Samsung and SK Hynix Gain Bargaining Power as Nvidia AI Server Costs Surge

Server manufacturers have notified major data center customers—including Microsoft, Google, and Oracle—that prices for Nvidia-based AI systems will increase by more than 15 percent in many cases for shipments starting in early 2027. According to reports, the price adjustments affect systems built on Nvidia's next-generation Vera Rubin and current Grace Blackwell platforms, with the exact scale of the increase varying by chip generation and memory configuration.

Memory components have emerged as a primary driver of this cost pressure. Nvidia's AI accelerators depend heavily on high-bandwidth memory (HBM), while AI data centers consume substantial volumes of conventional server DRAM. Supply remains tight as cloud service providers expand computing capacity and secure memory supplies through longer-term contracts, placing a small group of memory manufacturers in a stronger negotiating position.

Data from Counterpoint Research places SK hynix at 58 percent of global HBM revenue in the first quarter, while Samsung led the broader DRAM market in the second quarter with a 39 percent share. Together, both South Korean chipmakers account for 65 percent of global revenue. SK hynix established an early lead in HBM tied directly to AI accelerators, whereas Samsung maintained a larger footprint in conventional DRAM while expanding its share in advanced HBM products. The bulk of first-quarter HBM revenue stemmed from HBM3E, with HBM4 shipments expected to become more meaningful in the second half of the year.

Cost pressures are extending across the wider DRAM sector as well. TrendForce projected server DRAM contract prices to rise another 13 to 18 percent in the third quarter compared to the prior quarter, driven by continuous AI-related absorption of manufacturing capacity. South Korean investment bank Hana Securities noted that these price increases put renewed attention on memory suppliers, identifying Samsung and SK hynix as prime beneficiaries amid sustained infrastructure spending projected through 2027.

Demand for high-performance memory remains diversified, with companies including Amazon, Microsoft, Google, and Meta developing proprietary AI processors that also require advanced DRAM and HBM. However, climbing component prices are elevating overall infrastructure expenses. Server suppliers have warned customers of price hikes of approximately 17 percent on major Nvidia-based systems, estimating that chip-system costs for a 1-gigawatt data center could rise by at least $5 billion.

Addressing the long-term impact on technology firms, Shin Joong-ho, head of research at LS Securities, highlighted the constraints of rising component expenses. Big Tech can absorb higher costs temporarily, but there is a limit to how far infrastructure spending can rise without affecting returns. If memory inflation continues to push up Nvidia system prices, companies may eventually need to reassess the expected returns on new AI data center projects.

"The current surge in AI server costs highlights a critical bottleneck in the hardware supply chain, where component manufacturers hold significant leverage over downstream technology giants. As memory prices continue to climb and infrastructure expenses expand, capital allocation strategies for large-scale data centers will face increasing scrutiny. Businesses must carefully evaluate long-term return on investment as hardware inflation begins to impact project economics across the technology sector." — Dr. Shishir Gupta, Founder & CEO, StartupLanes