Upstage plans major AMD chip acquisition

Korean artificial intelligence startup Upstage is in discussions to acquire around 10,000 advanced AI accelerators from Advanced Micro Devices as it seeks to scale domestic computing capacity and reduce reliance on overseas infrastructure, signalling an intensifying race to secure high-performance hardware in Asia’s AI sector.

The proposed deal centres on AMD’s latest data centre-grade accelerators, designed to rival dominant offerings in the global AI chip market. Upstage, known for its large language model development and enterprise AI services, is positioning the potential purchase as a step towards building sovereign computing capabilities within Korea, where demand for generative AI infrastructure is rising sharply across finance, manufacturing and public-sector applications.

People familiar with the discussions indicate that the scale of the planned acquisition reflects both immediate computational needs and longer-term ambitions to establish a domestic AI ecosystem capable of training and deploying advanced models without depending heavily on foreign cloud providers. The move aligns with broader efforts in several countries to localise AI infrastructure amid geopolitical concerns and supply chain constraints.

AMD has been expanding its footprint in AI hardware as companies seek alternatives to entrenched suppliers. Its latest accelerators, part of the MI300 series, are designed to handle large-scale model training and inference workloads with improved memory bandwidth and energy efficiency. Industry analysts view Upstage’s interest as a sign that AMD’s strategy of targeting hyperscalers and emerging AI players is gaining traction beyond North America.

Upstage has grown steadily since its founding, focusing on document processing, natural language understanding and enterprise automation. Its proprietary models have been deployed across industries including banking and legal services, where demand for domain-specific AI solutions continues to expand. Executives at the company have previously emphasised the need for greater control over computing resources, citing cost volatility and limited access to high-performance GPUs as barriers to scaling.

Securing a large volume of accelerators could enable Upstage to train more advanced models locally while offering competitive AI services to regional clients. It also positions the company to compete with larger global players that benefit from extensive infrastructure investments. Market observers note that access to compute power has become a defining factor in the AI race, with startups increasingly seeking direct hardware procurement rather than relying solely on cloud-based capacity.

The discussions come amid tightening supply conditions for AI chips, driven by surging demand from technology firms, research institutions and governments. Export controls and trade restrictions have further complicated access to advanced semiconductors, prompting countries in Asia and Europe to accelerate domestic capacity-building initiatives. Korea, home to major semiconductor manufacturers, has been exploring strategies to strengthen its position across the AI value chain.

Government-backed initiatives have aimed to support AI development through funding, talent programmes and infrastructure investments. The potential deal between Upstage and AMD reflects private-sector momentum complementing these policy efforts, as companies seek to capitalise on the growing importance of AI across economic sectors.

Competition in the AI hardware market has intensified as AMD, Intel and other players attempt to challenge established leaders. While Nvidia maintains a dominant position, particularly in training large language models, customers are increasingly evaluating alternative suppliers to diversify risk and manage costs. AMD’s recent partnerships with cloud providers and enterprise customers highlight a shift towards a more competitive landscape.

Analysts caution that large-scale procurement of accelerators involves significant logistical and operational challenges, including data centre readiness, energy consumption and integration with existing software frameworks. Upstage would need to invest in infrastructure and engineering capabilities to fully utilise the hardware, particularly if it aims to build a self-sufficient AI platform within Korea.

At the same time, the strategic implications extend beyond a single company. A successful deployment of large-scale compute resources by Upstage could encourage other regional players to pursue similar investments, contributing to a broader decentralisation of AI development. This trend could reshape how AI capabilities are distributed globally, reducing the concentration of computing power in a handful of technology hubs.