Investors interested in Artificial Intelligence should note the rising costs of AI infrastructure, particularly with Nvidia's new Vera Rubin NVL72 rack. Morgan Stanley estimates the bill-of-materials cost for this next-generation AI supercomputing system at approximately $7.8 million, effectively doubling the cost of the current Blackwell NVL72 racks.
#What is Driving the Increased Costs?
The increase in the cost of the Vera Rubin racks can be attributed mainly to the substantial rise in memory prices. High-bandwidth memory technologies such as HBM4 and LPDDR5X have seen a staggering 435% price hike. As a result, memory components now comprise about 25-26% of the total system cost, accounting for nearly $2 million per rack.
Additionally, GPU costs are on the rise, increasing by 57% compared to the previous Blackwell generation, alongside other component price surges. Notably, the costs of printed circuit boards have increased by up to 233%. The growing demand for memory is largely influenced by a strong market for AI and ongoing supply constraints in the semiconductor sector.
#What Features Does the Vera Rubin Rack Offer?
Investing in the Vera Rubin platform offers a revolutionary approach to AI infrastructure, featuring a full-stack design that integrates cutting-edge Rubin GPUs with proprietary Vera CPUs. This new configuration is tailored to enhance performance for agentic AI workloads. The system’s capabilities are significant; it can execute Mixture-of-Experts training with four times fewer GPUs than the previous generation. Furthermore, inference costs are reported to be ten times lower per million tokens when compared to its predecessor.
The initial rollout of Vera CPU racks has already begun, with shipments to leading technology firms such as Anthropic, OpenAI, SpaceX, and Oracle. Volume production is slated for Q4 of 2026, following preliminary deliveries later in that year.
#The Implications for AI Infrastructure Spending
The supply-demand dynamics in the memory sector present a lucrative opportunity for suppliers. With HBM4 seen as a revolutionary technology and a limited manufacturing capacity, the competition for memory resources will inevitably affect pricing across the industry. The emerging trend of memory representing a larger percentage of total rack costs signals a fundamental shift in how profits are realized within AI infrastructure.
In summary, as AI infrastructure continues to evolve, investors should closely monitor these developments. The substantial upward adjustments in pricing for components like memory and CPUs will shape the future landscape of AI technology, influencing both costs and performance efficiencies. Taking note of these trends could be vital for strategic investment decisions in the tech sector.