VESSL AI outlines 300MW AI factory plan at OCP Korea Tech Day

VESSL AI said it will present its AI factory strategy at the 2026 OCP Korea Tech Day in Seoul on August 21, including a roadmap to reach 100MW of AI datacenter capacity by next year and 300MW by 2028. The company also plans to discuss multi-datacenter GPU cloud operations, next-generation GPU procurement and reliability engineering for agentic AI workloads.

Why it matters: - VESSL AI is positioning infrastructure, not just GPU count, as the next battleground for enterprise AI workloads. - The company’s 300MW target signals a push toward large-scale capacity for distributed training and inference across Korea and overseas. - Enterprise and institutional customers could use the event to pre-reserve next-generation GPU allocations.

What happened: - CEO Jaeman An said VESSL AI will participate in the 2026 OCP Korea Tech Day at COEX in Seoul on August 21. - VESSL AI plans to showcase its VESSL Cloud technology and its datacenter expansion strategy. - The company also plans to present its AI Factory Service, an end-to-end framework for computing resources, development and operational environments, and infrastructure management.

The details: - VESSL AI aims to bring 100MW of AI datacenter capacity online by next year, split evenly between 50MW domestically and 50MW internationally. - The company plans to expand that total to 300MW by 2028. - VESSL AI said its multi-datacenter model will span multiple Korean regions and overseas locations. - The model is designed to support distributed AI training and inference workloads. - The company will outline procurement and deployment plans for new GPUs, including NVIDIA’s next-generation architecture. - VESSL AI will offer pre-reservation consultations for enterprise and institutional clients seeking next-gen GPU allocations. - Jaeman An will speak in the Datacenter Track on “From Datacenters to AI Factories.” - An plans to discuss how AI agent growth is shifting token demand and why GPU clouds depend on power, silicon, capital and operational efficiency. - Jihwan Chun, VESSL AI’s CTO, will speak in the AI Solution Track on “GPU Infrastructure Reliability Engineering in the Era of Agentic AI.” - Chun will discuss infrastructure safeguards developed while operating H100, H200 and B300 clusters across multiple Cloud Service Providers in Korea. - Those safeguards include GPU monitoring, a multi-cluster control plane and an automated recovery system optimized for inference service level agreements.

Between the lines: - VESSL AI is framing AI infrastructure as a throughput problem, where the winning metric is tokens extracted per unit of hardware rather than raw hardware ownership. - The multi-datacenter approach suggests the company is targeting resilience and geographic flexibility as demand grows for both training and inference. - The emphasis on next-generation GPUs and pre-reservations points to continued supply pressure in advanced accelerator markets.

What's next: - VESSL AI is expected to use the event to market its capacity roadmap and technical capabilities to enterprise, institutional and infrastructure-focused partners. - The company’s near-term milestone is 100MW of AI datacenter capacity by next year. - Its longer-term target is 300MW by 2028. - The company also intends to keep expanding its GPU and reliability engineering stack as agentic AI adoption increases.