<\/p>\n
China\u2019s LineShine tops the TOP500 with 2.198 exaflops\u2014a pure CPU architecture that sends a clear geopolitical signal. But raw HPL performance doesn\u2019t tell the whole story when it comes to AI dominance. Here\u2019s a look at Europe\u2019s response. <\/em><\/p>\n TL;DR \u2014 Summary in three points<\/p>\n China\u2019s LineShine dominates the TOP500 with 2.198 exaflops, but as a pure CPU architecture, it isn\u2019t automatically the best for AI. Europe has JUPITER (exascale), LUMI, and the upcoming Arrhenius. For Nordic companies, the key is not to compete with national supercomputers\u2014it\u2019s to deploy the right computing power in the right place. <\/p>\n<\/div>\n China\u2019s new supercomputer, LineShine<\/strong>, has secured the top spot on the TOP500 list with a measured HPL performance of 2.198 exaflops<\/strong>. Reaching over two exaflops\u2014that is, more than two trillion floating-point operations per second\u2014represents a level of computing power that dwarfs traditional national research programs. However, LineShine is about more than just raw speed; the machine sends a clear geopolitical signal. <\/p>\n The system is installed at the National Supercomputing Center<\/em> in Shenzhen and is built entirely on China\u2019s own LingKun platform<\/strong>: Arm-based LX2 processors with 304 cores per chip, the proprietary LingQi network, and the Kylin operating system. In total, this amounts to nearly 13.8 million cores<\/strong>. LineShine thus becomes the first supercomputer on the list to break the two-exaflops barrier in sustained double-precision performance using only CPUs. <\/p>\n This poses a critical question for Europe: How should we compete now that supercomputers have evolved from research tools into strategic infrastructure?<\/p>\n With its latest HPL results, LineShine has edged out the previous leader, the<\/em> American El Capitan<\/em> (built on the AMD MI300A), by just over 21 percent. It\u2019s a technical feat, but the number doesn\u2019t tell the whole story. <\/p>\n The TOP500 list primarily measures traditional HPC performance\u2014the kind of double-precision floating-point calculations that power climate models, materials research, and advanced physical simulations. The modern AI landscape, however, requires entirely different characteristics: <\/p>\n Analysts therefore emphasize that LineShine isn\u2019t automatically the world\u2019s best AI machine just because it tops the list. A pure CPU architecture can be fantastic for heavy scientific simulations, but less optimal for modern AI training compared to GPU-intensive systems. What makes LineShine interesting is that China has chosen not only to copy the West\u2019s focus on GPUs\u2014it has built its own, alternative architecture. <\/p>\n The obvious driving force is the U.S. export restrictions that have for several years curtailed China\u2019s access to advanced HPC and AI chips from Nvidia and AMD. But beneath the surface, it\u2019s all about sovereignty. <\/p>\n As computing power becomes the driving force behind everything from defense and energy supply to biotechnology and AI, supercomputers can no longer be viewed as ordinary IT purchases. They are national assets worthy of protection, on the same level as the power grid or transportation infrastructure. China simply wants to eliminate all dependence on foreign chips, networks, and operating systems. LineShine is therefore as much a geopolitical statement as it is a technical achievement. <\/p>\n The perception of a Europe that has fallen behind the U.S. and China is widespread, but it\u2019s too black-and-white. We\u2019ve already come a long way. <\/p>\n JUPITER<\/strong><\/a> in Germany, operated by the J\u00fclich Supercomputing Center<\/em> as part of EuroHPC, is the continent\u2019s first true exascale system. With its modular architecture\u2014divided into a GPU-accelerated booster section and a general-purpose cluster section\u2014it is built to handle both traditional research and heavy AI workloads. <\/p>\n At the same time, the Finnish company LUMI<\/strong><\/a> in Kajaani delivers approximately 380 petaflops of sustained performance. The system is also a key resource for Sweden, as we are part of the consortium and Swedish researchers have direct access to the resources. On-premise, we also have a solid foundation with NSC in Link\u00f6ping (Tetralith<\/strong>, Berzelius<\/strong>) and PDC at KTH (Dardel<\/strong>). <\/p>\n The next big step is Arrhenius<\/strong><\/a>, which is being installed in Link\u00f6ping under the NAISS banner:<\/p>\n The GPU performance comes in at 66.8 petaflops HPL<\/strong> \u2014which is about seven times more than what Dardel currently delivers. This is a significant boost in raw computing power for Swedish industry and academia. <\/p>\n Although JUPITER and LUMI are important milestones, they do not address all Nordic needs. For Swedish companies developing AI, digital twins, or autonomous systems, it\u2019s rarely enough that \u201cthere\u2019s a supercomputer somewhere in Europe.\u201d The practical challenges are more about accessibility, lead times, and regulatory compliance: <\/p>\n This is where the gap between public research infrastructure and industry-oriented AI environments becomes apparent. University supercomputer centers do a fantastic job for academia, but commercial projects often require shorter lead times, strict SLAs, and total control over the environment. That doesn\u2019t mean every company has to build its own exascale systems, but it does require a well-thought-out HPC strategy. <\/p>\n HPC has evolved from being a concern solely for universities and industry giants to becoming a core requirement for broader business operations, driven by advancements in AI.<\/p>\n As AI models grow, the difference between having immediate access to computing resources and having to wait in line is becoming a critical business factor. Having computing power close to the business enables faster iteration, better data protection, and a shorter path to production. <\/p>\n The EU\u2019s initiative on AI Factories<\/em> is clear evidence that the seriousness of the situation has been recognized. The idea is to provide startups, companies, and the public sector with access to AI-optimized supercomputers and expert support. <\/p>\n Mimer AI Factory is now being established in Sweden Mimer AI Factory<\/strong><\/a> is being established in Link\u00f6ping, with a focus on areas such as life sciences, materials science, and autonomous systems. EuroHPC has also signed an agreement for a brand-new AI supercomputer for Mimer, which is scheduled to be installed in 2026. <\/p>\n This is a major and welcome step, but it does not meet all needs. Large-scale public initiatives are excellent resources, but many companies require fully dedicated, isolated environments where security, performance, and day-to-day operations are fully aligned with their own organization. It is in this segment that enterprise-grade GPU clusters serve their purpose. <\/p>\n You don\u2019t need to match China\u2019s millions of cores to stay at the forefront. For most Swedish companies, it\u2019s more a matter of building a smart, customized GPU cluster with the right balance between storage, networking, and orchestration. This could be an inference cluster for AI agents, a secure environment for sensitive data, or a hybrid model where you offload heavy workloads to national resources while keeping the core in-house. <\/p>\n The key is to flip the process: start with the workload instead of the hardware.<\/strong><\/p>\n \nBy first defining which models will be trained, how much data will be transferred, and what the requirements are for confidentiality and response times, the design of the infrastructure becomes both more precise and cost-effective.\n<\/p>\n<\/blockquote>\n At Aixia, we build AI and HPC infrastructure for Nordic companies that want to maximize their performance without losing control over their data, costs, or operations.<\/p>\n As the only DGX SuperPOD-certified partner<\/a> in the Nordic region, we design, install, and operate GPU clusters tailored for generative AI, inference, and high-performance computing environments. This includes everything from customized on-premises solutions and private AI clouds to hybrid environments that complement major national resources. <\/p>\n It\u2019s not a matter of choosing one or the other. Europe needs major flagship projects like JUPITER, LUMI, Arrhenius, and Mimer, but Nordic companies also need flexible, secure, and business-oriented alternatives. LineShine is clear proof that computing power has become a matter of hard geopolitics. However, the answer isn\u2019t just to build the largest machines, but to create an accessible ecosystem where the right computing power is available in the right place. <\/p>\n Because, in the end, what matters most isn’t who owns the fastest supercomputer on the list\u2014it’s where the innovation is actually taking place.<\/p>\n LineShine delivers 2.198 exaflops of HPL performance, topping the June 2026 TOP500 list.<\/p>\n Not necessarily. It excels at traditional HPC performance (double precision), but since it is entirely CPU-based, it is not automatically optimal for the types of low-precision computations required by modern AI training. <\/p>\n Yes. JUPITER<\/a> in Germany is Europe’s first exascale system and is operated within EuroHPC at the J\u00fclich Supercomputing Center.<\/p>\n Finland has the pre-exascale system LUMI<\/a>. In Sweden, there is the NSC<\/a> in Link\u00f6ping (with Tetralith and Berzelius), the PDC<\/a> at KTH (Dardel), and the new Arrhenius<\/a> initiative. <\/p>\n Yes, through access via NAISS, EuroHPC, and Mimer AI Factory<\/a>. For business-critical, continuous, or highly sensitive data, dedicated, on-premises GPU clusters are often the best commercial option. <\/p>\n Advanced AI requires customized computing power, networking, and storage. Without the right infrastructure, an AI strategy can easily remain just a vision rather than an actual, functioning production environment. <\/p>\n
\n1. LineShine is fast, but the figure needs to be put into context<\/h2>\n
\n
2. Why is China building its own processors?<\/h2>\n
3. Europe Is Not Without Hope<\/h2>\n
\n
4. However, the Nordic region still lacks its own exascale capacity<\/h2>\n
\n
5. HPC is no longer just for researchers<\/h2>\n
\n
6. Mimer and AI Factories are a step in the right direction<\/h2>\n
7. Here\u2019s How Swedish Companies Can Gain HPC Capacity Without Building Their Own LineShine<\/h2>\n
\n
Aixia’s Perspective<\/h2>\n
\nFrequently Asked Questions<\/h2>\n
How fast is China’s LineShine?<\/h3>\n
Is LineShine the best in the world at AI?<\/h3>\n
Does Europe have an exascale supercomputer?<\/h3>\n
What key supercomputer resources are available in the Nordic region?<\/h3>\n
Can Swedish companies use HPC without building their own supercomputers?<\/h3>\n
Why is HPC important for an AI strategy?<\/h3>\n
\nSources and Direct Links<\/h2>\n
\n
\n