Report: As NVIDIA Rubin Ramps Up, Cloud Giants Have Booked All Long-Term Contracts for 2027, Memory Shortage to Extend into 2028

As NVIDIA's next-generation AI accelerator, Vera Rubin, enters its mass production phase, the global memory market is facing a deepening supply crisis. Supply chain sources revealed that major cloud service providers have locked in all long-term agreement (LTA) capacity for 2027 and are positioning early for 2028 supplies, with memory shortage pressures expected to intensify continuously from the second half of 2026.

According to Digitimes Asia on Wednesday, citing supply chain sources, leading cloud service providers (CSPs) have booked all available LTA capacity for 2027 and are actively advancing supply negotiations for 2028.

Meanwhile, several memory module manufacturers recently received notifications from original equipment manufacturers (OEMs) stating that no additional supply can be provided beyond previously committed volumes, as upstream capacity has been largely redirected to meet demand for NVIDIA-related AI servers and CSPs.

This supply landscape is transmitting effects throughout the entire industry chain. Shortages in the DRAM and NAND markets are expected to widen further starting in the second half of 2026, with the industry widely anticipating that the severity of the shortage in 2027 will surpass that of 2026. For investors, memory prices are expected to remain strong over the next two to three years, and the market remains broadly optimistic about the potential for contract price increases among memory manufacturers.

Cloud Giants Rush to Lock Capacity, 2027 Long-Term Contracts Exhausted

According to supply chain sources, as NVIDIA Vera Rubin is scheduled to begin shipments in the second half of 2026 and HBM4 enters mass production, major cloud service providers are continuously increasing investments in AI data center expansion, turning the competition for memory capacity into a white-hot battle.

Upstream capacity has been massively directed toward NVIDIA AI server and CSP demands, with nearly all capacity for 2027 already committed to CSP clients. Several memory module manufacturers recently received notices from OEMs explicitly stating that no additional supply can be provided beyond previously committed amounts, leaving extremely limited available surplus in the market.

At the same time, supply chain sources pointed out that Apple has continuously locked in memory capacity for the third quarter for new product launches, while other brand customers are forced to advance production plans under the pressure of tight supply, further compressing the available capacity space in the market.

2028 Supply Negotiations Start Early, Price Expectations Remain Strong

After securing 2027 capacity, leading CSPs have initiated LTA supply negotiations for 2028 ahead of schedule. According to industry sources, upstream suppliers were previously unwilling to release 2028 capacity before May 2026, but have recently begun accepting discussions for Q1 2028 orders, with initial allocations for some HBM and server capacity already completed.

Industry sources noted that LTA structures vary by supplier; only a few require upfront deposits, while most models involve customers committing to expected purchase volumes first, allowing suppliers to adjust expansion timelines accordingly, with final selling prices confirmed only before actual shipment. This arrangement indicates that memory manufacturers maintain full confidence in the upside potential for contract prices over the next two to three years.

There were previous rumors that CSPs intended to fund the construction of dedicated production lines for memory manufacturers and subsidize equipment purchases. However, industry insiders stated that upstream suppliers are currently enjoying robust profitability, with new fab and equipment budgets planned independently, without needing to rely on single customers for line construction. Nevertheless, CSPs are still actively striving for long-term memory contracts to avoid a recurrence of the supply crunch seen from 2025 through the second half of 2026.

Capacity Crowding-Out Effect Spreads, AI Server Configurations See Downgrades

Supply chain sources pointed out that server memory and standard memory are highly similar in design architecture, unlike embedded memory, meaning that suppliers' capacity allocation primarily squeezes standard memory and PC memory—together accounting for approximately 60% to 70% of the overall DRAM market. As a result, the share of PC memory in total DRAM volume has dropped from the previous 11% to 12% to about 9%.

Some memory module manufacturers have also adopted prepayment models, paying suppliers in advance at lower agreed-upon prices and making up the difference when the original manufacturers ship at market rates. While this approach increases short-term financial pressure, it helps secure critical capacity in advance.

On the demand side, high prices and tight supply are driving signs of configuration downgrades in AI servers. According to industry sources, 128GB was previously the mainstream specification, but some customers have shifted to 64GB or 96GB configurations to control costs. Although this change has a limited impact on overall memory usage, it highlights that soaring memory costs are slowing the upgrade pace of AI servers.

NVIDIA CEO Confirms Full Support for Vera Rubin from Three Major Memory Manufacturers

NVIDIA CEO Jensen Huang recently publicly confirmed that SK Hynix, Samsung Electronics, and Micron Technology have all completed HBM4 certification, started production, and are fully supporting the Vera Rubin platform.

Addressing rumors about a significant reduction in memory usage, Huang stated that future systems will still heavily utilize high-speed memory, but the shortage issue needs to be managed rationally across various systems, with further expansion of supply.

The challenge currently facing memory manufacturers lies in supporting the massive rollout of Vera Rubin while simultaneously meeting the rising demand for end-user devices and edge AI. Industry consensus is increasingly converging on the judgment that "the 2027 shortage will be more severe than that of 2026," with the cycle of tight memory supply expected to extend into 2028.