News
Hyperscaler memory costs projected to drive one-third of capital expenditure increases through 2027
Thursday, February 12, 2026 at 12:06 PM
UBS has significantly adjusted its capital expenditure forecasts for major hyperscalers, predicting $827 billion in 2026 and $915 billion in 2027. Memory costs are expected to be a primary driver, surging from $53 billion in 2025 to $252 billion by 2027. This shift implies memory will account for one-third of the annual CapEx increase. On a component level, memory's share of the bill of materials for an NVL72 rack is projected to rise from 6% to 16% as systems transition from the GB200 to the VR200 platform.
Context
UBS recently updated its forecasts for major hyperscalers, projecting capital expenditure to reach $827 billion in 2026 and $915 billion by 2027. A primary driver of this surge is the skyrocketing cost of memory components, which is expected to account for roughly one-third of the total CAPEX increase this year. Total memory spending by tech giants is set to balloon from $53 billion in 2025 to a staggering $252 billion by 2027. This represents a fundamental shift in data center economics as AI workloads demand significantly more high-performance storage.
For Nvidia, this trend signals a major transformation in system architecture. Within the high-end NVL72 rack, the memory component’s share of the total bill of materials is projected to jump from 6% in the GB200 platform to 16% in the upcoming VR200 series. Additionally, memory costs for general-purpose servers are expected to rise by $10,000 per unit. These figures underscore the increasing dominance of memory costs in the AI supply chain as next-generation platforms require denser, more expensive hardware configurations.
Related Companies
Nvidia
NVDA