Samsung Electronics and SK Hynix are getting ready to start mass production of HBM4 (High Bandwidth Memory) next year. These advanced memory chips are a key part of modern AI data centers and are mainly used in powerful processors made by Nvidia.
According to reports from South Korea, Samsung is moving faster than its rival and plans to begin production much earlier in 2026.
Samsung Aims to Start First
Supply chain sources say Samsung plans to begin mass production of HBM4 as early as February 2026. This early start could give Samsung a major advantage in winning large supply contracts.
SK Hynix, on the other hand, is expected to scale up full production around September 2026. While both companies are ahead of other rivals like Micron, Samsung’s head start may help it secure early orders from Nvidia.
Most of these HBM4 chips are expected to be used in Nvidia’s next-generation AI platform, known internally as Vera Rubin.
Different Approaches to Building HBM4
HBM4 introduces an important design change. For the first time, the base die—the logic layer that controls the memory stack—is being custom-built instead of using a standard design.
-
SK Hynix is working with TSMC, using its 12nm process to make the base die. This helps ensure smooth compatibility with Nvidia’s AI chips.
-
Samsung is doing everything in-house. As a fully integrated chipmaker, Samsung will use its own 10nm process to build both the memory and the logic layer.
Despite the different strategies, both companies are targeting big performance gains.
Big Performance and Power Efficiency Boost
HBM4 is expected to offer:
-
Up to double the bandwidth of current HBM3E memory
-
Around 40% better power efficiency
These improvements are especially important for AI data centers, where electricity use and heat management are becoming serious challenges.
High Production Capacity, But No Relief for Consumers
Samsung currently has a slight edge in production scale. Reports suggest:
-
Samsung can produce around 170,000 HBM units per month
-
SK Hynix produces around 160,000 units per month
In overall DRAM production, Samsung’s lead is even larger.
However, this does not mean cheaper RAM for consumers. Almost all HBM4 production planned for 2026 is already reserved by large AI companies. As a result, memory factories are prioritising AI chips over consumer RAM like DDR5, which means memory shortages for phones and PCs may continue through 2026.
Why This Matters for the AI Industry
With Micron lagging behind in HBM4 timelines, Samsung and SK Hynix are effectively the only major suppliers of this technology in 2026. This gives both companies strong pricing power and influence over the AI hardware market.
For Samsung, the early HBM4 launch is especially important. The company reportedly struggled to get Nvidia approval for earlier HBM3E chips, allowing SK Hynix to gain ground. By moving early in 2026, Samsung is trying to regain its position as Nvidia’s top memory supplier.
For Nvidia, having two strong suppliers is a big advantage. It reduces the risk of shortages and delays when launching new AI hardware.
What to Watch Next
As early 2026 approaches, the industry will closely watch:
-
Whether Nvidia officially approves Samsung’s HBM4 chips
-
How smoothly Samsung’s early production runs perform
-
Whether SK Hynix can catch up later in the year
The success of these early HBM4 launches could shape the future of AI hardware—and the global memory market—for years to come.
