Core42 unveils Maximus-384 supercomputer, ranked 20th worldwide
Core42's US deployment features more than 9,000+ AMD Instinct MI300X GPUs
#UAE #USA #HPC - G42 Croup’s sovereign cloud, AI infrastructure, and services company Core42, has revealed that it has commissioned the world’s 20th most powerful supercomputer at in Buffalo, New York. The Maximus-384, which is powered by over 9,000 of AMD’s Instinct MI300X GPUs, is ranked in the Top500 list of supercomputers with a LINPAK performance of 114.50 petaflops and a theoretical peak (Rpeak) of 251.15 petaflops. The system is configured for large-scale AI training and inference. The Maximus-384 cluster operates within US energy company TeraWulf’s state-of-the-art Lake Mariner data centre campus on the shores of Lake Ontario. Core42 signed long-term lease agreements for data centre space with TeraWulf in 2024.
SO WHAT? - The TOP500 placement of Maximus-384 both validates Core42’s technical capabilities in deploying and operating world-leading supercomputing infrastructure and illustrates G42’s increasing investment in the United States. The US market is a key pillar of the group’s global expansion, which has already seen it hire employees in California, New York, Texas, Washington and elsewhere. Meanwhile, G42’s investment in the US aligns with the US-UAE AI Acceleration Partnership announced during US President Donald Trump’s official visit to the UAE in May 2025.
Here are some key points about the new supercomputer:
Core42 this week revealed the US-based Maximus-384, the world’s 20th most powerful supercomputer, as ranked by Top500 with a LINPAK performance of 114.50 petaflops and a theoretical peak (Rpeak) of 251.15 petaflops.
Maximus-384 is located in Buffalo, New York at TeraWulf’s state-of-the-art Lake Mariner data centre campus on the shores of Lake Ontario. Core42 signed long-term lease agreements for data centre space with TeraWulf in 2024.
Core42’s Buffalo cluster features more than 9,000 AMD Instinct MI300X GPUs (graphics processing units) configured for large-scale AI training and inference via G42’s strategic collaboration with AMD.
The TOP500 List tracks performance trends across global supercomputing systems using the High Performance Linpack benchmark.
The architecture is powered by US networking company Broadcom’s high-bandwidth networking chips handling data transfer between graphics processing units, central processing units and storage, alongside US network equipment manufacturer Arista Networks’ Ethernet switches and EOS software managing traffic flow and optimising data routing.
Core42’s expanding infrastructure spans sovereign clusters and international supercomputing deployments. The company is building a global backbone for ultra-low latency and regulatory compliance to support organisations from startups to multinationals.
Optimised for AI workloads, Core42’s AI Cloud infrastructure operates purpose-built high-performance computing system compute islands functioning as self-contained environments featuring dedicated high-performance storage and backbone networks delivering exceptional speed and reliability.
The Buffalo cluster operating within TeraWulf’s Lake Mariner facility provides power resiliency, cooling efficiency and operational scale required for large GPU deployments, whilst maintaining the zero-carbon, energy-efficient compute environment standards applied across Core42 regions.
Core42’s segmented architecture strategy supports deployment of multiple silicon types enabling Core42’s heterogeneous AI Cloud, including AI infrastructure from NVIDIA, Cerebras and AMD.
ZOOM OUT - Core42’s Maximus-384 ranking marks a significant advancement from the company’s previous highest-ranked system, an NVIDIA DGX SuperPOD that secured 25th position on the TOP500 list in November 2024. That system, featuring 185,712 CPU and GPU cores powered by Intel Xeon Platinum processors and NVIDIA DGX H100 GPUs, achieved Linpack performance of 55.81 petaflops against theoretical peak of 87.27 petaflops. Ranked as the Middle East’s most powerful high-performance computing system at the time, and the second-largest NVIDIA DGX SuperPOD deployment globally, the UAE-based infrastructure addressed growing regional demand for sovereign AI capabilities. The jump from 25th to 20th position within twelve months, whilst more than doubling performance from 55.81 to 114.50 petaflops, shows a rapid expansion of compute capacity across continents.
[Written and edited with the assistance of AI]
LINKS
Top500 list - Nov. ‘25 (Top500 website)
Read other related articles:
G42, iGenius to build €1B AI supercomputer in Italy (Middle East AI News)
New US-UAE agreement sure to accelerate G42’s US plans (Middle East AI News)
Core42’s NVIDIA DGX SuperPOD (Middle East AI News)
G42 to deploy 2GW data centre and supercomputer in India (Middle East AI News)
Cerebras and G42 begin building CG-2 (Middle East AI News)
Cerebras & G42 build massive cloud supercomputer network (Middle East AI News)