According to Samsung Electronics, high-bandwidth memory will make up close to 30% of the overall wafer capacity of the global DRAM market in 2027, which demonstrates just how fast the development of artificial intelligence is shifting the focus of semiconductor fabrication and putting stress on regular memory production.
Artificial intelligence development does not only generate the need for better processors, but also requires fast memory that can transfer large amounts of information from accelerators to computer systems. High-bandwidth memory (HBM) has become such an essential technology in fulfilling this need, that even its production requirements are now having an effect on the whole DRAM market.
In an article posted by Reuters on September 29, 2026, Samsung Electronics Executive Vice President, Kim Taewoo stated that High Bandwidth Memory could occupy up to 30% of the total wafer production capacity of DRAM chip manufacturers, compared to just 20% at present. Because HBM and conventional DRAM compete for wafer capacity, the shift could place additional pressure on supplies of standard memory products.
AI Is Redirecting DRAM Capacity Toward HBM
The design of HBM is meant to deliver high memory bandwidth through stacking of multiple DRAM dies and tight coupling with AI processors and other high-performance computing hardware. Such a setup is increasingly crucial as AI models get more resource-intensive. The training and inference of AI models necessitates processors to access huge amounts of data fast; hence, memory bandwidth becomes a critical aspect.
Therefore, the rising demand for HBM is driving the way memory producers distribute their manufacturing capabilities. Rather than considering HBM as a somewhat niche segment within the memory market, the producers are starting to use sizable amounts of DRAM wafer capacity on AI-specific memory products. The latest projection by Samsung highlights the extent of this trend. An increase from about 20% of DRAM wafer capacity industrywide today to 30% by 2027 would be a major rise in production capacity dedicated to HBM.
Why the Shift Matters for Conventional DRAM
The impact extends beyond HBM itself.
Both HBM and traditional DRAM utilize similar production facilities, and therefore, manufacturers need to allocate capacity for AI memory and other devices, such as personal computers, phones, servers, and electronic components.
As stated by Samsung’s Kim, the production capacities of HBM and traditional DRAM are in competition for the same number of wafers. The more resources are used to produce HBM memory, the less room is left for the production of traditional DRAM.
Nevertheless, it would be wrong to consider the transition process as a direct substitution of the production volumes. Manufacturing effectiveness, technical upgrades, increases in yields, and new capacities may affect the final result.
The key issue is that AI is changing the composition of memory demand, making capacity allocation an increasingly important consideration for semiconductor manufacturers.
HBM Demand Is Creating a New Memory Supply Equation
The changing capacity mix comes as AI server investment continues to support demand for HBM.
According to a report from TrendForce published on September 29, the growth in AI server demand is putting HBM and traditional DRAM chips into a competitive state as far as advanced manufacturing process and wafer capacity are concerned. The company predicts that the memory market will be tight until at least 2027.
This creates a more complex supply equation for memory manufacturers.
On one side, AI companies and data center providers will need more and more HBM to support their advanced computing capabilities. And on the other hand, conventional memory segments still rely on conventional DRAM technology. As such, manufacturers will need to find the right balance in terms of the production volumes for different segments depending on profitability and demands.
It is possible that in this manner, memory industry might start experiencing the impact of the developments in AI inference infrastructure.
HBM4 Is Adding to the Industry's Capacity Challenge
The transition to newer generations of HBM is another important factor.
According to TrendForce, its 2027 HBM forecast takes into account the growing role of high-priced HBM4 chips and the slow rise of HBM4e in the second half of 2027. TrendForce expects the HBM market to continue facing supply constraints, as it will take some time for the next generation of products to increase in volume and production yield.
Next-generation HBM products will be produced through advanced manufacturing and packaging processes as well. As AI accelerators get increasingly complex, memory needs to catch up with the changes in processor architecture.
What this implies is that increasing HBM manufacturing is about more than producing more conventional DRAM chips. There is also stacking, packaging, testing, yields, and many other requirements to consider.
Samsung Is Expanding Its HBM Production
Samsung's forecast comes as the company increases its focus on the HBM market.
TrendForce's industry estimates point to the possibility of Samsung's average monthly HBM wafer input to grow from 180,000 wafers in 2026 to 250,000 wafers in 2027. This will amount to a growth of almost 40%. According to the same estimate, HBM4-family memory chips can make up about 80% of Samsung's HBM shipment in 2027, compared to 40% in 2026.
This provides a picture of the extent of the transition occurring at Samsung's memory division. It is not merely a case of ramping up HBM memory chip production but also one of switching to newer generations of the product. From a broader market perspective, higher production at Samsung will add to supply but also point to how much DRAM production capacity is being directed toward AI memory.
What the Shift Means for the Global Semiconductor Supply Chain
Increased amounts of DRAM capacity used for HBM have multiple consequences for the electronics industry. AI infrastructure companies require increased amounts of HBM capacity to scale the production of accelerators. At the same time, companies producing memory can use HBM to enter one of the fastest-growing sectors of semiconductor demand. However, for those firms that are dependent on the conventional types of DRAM, the issue of capacity allocation becomes especially important.
PC producers, smartphone producers, server producers, and other electronics companies can find themselves in a market situation where AI-driven demand determines the availability of the memory components traditionally used in the production of their own devices. It does not mean that the shortage of conventional DRAM will occur immediately; however, this balance will become more and more important in the future.
Samsung's Projection Signals a Structural Change in Memory Manufacturing
The importance of Samsung’s forecast is not limited to the percentage value mentioned above. The fact that the ratio of wafer capacity has changed from roughly one-fifth of the DRAM wafers to almost one-third in a relatively short period shows how quickly AI is reshaping the priorities of the semiconductor manufacturing industry.
Currently, memory chipmakers have to produce their products based on the demands of AI accelerators, and not just the standard memory requirements. HBM requires better collaboration between memory chipmakers, processor makers, advanced packaging vendors, and AI infrastructure firms.
Outlook for the AI Memory Market
It seems unlikely that the fight for the available DRAM wafer capacity will stop in the coming years as investments in AI data centers continue. The three most notable players in the HBM market are Samsung, SK hynix, and Micron, and the producers of AI accelerators keep working on more and more complex systems.
Thus, the coming years will not be only about the growth of HBM output. Producers will have to boost their yields, develop the next HBM generations, expand advanced packaging techniques, and balance the production of AI memory and standard DRAM.
According to Samsung's projection, in 2027, the HBM market might take up almost 30% of the DRAM wafer capacity in the industry clearly illustrates the direction in which the memory industry moves. AI is no longer just another area of application generating demand for semiconductors. It changes the allocation of the existing manufacturing capacity.
