The Unprecedented Ascent of the Global Server Market: A Vertical Climb into the AI Era

However, the ASP of $170,220 also suggests that the market is more diverse than the headlines might imply. While there is significant industry chatter regarding the massive Grace-Blackwell NVL72 rackscale systems from Nvidia—which can carry price tags of $4 million to $5 million depending on the configuration—the IDC average indicates that a vast number of customers are still purchasing smaller-scale equipment. A significant portion of the market remains rooted in "fat" server nodes containing four or eight GPUs, or even lower-end systems utilizing PCI-Express cards like the Nvidia RTX Pro 6000. A four-way node in this class might cost between $90,000 and $125,000, while an eight-way version might range from $180,000 to $250,000. These lower-cost configurations are keeping the overall ASP from reaching the multi-million-dollar levels associated with the most advanced rackscale deployments.

The acceleration of the market is not limited to GPUs alone. IDC also reported on the rise of "XPU" systems—machines utilizing specialized matrix math engines designed specifically for AI workloads rather than the more general-purpose capabilities of a standard GPU. Revenue for XPU systems grew by a factor of 3.4X, reaching $27.5 billion in the second quarter. While specific ASPs for these specialized chips are not publicly detailed, estimates suggest they likely cost less than half of a high-end GPU system, placing their ASP around $85,000. This would imply quarterly shipments of approximately 325,000 units. This segment is expected to grow as hyperscalers, cloud providers, and AI model builders increasingly look toward custom, in-house silicon to optimize inference tasks and reduce reliance on third-party vendors.

When these two categories of accelerated computing are combined, the resulting picture is one of total market dominance. Accelerated machines now account for roughly 16.1 percent of total unit shipments but command a staggering 69.1 percent share of total market revenue. In total, approximately 840,000 accelerated units were shipped in the second quarter, accounting for $137.35 billion of the market’s total value.

Despite the focus on AI, "plain vanilla" servers—those without specialized accelerators—continue to see healthy demand. Sales of these traditional systems reached $51.42 billion, a 55.8 percent increase. This growth is being fueled by two primary factors: a general server refresh cycle as aging infrastructure is replaced, and the emergence of "AI sandbox" systems. These are standard servers used to execute the code and management processes generated by GenAI models, acting as the bridge between AI logic and real-world application. North of 4 million such servers were sold in the quarter, representing a 16.7 percent increase in volume. Across all categories, the total global market moved approximately 5.16 million machines during the three-month window.