Amid the RAM Crisis, Intel Launches Counterattack with ZAM: The Chip Designed to Challenge South Korean Dominance
Not long ago, few could have predicted Intel’s dramatic turnaround. The company that once ruled the consumer and server processor markets for decades endured a tough period under AMD’s shadow. Now, they’re making a strong comeback, not only positioning themselves as a major U.S. foundry through rescues but also aiming to take a bite out of South Korea’s massive RAM industry with their new ZAM memory.
Their key innovation lies in three-dimensional design.
Z for ‘Zolution’
Think back to math class when you first drew a cube. The X-axis runs east-west, the Y-axis north-south. To turn a square into a cube, you need the Z-axis for up and down. That’s the concept engineers at SAIMEMORY—a collaboration between Japan’s SoftBank and Intel—have applied to traditional DRAM memory. Their goal is to capture a share of the high-bandwidth memory (HBM) market that dominates data centers.
Recently, we reported that RAM manufacturers are frustrated with tech companies buying components “just in case,” leading to stricter policies.
Layered Like Pastry
A few months back, we covered how these two firms joined forces to challenge the dominance of Samsung, SK Hynix, and Micron in high-performance memory production. HBM is favored in data centers for its massive bandwidth, enabling more simultaneous operations—like a vast highway. However, it has drawbacks: high production costs, significant energy use, and intense heat requiring expensive cooling systems.
Standard DRAM wasn’t a viable alternative until Intel and SoftBank experimented with stacked DRAM. Simplifying it, this is like layering memory in a pastry, but the main challenge was connecting those thin layers to match HBM’s highway-like capabilities.
Introducing ZAM
After months of research, SAIMEMORY and Intel unveiled the ZAM prototype at the Intel Connection event in Japan. According to the companies, a ZAM module can reach up to 512 GB capacity, is straightforward to produce through vertically stacked chips, and crucially, it cuts energy consumption by 40% to 50% compared to conventional HBM.
While HBM is expensive and slow to manufacture, ZAM is cheaper, could ease supply chain bottlenecks, reduce data center energy use (a major issue), and simplify cooling. Current research suggests a theoretical limit of 20 layers, though designs hover around 16, so performance could improve by exceeding that.
The RAM price situation has become so dire that some are even building their own memory at home.
A Genuine Alternative
Intel’s ambitions are bold, claiming their DRAM module bonding technology delivers two to three times the capacity of HBM modules while being up to 60% cheaper to produce.
It all sounds advantageous, and it’s not a fringe idea—established HBM giants like Samsung are also exploring ways to overcome connection limits in stacked DRAM.
Ambitious Alliance
Equally significant as the ZAM prototype is the partnership itself. Intel has been absent from the memory market for years, with failed attempts in the 1980s and later with Optane technology, which flopped without gaining traction. SoftBank represents Japan, which led the sector in the 1980s but was overtaken by rising South Korean firms. In fact, Japanese memories displaced Intel’s, only to be eclipsed by South Koreans.
SAIMEMORY is backed not just by these heavyweights but also by Japanese companies like Fujitsu, Shinko Electric Industries, PowerChip Semiconductor Manufacturing, and the University of Tokyo. If ZAM succeeds commercially, it could relieve memory production chains (potentially easing the strained consumer market overwhelmed by data center demands) and establish a new contender to disrupt the current leading trio.
We’ll see it in action in a few years, as SAIMEMORY plans to complete prototypes in fiscal year 2027 and begin sales in 2029.
Image | Samsung, Maxence Pira
In Xataka | Nothing’s CEO believes we don’t need a high-end phone every year. It’s a mix of RAM crisis and common sense.
