The memory industry is now witnessing a major shift, as DRAM suppliers are now looking to lock in customers into 'multi-year' contracts, essentially guaranteeing demand.
Samsung Entering Into Long-Term Contracts Means Shortages Might Last Till The End of the Decade
It appears that the demand for memory products among Big Tech is so massive that they are willing to take the risk of entering long-term agreements with suppliers like Samsung and Micron, which is seen as a rather 'untraditional' move within the industry. A report by the Korean media outlet EBN reveals that hyperscalers are entering into long-term memory contracts with firms like Google and Microsoft, which lead the list. The idea with such contracts is to fix long-term volume while keeping spot pricing at its current level. By bagging such deals, memory manufacturers are looking to factor in demand that potentially extends the cycle much further.
If concluded in this manner, Samsung Electronics secures long-term demand visibility, enabling it to accelerate capacity expansion and prevent sharp price drops like in the past by avoiding inventory buildup.
- EBN
We have already discussed Samsung's co-CEO proposing such a model, and, from a business perspective, such contracts factor out the uncertainty involved in investing in capacity buildout and the risk of overinvestment. But at the same time, in the consumer industry, this basically means most of the future capacity built out would be dedicated to the AI sector, so shortages wouldn't ease anytime soon and instead would extend beyond traditional timelines, which is disappointing.
Hyperscalers' DRAM demand isn't driven solely by the infrastructure buildout; rather, efforts to deploy custom chips are expected to increase significantly as we move towards inference. This particular workload area is one where performance isn't the only priority; factors like throughput, TCO per token, latency, and others become significant, which is why ASICs are seen as a viable replacement. Companies like Google with TPUs, Microsoft with Maia, Meta with MTIA, and Amazon with Trainium are looking to increase the split between GPU-ASIC-based compute for inference, which is why locking HBM supply is vital to them.
It is indeed disappointing to see how the memory industry is evolving, and how, with each development, the priority given to the average consumer is decreasing. Initial estimates suggested the shortages would ease by the middle of 2027, but this could extend well beyond that.
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
