As chip makers feed the demand for AI, the shortage of DRAM and NAND chips could continue for the rest of the decade. But organizations can take steps to steady their ships.
By: Dave Dimlich
President of SD3IT
In an Inside the Mission post earlier this year, I warned that the growing demand for artificial intelligence infrastructure was creating a perfect storm in the memory chip market. At the time, a serious shortage was approaching as chip manufacturers shifted production toward the specialized memory required by AI data centers, and prices for chips used in smartphones, laptops and other devices started to climb as supplies began drying up.
Since then, the storm has hit with full force, with prices surging for dynamic random-access memory (DRAM) chips used in phones, computers and many other devices, as well as for NAND flash memory chips. Companies are competing for supplies that, in many cases, have already been committed to other customers. And this problem isn’t going away anytime soon.
Some industry projections suggest that memory shortages could persist through 2029 or even 2030. South Korean company SK Hynix, one of the world’s largest memory manufacturers, recently warned that supply constraints could continue through the end of the decade. More than half of its production capacity for 2027 and 2028 is already covered by long-term customer agreements. Samsung likewise expects supplies to be extremely tight through 2028.
Beyond foreseeable projections concerning supply and demand, there is also the possibility of unexpected events intensifying the crunch. J.P. Morgan’s research raises the possibility that a major cyberattack, for example, could generate as much as $145 billion in additional chip demand, which would further tighten supplies. That’s a hypothetical scenario, but it illustrates how dependent critical infrastructure is on physical components such as chips.
The Impact Is Industry-wide
AI’s unparalleled rate of growth is devouring a lot of computing resources. As Inside the Mission recently noted, the Electric Power Research Institute (EPRI) estimates that data centers could consume as much as 17% of U.S. electricity by 2030, a substantial leap from the 4% to 5% it consumes today, as the number of large data centers in the United States could triple in the same timeframe. Memory chips are another example of where feeding AI’s demands have taken precedence in the industry.
AI data centers depend heavily on high-bandwidth memory (HBM), which allows processors to access enormous amounts of data at extremely high speeds. Producing HBM requires specialized manufacturing processes and significant production capacity. As demand has accelerated, manufacturers including Samsung and SK Hynix have shifted resources toward the higher-value products.
The impact of that shift is being felt throughout the industry, as traditional DRAM, the memory used in PCs, servers, smartphones and countless other electronic devices, is becoming harder for companies to get their hands on. NAND flash storage, which is a feature in everything from smartphones, digital cameras and gaming consoles to SD cards and USB flash drives, is also being squeezed.
Price tags tell the story. J.P. Morgan Global Research estimates that by the end of this year DRAM prices will have increased by more than 400% since the beginning of 2024. And prices for software, accessories and storage devices have climbed by about 23% since the end of 2024, according to the Consumer Price Index and the Producer Price Index.
The problem is compounded by the limited number of companies capable of producing memory at scale. Samsung, SK Hynix and Micron dominate much of the global market. When those manufacturers prioritize one category of memory, customers relying on other categories have fewer alternatives. Deloitte projects that the three largest chip makers will invest in increasing their production capacity but that won’t increase supply or cut prices anytime soon, since it can take three to five years to get those types of facilities into production. Chinese chip makers like CXMT also are ramping up production, but U.S. national security rules discourage or prevent buying chips from China.
Steps Toward Weathering the Storm
The market will eventually stabilize, as the semiconductor industry has done before in the face of changing demand. But organizations can’t make plans on the assumption that prices and availability will return to previous levels next year or even the year after.
How much memory the world needs will continue to change. AI data centers may be getting all the attention now, but they’re not the only source of future memory demand. Robotics, autonomous systems, connected devices and other emerging technologies—many if not all of them contingent on AI—will place additional demands of their own on the industry.
So, what should organizations do? Step 1: Don’t panic, since that never solved a problem. But don’t assume the market will correct itself before your next technology refresh. Instead, you can take practical steps toward ensuring your organization can weather the storm.
Improve forecasting: Organizations need to take a proactive approach to refresh cycles. Start with a clear understanding of existing infrastructure, upcoming replacement requirements and projected memory needs. Waiting until a server reaches the end of life to begin looking for replacement hardware is a risky approach in this market. A 30-, 60- or 90-day planning cycle can help procurement and IT teams identify immediate requirements, lock in delivery commitments and standardize platforms where appropriate.
Keep what works: When memory costs increase dramatically, replacing equipment in line with traditional refresh milestones might not always make financial sense. Sometimes, extending the useful life of existing infrastructure, reallocating available resources and optimizing workloads can help organizations postpone unnecessary purchases. You don’t want to hold onto outdated equipment indefinitely, of course, but it can help to base refresh decisions on operational requirements, performance and total cost rather than a preset replacement schedule.
Diversify sourcing: This is a good time to make sure you’re not locked into a single supplier for key components, since supply shortages can hit everyone along the supply chain. Explore alternative sourcing options, identify compatible components and consider forward purchasing agreements when the economics make sense. It could also be beneficial to look at hardware architectures that allow greater component reuse and flexibility. Standardization and open interconnect approaches can make it easier to reuse available memory and reclaim components from decommissioned equipment.
Increase efficiency: The answer isn’t always buying more memory. Better workload management, virtualization strategies, storage optimization and more efficient infrastructure configurations can reduce the amount of additional capacity an organization needs.
Playing the Long Game
Ultimately, the memory shortage is a reminder that no organization is an island, and plans can’t be made in isolation. You need to account for market conditions, potential supply chain disruptions and changing operational requirements. The ability to adapt when circumstances change is essential.
That’s our focus at SD3IT, where we work with federal, military and commercial enterprises to evaluate those challenges, identify the right technology solutions and bring together the manufacturers and partners needed to implement them. As a solution aggregator, our role is to help customers look beyond individual products and find the right combination of technology, performance and cost.
We can’t make memory chips appear when the market is short, but we can help organizations make smarter decisions about the resources they have, the equipment they need and the investments they make.
The perfect storm we saw coming nine months ago has hit, and it isn’t going away. But organizations that plan ahead, stay flexible and make informed decisions, rather than reacting to every change in the market, can limit their damage and emerge in strong shape.
________________________________________
About SD3IT
Solution Driven, Designed and Delivered Technology (SD3IT) provides advanced IT solutions that help organizations modernize infrastructure, enhance security and improve operational performance. The company specializes in zero trust architecture, edge computing, cybersecurity, IoT visibility, data management and supply chain risk management to support mission-critical operations in complex and demanding environments.