At SC26 in Chicago
KONST, an AI compute services company based in Taiwan, is exhibiting at SC26, the supercomputing conference held November 15–20 at McCormick Place in Chicago, at Booth 1062. It is presenting its AI compute services under the theme "24/7 GPUs Delivered in Asia," aimed at cloud and compute providers, HPC and AI decision-makers and research institutions. The company was founded in 2024, is headquartered in Taipei and has offices in Singapore and Japan.
Ben Chang, co-founder and chairman, said SC26 is where the global HPC community gathers, and that the company wants international partners to see compute services in Asia that connect GPU clusters, cloud platforms and model governance. He said Taiwan builds the world's AI hardware and that KONST's job is to turn that hardware strength into compute that all of Asia can use, and to keep it running around the clock.
Validation before delivery
Konstra AI handles architecture design, rack deployment and tuning, then validates each cluster with tests such as NCCL and HPL, so that customers receive compute ready for training and inference from the first day. Configurations on show range from NVIDIA H100 and H200 to B200 and B300. The company describes this preparation as the foundation of the service chain, aimed at customers with long-term, large-scale training and inference needs.
Round-the-clock operations
In Taiwan, Konstra AI already runs 24/7 remote and on-site operations for several large GPU clusters, both self-built and managed. The work covers monitoring of GPU health, network performance and power load; incident response with root cause analysis reports; continuous tuning to raise utilisation; and access management with two-factor authentication, per-site role-based permissions and an audit trail for every action. Each customer also has a dedicated support contact.
Flexible compute and model access
Partner Glows.ai offers compute billed by the second, with users choosing GPU or CPU instances and able to check the underlying processor and run large language models on one platform. At SC26, Glows.ai was invited to join Intel's Innovation Pass partner co-marketing initiative and is showing CPU cloud services powered by Intel Xeon Platinum 8558 and 8468 processors. At the model end, Horizon AI's (k) ATP Token connects more than 70 models through a single key, with prepaid billing and no monthly fee. Horizon AI is a member of the NVIDIA Inception program.