Nvidia has begun delivering its first Vera Rubin systems to Microsoft while simultaneously notifying the broader customer base of a 15% increase in AI server pricing.
Nvidia’s first production units of its next-generation AI computing platform, “Vera Rubin,” have arrived at Microsoft data centers. Microsoft CEO Satya Nadella announced the milestone on social media, writing, “Delivery day for the first production Vera Rubin systems arriving at Microsoft data centers. Grateful to Nvidia and the Azure hardware and data center teams for making this important milestone possible.” Nvidia simultaneously confirmed via its official channels that Vera Rubin has entered full-scale production ramp-up.
In a separate development, Nvidia has reportedly notified major customers—including Microsoft, Google, and Oracle—that prices for AI chip systems will rise by more than 15% beginning with shipments early next year. According to Bloomberg, citing industry sources, the adjustment applies across multiple product lines, including systems equipped with Vera Rubin and Grace Blackwell architectures. The exact magnitude of the increase varies by configuration depending on the Nvidia chip generation and memory setup.
The flagship configuration, the Vera Rubin NVL72, pairs 36 “Vera” central processing units with 72 “Rubin” graphics processing units. Compared to the preceding Blackwell platform, it delivers up to 10 times higher AI inference throughput per watt and reduces cost per token to approximately one-tenth, according to figures disclosed during Nvidia’s 2026 CES keynote and reported by Tom’s Hardware.
The impending price hike stems from severe cost pressures in the memory semiconductor sector, driven by an acute high-bandwidth memory (HBM) supply crunch. As AI accelerator demand skyrockets, so does the requirement for high-performance memory. South Korea’s Samsung Electronics (005930.KS) and SK Hynix (000660.KS), alongside Micron, dominate the global DRAM market as a three-way oligopoly. Although all three are expanding production, AI-driven demand continues to outpace supply growth. Market research firm TrendForce reports that DRAM contract prices surged more than 75% quarter-over-quarter in the fourth quarter of 2025, with an additional average increase of 90–95% projected for the first quarter of 2026. Server DRAM alone is expected to rise more than 60% in Q1, underscoring how server and AI workloads are fueling the escalation. To mitigate HBM constraints, Nvidia is reportedly even considering scaling back the memory capacity originally planned for “Rubin Ultra,” the high-performance successor platform to Vera Rubin.
Attention is also focused on when Vera Rubin will reach South Korea. Following a meeting with Nvidia CEO Jensen Huang on June 8, Deputy Prime Minister and Minister of Science and ICT Bae Sang-hoon stated that “South Korea has been designated as a top-priority recipient for Vera Rubin supply.” However, specific delivery timelines and volumes for the region remain undisclosed. While the South Korean government and Nvidia previously agreed to supply 260,000 GPUs to the country, Vera Rubin allocations were not separately specified. During the same meeting, Bae noted that GPUs required for future expansion projects would also be “supplied without issue.”
The announced price adjustments are widely viewed as evidence that memory manufacturers—Samsung Electronics, SK Hynix, and Micron—are gaining substantially stronger bargaining leverage. Nvidia’s AI accelerators rely heavily on integrated DRAM, including HBM, to rapidly feed data into computational processes. Meanwhile, hyperscalers such as Microsoft, Google, Amazon, and Meta continue developing proprietary AI chips, yet still depend on Nvidia for large-scale data center deployments due to its optimized silicon and mature ecosystem. Concurrently, securing reliable memory supplies from Samsung, SK Hynix, and Micron remains critical for these cloud providers seeking to reduce their reliance on Nvidia.
Industry observers warn that Nvidia’s price increase could further complicate large-scale infrastructure investment plans. Major AI data center projects are already grappling with construction delays, labor shortages, deteriorating capital market conditions, and community opposition. The added cost pressure may intensify financial and logistical challenges across the sector.