South Korea's Ministry of Science and ICT has detailed its budget for introducing Nvidia's next-generation AI chip, Vera Rubin, in 2027, setting the cost at 27.5 billion won per rack, according to Money Today. This figure is significantly higher than earlier media expectations.
Budget and Configuration
The total budget for GPU server expansion next year stands at 3.85 trillion won, entirely allocated for purchasing 140 Vera Rubin NVL72 server racks. Each NVL72 rack integrates 72 Rubin GPUs and 36 Vera CPUs, resulting in a total of 10,080 Rubin GPUs. This procurement is separate from the ministry's previously announced plan to secure 52,000 advanced GPUs by 2028. Actual supply may be adjusted depending on Nvidia's delivery status.
Exceeding Earlier Estimates
The 27.5 billion won per rack price tag notably surpasses earlier predictions. Investment bank Bernstein estimated in June that a Vera Rubin NVL72 rack would cost around $9.1 million (approximately 12.3 billion won), while Morgan Stanley estimated about $7.8 million (around 10.5 billion won). The South Korean government stated that the budget accounts for potential price increases due to GPU shortages and auxiliary infrastructure costs.
Purpose for Frontier Model Development
The batch of chips will be used for developing next-generation frontier models, including AGI. Deputy Prime Minister and Minister of Science and ICT Lee Jong-ho said in June that training a model comparable to Anthropic's Mythos would require roughly 10,000 Vera Rubin GPUs, making the scale of these 140 racks sufficient for that purpose.