salud Guillem, from the Polytechnic University of Valencia, believe that these digital cardiac twins will be key to treating complex arrhythmias, deciding who benefits from an implantable defibrillator, and anticipating the risk of sudden death. And the long-term goal is even more ambitious: to build a complete digital twin of the human body that will allow for testing therapies, adjusting drug dosages, and personalizing medicine to the fullest extent.
AI, supercomputing, and the shift from training to inference
For years, the bulk of AI investment has gone toward training increasingly large models, especially in generative AI. Today, the focus is clearly shifting toward inference : using those models at scale in production, continuously, and at a lower cost per operation.
This shift was clearly evident at CES 2026. Manufacturers like Lenovo unveiled servers specifically designed for inference, such as the ThinkSystem SR675i, SR650i , and ThinkEdge SE455i , ready to run AI models near the data source, at the so-called edge.
Origin PC, now part of the Corsair ecosystem, has showcased the S-Class Edge AI Developer Kit , a compact, ready-to-use platform for developing AI at the network edge. The idea is that small development or research teams can test and deploy AI without always relying on the cloud or massive external data centers.
Most PC manufacturers present at CES followed the same path: Acer with its RA100 AI mini-station and its updated Veriton desktops; LG GRAM with dual AI capabilities (local + cloud); Asus with a range of new Vivobooks and the ProArt PX13 convertible aimed at creators working with AI; Dell updating the XPS range for AI workloads; and HP updating EliteBook, EliteBoard, Omnibook and OmniStudio, all of them with AI acceleration and data power.
Supercomputing that comes down to the desktop: the “desktop supercomputer”
One particularly interesting trend is local supercomputing , with machines that, while not reaching the scale of a national center, offer incredible power in a desktop format. At CES 2026, Gigabyte (through its subsidiary Giga Computing) unveiled the Gigabyte W775-V10 , a true “desktop supercomputer.”
This system integrates the NVIDIA AI stack and the NVIDIA GB300 Grace Blackwell Ultra Desktop accelerator , among other top-tier components. Its goal is to enable AI teams to train and infer complex models without relying on the cloud or external data centers, while maintaining complete control over the data and runtime environment.
Alongside it, CES has served to refresh the ecosystem of components: new Intel Core Ultra CPUs , new AMD Ryzen processors, Qualcomm's Snapdragon X2 Plus chip , Kioxia's BG7 SSDs, advanced DDR5 memory from Gigabyte, and updated motherboards from MSI, all designed to support data-intensive and AI-intensive workloads.
In the peripherals sector, brands like Corsair showcased their latest high-end mice and keyboards, while Anker, eufy, and soundcore focused on connected devices. Even some intriguing gadgets appeared, such as the Plaud Notepin S , a small note-taking device powered by AI.
What are supercomputers being used for today: from COVID to air quality
Supercomputers are almost always used for advanced research in fields where a regular PC would literally take ages to calculate. Their classic uses include meteorology and climate , earthquake simulation, research in astrophysics, geophysics, biology, medicine, drug design, and aerospace engineering.
During the COVID-19 pandemic , several supercomputers were used to simulate the behavior of viral proteins, test combinations of molecules, and accelerate the search for drugs. Massive simulations allowed researchers to discard unpromising approaches and focus their efforts on the compounds with the highest probability of success.
Centers like the Barcelona Supercomputing Center have shown very concrete examples: with sensor data and fluid dynamics models, they trained neural networks to control incinerators and improve fuel use by reducing emissions; or to predict air quality in large cities with remarkable accuracy, based on years of historical data.
Another striking example is AlphaFold , DeepMind's system for predicting protein folding from its amino acid sequence. This Nobel Prize-level problem has benefited from an explosive combination of AI, big data, and supercomputing . The impact on biomedicine and drug design has been enormous, to the point that tens of thousands of researchers worldwide are already using its results on a daily basis.
Common uses of supercomputing
- Weather and climate forecasting in the medium and long term.
- Simulation of earthquakes, tsunamis and natural hazards to reduce damage.
- Design and testing of aircraft, vehicles and rockets using aerodynamic models.
- Drug discovery and design and molecular interaction studies.
- Astrophysics and cosmology: formation of galaxies, stars and black holes.
- Air quality and atmospheric composition in regions and cities.
- Big Data and social simulation: cultural evolution, population movements, smart cities.
- Security and defense: from nuclear weapons simulation to digital twins of radars and complex systems.
Where are the large supercomputers located and what role does Spain play?
The Top500 list compiles and ranks the 500 most powerful supercomputers in the world twice a year since 1993. Although China leads in the number of systems on that list, the United States maintains the lead in total aggregate power , especially with machines like Frontier.
Among the current giants are Fugaku in Japan, which led the way for years; Summit and Sierra in the United States; and Sunway TaihuLight and Tianhe-2A in China, which also once held the top spot. Italy is home to systems like HPC5 and Marconi-100 , and Switzerland boasts Piz Daint , a long-time leader in Europe.
In Spain, the benchmark is MareNostrum at the Barcelona Supercomputing Center. Since its first version in 2004, with approximately 42,4 teraFLOPS, it has scaled up to the current MareNostrum 4, with around 13,7 petaFLOPS . The next generation, MareNostrum 5, will represent a leap of several orders of magnitude in power and energy consumption, and is part of the European strategy to develop exascale infrastructure.
A particularly important network is the Spanish Supercomputing Network (RES) , which brings together centers and machines distributed across different autonomous communities and provides services to researchers throughout the country. Building on this foundation, an Ibero-American Supercomputing Network has also been created , connecting resources from countries such as Mexico and other Latin American partners for joint projects.
At the regional level, facilities such as the Picasso supercomputer at the University of Málaga stand out, with some 40.000 computing cores and 180 TB of RAM. Picasso serves both researchers at the university itself and Andalusian users through the Andalusian Bioinformatics Platform, as well as scientists from all over Spain through the RES (Spanish Network for Bioinformatics).
These systems almost always run on Linux or its derivatives, due to their open-source nature, stability, and low resource consumption compared to other commercial operating systems. On this foundation, a specialized and refined ecosystem of scientific tools, programming environments, and AI libraries has been built.
Reference centers: Barcelona Supercomputing Center and the European race for its own hardware
The Barcelona Supercomputing Center (BSC) is one of Europe's leading players in supercomputing and computer architecture research. Led for decades by Mateo Valero, the BSC has grown from managing a single supercomputer to becoming a center with over a thousand people from more than 50 countries, organized into departments of Computer Science, Life Sciences, Earth Sciences, and Social Applications.
One of BSC's distinguishing features is that it doesn't just operate machines; it also develops its own software, algorithms, and even processors . For years, it has been involved in European projects such as EuroHPC and initiatives like the European Processor Initiative (EPI) and European chips based on open architectures such as RISC-V , with the aim of reducing Europe's dependence on American and Asian manufacturers.
In collaboration with other partners, the BSC has promoted prototypes of vector processors , ARM- and RISC-V-based platforms, and a whole family of designs with names like Turtle, Lizard, and Chameleon , which have grown in complexity with each generation. The goal is to create, in the medium term, chips capable of powering “MareNostrum 6” supercomputers with critical computing technology developed in Europe.
This effort is framed within an uncomfortable reality: Europe used to design part of the ARM architecture, but the sale of ARM to non-European companies and the lack of large foundries of its own have left the continent in a precarious position . While some initiatives, such as the United States guaranteeing TSMC's advanced production in Arizona, or the chip factories that Germany and France attract with substantial public subsidies, take place in Spain, Spain faces the challenge of combining design, manufacturing, and an industrial ecosystem with comparatively fewer resources.
In this context, the Spanish strategy involves consolidating centers like the BSC, promoting networks like the RES, supporting open-chip projects, and training highly sought-after professionals in computer architecture and AI. It is no coincidence that, as the center's own directors acknowledge, PhDs specializing in these fields receive offers from private industry with salaries difficult to match from academia, making it challenging to retain talent.
Meanwhile, other Spanish universities are strengthening their infrastructure, as demonstrated by the Picasso Institute in Málaga and the RES nodes distributed throughout the country. In many cases, these systems serve particle physics , climate change studies, engineering, bioinformatics, and smart city projects, demonstrating that supercomputing is no longer a luxury reserved for a select few laboratories.
Looking at the big picture, it's quite clear that supercomputing has gone from being a laboratory curiosity to becoming a critical infrastructure for climate, health, security, the economy, and even for how we design cities, vehicles, and pharmaceuticals. At the same time, the leap of AI from the lab to production and the rise of digital twins are bringing some of this power to specialized servers and even the desktops of engineers and scientists, creating a scenario in which, far from disappearing, supercomputing is becoming increasingly integrated into our daily lives, even if we often don't see it.
