AI firm PanaAI and Junee have partnered up to build an AI supercomputer in Australia.

Dubbed PanaAI AUS AISF, the supercomputer will comprise up to 4,088 Nvidia H200 Tensor Core GPUs and be interconnected via the Nvidia Quantum-2 InfiniBand platform.

Set to launch in early 2025, it is expected to achieve a theoretical peak performance of around 16.4 exaflops for FP8 operations, making it the most powerful Nvidia-based AI supercomputer in the Southern Hemisphere.

The system will offer cloud services to both public and private sectors, supporting the development of large foundational models for generative AI and the advancement of research, innovation, and societal applications.

In a statement to ITBrief Australia, a spokesperson for PanaAI stated: "To meet the growing demand for generative AI and computing resources, PanaAI is planning to establish multiple world-class, powerful supercomputing systems to provide unparalleled accelerated services to users worldwide.

"We highly value our collaboration with Nvidia to deliver AI cloud services and rapidly train large foundational models built on ever-expanding datasets of images, acoustics, and sensor information. We look forward to closer collaboration to support users around the globe."

PanaAI specializes in AI computing services, particularly in GPU capabilities for areas such as high-performance computing and machine learning.

Junee Limited, and its wholly owned subsidiary ASPAC AI Computing, describe themselves as specializing in supercomputing infrastructure. Hong Kong-based Junee was previously an interior design, fit-out and maintenance services provider to both residential and commercial clients. Earlier this year the firm announced a funding round for an AI supercomputing center in Australia.

According to an SEC filing, PanaAI will supply ASPAC with high-performance computing resources in a deal totaling $100 million.

Eviden to triple Finland’s computing power with new national AI supercomputer

Atos’ Eviden has been awarded up to €60 million ($63m) as part of a framework agreement with the IT Center for Science in Finland to provide the country with a new national supercomputer.

Named Roihu, the system will triple the computing capacity of the center’s existing Mahti and Puhti machines, whilst simultaneously significantly increasing its AI performance.

Roihu will be built on Eviden’s BullSequana XH3000 hybrid platform and increase the center’s GPU capacity by more than 10x, providing 49 petaflops of theoretical peak performance.

Stulz Modular completes installation of hybrid cooled data center for the University of Göttingen’s Emmy supercomputer

Modular data center provider Stulz Modular has announced the completion of a new data center to house the University of Göttingen’s Emmy supercomputer in Germany.

The modular data center was commissioned as the university’s existing facilities did not have the required space or cooling infrastructure to support Emmy.

The new data center has a total capacity of 1.5MW but can accommodate further expansion in the future. It also contains a direct-to-chip liquid cooling system that can remove heat density of up to 100kW per rack.

It comprises four prefabricated modules – two larger modules covering an area of 85 sqm (914 sq ft) joined along the spine to accommodate the direct-to-chip liquid-cooled supercomputer – and two smaller modules, also joined along the spine to accommodate air-cooled IT equipment in 70 sqm (753 sq ft) of space.

University of Alabama approves construction of high-performance computing and data center

The University of Alabama has approved plans for the construction of a high-performance computing and data center on its campus.

Slated to be completed by late 2026, the facility will hold an estimated $25 million worth of computing power and support research across all the university’s campuses, in addition to projects in collaboration with external partners.

Most of the funding for the center comes from a disbursement from the US Department of Commerce’s National Institute of Standards and Technology, alongside a supplemental funding package from the state’s Education Trust Fund.

Sesterce launches Nvidia-powered AI supercomputer in Digital Realty’s Marseille data center

AI infrastructure provider Sesterce has deployed an AI supercomputer featuring an undisclosed number of Nvidia H100 Tensor Core GPUs, connected via InfiniBand technology.

Hosted in Digital Realty’s Marseille data center in France, it has been designed to address the growing demand for cloud GPU solutions and will offer AI-as-a-Services to customers, with a focus on scalability and sustainability.

Sesterce says the deployment will allow customers to develop scalable AI projects of any size while ensuring optimal resource usage and environmental responsibility.

New astronomy supercomputer ATERUI III deployed in Japan

The Center for Computational Astrophysics at the National Astronomical Observatory of Japan (NAOJ) has deployed an HPE Cray XD2000 supercomputer that will replace its existing ATERUI II system.

Housed at the Mizusawa Campus in Oshu City, Iwate prefecture, the new system became operational on December 2.

It has a total theoretical peak performance of 1.99 petaflops and consists of two types of subsystems: System M which focuses on memory bandwidth and System P which focuses on memory capacity.

In a statement, NAOJ said ATERUI III will further serve as a “laboratory for theoretical astronomy," allowing the observatory to investigate a wide range of astrophysical phenomena.

Kaytus partners with NEC Germany for liquid-cooled HPC cluster at University of Cologne

Kaytus has partnered with NEC Germany to provide the University of Cologne with a liquid-cooled high-performance computing cluster.

Dubbed RAMSES and housed at the university’s data center, the 4.8 petaflops cluster will replace the university’s previous CHEOPS systems, with the upgrade intended to support software applications in the fields of astrophysics, quantum physics, life sciences, and genome analysis.

The system contains more than 140 liquid-cooled nodes, with Kaytus and NEC claiming its liquid cooling solutions were able to reduce cooling costs at the university’s data center by 40 percent, despite improved performance and high data security.