Acryl pushes GPUBASE into North America after revenue jumps 150%

Acryl is expanding its GPU operations platform GPUBASE into North America after first-half revenue rose 150% year over year to KRW 12 billion and its order backlog reached KRW 21.2 billion at the end of June. The Seoul-based AI infrastructure firm used a 1,272-GPU performance verification and meetings in Florida with U.S. research and economic development groups to build its market entry case.

Why it matters: - Acryl is trying to turn AI infrastructure performance into a North American sales pitch. - The company is targeting universities and public research institutions first, then data centers and cloud operators. - Faster GPU utilization and lower networking costs matter because AI clusters are expensive to buy and even more expensive to run efficiently.

What happened: - Acryl began its North American market entry with GPUBASE, its GPU operations platform. - The move follows first-half revenue of KRW 12 billion, up 150% year over year. - Acryl ended June with an order backlog of KRW 21.2 billion. - The company presented at the 39th US-Korea Conference on Science and Technology, or UKC 2026, in Orlando, Florida, from August 5-8. - Acryl was the only Korean company on the panel of the official Korea-US AI cooperation symposium. - Acryl delivered the anchor presentation at the official sponsor forum. - Company representatives visited HiPerGator, the University of Florida’s supercomputing facility. - Acryl also held a roundtable with SelectFlorida and met NVIDIA co-founder Chris A. Malachowsky on August 7.

The details: - GPUBASE pools GPUs and NPUs from multiple manufacturers into a single resource. - The platform can partition a single GPU so multiple jobs share it. - Acryl says the system reduces network bottlenecks and raises the realized performance of existing equipment. - The commercial product includes Booster, Control Tower and Monitor Tower. - Booster is designed to reduce data-transfer bottlenecks between GPUs. - Control Tower manages GPU, server, storage and network resources as one pool. - Monitor Tower shows infrastructure and job status in real time. - GPUBASE network technology is based on research by Acryl CTO Yeom Ik-joon, a professor of software at Sungkyunkwan University. - The research includes PeRF, which distributes data-transfer paths, and UL-MPRDMA, which uses multiple paths at the same time. - Acryl’s U.S. market entry is focused on heterogeneous accelerator environments and Ethernet-based clusters rather than single-vendor systems or InfiniBand. - The company says universities and public research institutions will be its first references for expansion into data center and cloud operators. - Acryl’s K-Scale Evaluation began in June in Amazon Web Services and other global cloud environments. - The Horizontal Phase of that evaluation finished in July. - Independent tests ran on 1,272 GPUs across seven heterogeneous types in the three major global cloud environments, each with a different network architecture. - GPU utilization rose above 90% in the tests. - Training time under high load fell by up to 96%, equal to a speed-up of up to 24 times. - Job queue wait time fell by up to 93%. - NCCL all-reduce communication bandwidth improved by 141% to 2,375% depending on load. - Fault detection and recovery finished within five minutes. - Acryl plans a next-stage performance test at single-cluster scale of 1,000 GPUs. - A KOLAS-accredited test report is expected afterward. - Acryl’s cost-saving figures are its own estimates. - On a compiled basis from published papers and commercial cases, network total cost of ownership on a 512-GPU NVIDIA H100 cluster is 48% of the InfiniBand figure. - That implies a saving of about US$2.2 million. - Raising average GPU utilization from 55% to around 85% would cut the GPU purchase cost needed for the same throughput by about 35%. - In July, Acryl signed a contract to supply GPUBASE to Hancom’s modular data center construction project. - An advanced-tier contract worth KRW 3.3 billion was added on July 29. - Cumulative supply for the Hancom project reached KRW 5.5 billion. - The advanced tier runs on NVIDIA H200 NVL GPU servers and direct-liquid-cooled racks. - Delivery is scheduled to finish in the third quarter. - Acryl signed a memorandum of understanding with Megazone Cloud on July 24. - The agreement extends coverage to large AI data centers and distributed AI infrastructure. - Earlier agreements with MangoBoost and FuriosaAI expanded verification to Korean-made DPUs and NPUs. - In June, Korea’s Institute of Information & Communications Technology Planning & Evaluation selected Acryl as lead organization for an Ethernet-based GPU cluster network fabric project. - The project has a total budget of about KRW 6.7 billion and runs through December 2028. - Acryl is working on the project with Yonsei University, Sungkyunkwan University and Ajou University. - The project aims to replace InfiniBand with open Ethernet. - Acryl reported August 18 first-half revenue of KRW 12 billion, equal to 90% of full-year 2025 revenue. - The operating loss ratio narrowed to 19% from 44% a year earlier. - The order backlog at the end of June was 49% higher than at the end of March. - Medical AI contributed KRW 8.5 billion in new orders this year through NADIA-ANE and ALLM.H. - NADIA-ANE is an AI-native electronic medical record platform. - ALLM.H is a healthcare foundation model deployed on hospitals’ own GPU servers. - On September 1, Acryl said its 31-billion-parameter successor model, ALLM.H2 FRANK, answered all 435 physician-examination questions on the KorMedMCQA benchmark.

Between the lines: - Acryl is positioning GPUBASE as an efficiency product, not just a capacity product. - The North American pitch leans on verified performance, open Ethernet and heterogeneous hardware support, which could appeal to research buyers trying to avoid lock-in. - The company is also showing commercial traction in Korea at the same time, which strengthens the expansion story. - The mix of research, public-sector projects and enterprise deals suggests Acryl wants validation before a broader U.S. push.

What's next: - Acryl plans a 1,000-GPU single-cluster test next. - The company expects a KOLAS-accredited report after that test. - Delivery on the Hancom advanced-tier supply is due in the third quarter. - Acryl will likely use the U.S. conference meetings and verification results to pursue university, research and infrastructure customers in North America.

The bottom line: - Acryl is trying to convert fast-rising revenue and benchmark results into a North American foothold for GPUBASE.