How AI Is Redesigning the Memory Supply Chain
19 November 2025
The rapid expansion of AI is placing intense pressure on global chip and memory production, making component availability a key factor in industrial competitiveness.
OpenAI, Samsung, and SK Hynix have signed an agreement to purchase 900,000 DRAM (dynamic random-access memory) wafers per month accounting for nearly 40% of global production. This marks the beginning of a new era for the semiconductor market. The deal is part of the “Stargate” project by OpenAI in collaboration with SoftBank and Oracle and aims to build a global network of AI data centers in an initiative worth approximately $500 billion that seeks to strengthen OpenAI’s U.S. AI infrastructure through the construction of large facilities by the end of 2025. However, the Stargate project is already reshaping the global supply chain: while hyperscalers (large cloud service providers) and big players like OpenAI are securing their supplies, medium and small manufacturers are facing rising prices and component shortages.
Under the agreements signed between OpenAI, Samsung and SK Hynix, Samsung has committed to providing advanced DRAM and heterogeneous integration solutions.
This agreement has prompted other hyperscalers to secure multi-year supplies , exhausting entire V9 NAND and Micron 2026 HBM production lines ahead of schedule. Micron has also almost exhausted its HBM3E production and started manufacturing the new 11Gbps HBM4, which has intensified global competition for resources.
The result is a market dominated by hyperscalers, while smaller players, such as European suppliers, face high prices, limited supplies, and longer lead times.
The current shortage of chips and memory devices is not a spontaneous occurrence: rather it is the result of targeted industrial strategies, a shortage of skilled labor, and geopolitical tensions.
In recent years, major chipmakers have shifted production toward high-margin segments such as HBM (high-bandwidth memory) and AI memory, reducing production of traditional DRAM and NAND (flash memory). The DRAM and NAND sectors were already affected by the crashes in 2019 and 2022.
Accumulated losses after overproduction have prompted many companies to downsize their less profitable operations and focus on AI-driven technologies.
At the same time, the shortage of skilled technicians is limiting production capacity. In South Korea and Taiwan, retirements among experienced staff are causing bottlenecks in production and plant maintenance.
Complicating matters further, China controls the majority of the rare earth elements required for memory and magnetic components. Export restrictions to the US and the EU are exacerbating the vulnerability of the global supply chain and making manufacturers dependent on a small number of key players.
In short, the current crisis stems from a combination of aggressive industrial strategies, skills shortages, and geopolitical tensions.
This imbalance between demand and production affects more than just DRAM chips; it is also transforming the storage market.
Data centers for AI are rapidly replacing HDDs with SSDs, even in cases where the cost had previously hindered adoption. The widespread use of QLC flash arrays for “warm” data is accelerating the demand for NAND and reducing its availability even further.
Manufacturers are allocating much of their available capacity to large cloud operators, which is reducing resources for other market segments.
The paradox is clear: the more SSDs are adopted, the scarcer and more expensive they become. Delivery times for QLC arrays now exceed six months, impacting even sectors not directly related to AI.
Since Q22025, DRAM and NAND prices have been rising steadily, with DRAM spot prices increasing by 160% in just one month and NAND up 50% since April, surpassing the peaks seen during the 2016–2018 memory boom, according to Morgan Stanley. Samsung has raised DDR5 prices by up to 60%, while SK Hynix reports that production of HBM, DRAM, and NAND for 2026 is already “essentially sold out.”
These increases are reflected in concrete examples: 32GB DDR5 modules jumped from $149 to $239, while 16GB and 128GB modules rose around 50%, and 64GB and 96GB modules saw gains above 30%. Supply is so tight that Samsung even delayed the announcement of October pricing, putting server makers and data center operators under significant pressure. Many customers are now accepting that they will not receive the volumes requested and are willing to pay substantial premiums to secure part of the available supply.
In this context, large hyperscalers can manage scarcity and price hikes thanks to their scale and long-term supply contracts, while smaller players face longer lead times and rising costs. With Samsung’s new fabs not expected to be operational before 2028, supply chain pressures are likely to persist, consolidating a scenario of sustained tension in the memory market.
Looking ahead, the memory market is anticipating two possible scenarios.
If AI growth continues, demand for high-performance memory will remain high.
The AI boom is already pushing up prices and reducing chip availability — a situation that, according to analysts, could last one or two years.
Demand for AI-driven storage is accelerating the shift from HDDs to nearline SSDs, driving growth across the entire memory sector.
Conversely, if enthusiasm for AI were to wane, demand could contract rapidly.
The Bank of England warns that markets could undergo a sudden correction if the AI euphoria fades. In such a scenario, overcapacity and advance investments could generate a boom-bust cycle, resulting in a decline in prices similar to that seen in previous phases of overproduction.
The global race for memory storage is a turning point for the digital economy. Artificial intelligence is reshaping industrial priorities, production models, and supply chains. In a market dominated by large hyperscalers, European companies are faced with the challenge of ensuring continuity and technological autonomy.
This is where Electronic Partner comes in.
In a market dominated by hyperscalers and characterized by growing scarcity of chips and memory, maintaining stable and reliable supplies is a decisive competitive advantage. Electronic Partner is a key partner for European companies in this scenario, offering flexibility, access to a global network of trusted suppliers, and specialized consultancy services.
Thanks to its independent position and diversified supply chain, Electronic Partner enables companies to mitigate the risks associated with market volatility and ensure business continuity, even during periods of heightened supply chain pressure.
Our goal is to transform the supply crisis into an opportunity for resilience and growth by leveraging expertise and relationships to promote the stability of our European customers.
Newsletter archives:
How AI Is Redesigning the Memory Supply Chain
Blog article archives:
Search news articles: