Listen to this Post
The NVIDIA GTC (GPU Technology Conference) 2025 has been the epicenter of cutting-edge AI hardware announcements, with the global leader in AI accelerators preparing to introduce its next-generation products. Of particular interest was the unveiling of the Rubin Ultra, NVIDIA’s anticipated successor to the Blackwell series, slated for release in the second half of 2027. This new model will feature HBM4e memory, marking a significant leap in performance for AI applications. However, it’s the ongoing rivalry in high-bandwidth memory (HBM) that has truly captured the industry’s attention.
SK Hynix and Samsung, two of the biggest players in the memory space, are locked in a fierce battle for dominance in the HBM market, with each company vying to secure its place as the primary supplier for NVIDIA’s latest AI chips. SK Hynix, which has been the main supplier of HBM3E to NVIDIA, showcased its advanced HBM4 memory prototype at the conference, while Samsung, still struggling with approval for its HBM3E chips, made strides with its next-generation HBM4 chips.
The Memory Wars: SK Hynix vs. Samsung
At the heart of the competition is the transition from HBM3E to HBM4 memory, a crucial upgrade for AI accelerators like NVIDIA’s future GPUs. The current leader in the space, SK Hynix, has been the dominant supplier of HBM3E memory, with Samsung’s offerings falling behind in terms of approval. Despite a strong push from Samsung, which has now focused on HBM4 chips, it is clear that the competition is far from over. Here’s what happened at GTC 2025:
- NVIDIA’s Rubin Ultra: The long-awaited successor to Blackwell, the Rubin Ultra, is expected to launch in the second half of 2027. It will feature cutting-edge HBM4e memory, setting a new standard for AI accelerators.
– SK
- Samsung’s Setback: Samsung, despite receiving approval from NVIDIA’s CEO Jensen Huang at GTC 2024, has struggled to secure approval for its HBM3E modules. In an attempt to recover, Samsung has shifted its focus toward HBM4, planning to meet the Production Readiness Approval procedure six months ahead of schedule, aiming for completion by mid-2025.
What Undercode Says:
The ongoing HBM memory race underscores a pivotal moment in the AI hardware industry, where the battle for next-gen memory technology will determine the future of high-performance computing. SK Hynix’s continued success with HBM3E and its early lead in HBM4 is not just about securing deals with NVIDIA but also ensuring that it remains a key player in the broader AI chip landscape. By showcasing the 12-layer HBM3E at GTC 2025, SK Hynix sends a clear message to its competitors: it is ahead in the game, both in terms of technology and readiness for the next leap forward.
Samsung, on the other hand, has faced significant setbacks. Despite its strong backing from NVIDIA’s CEO and other strategic moves, its inability to break into NVIDIA’s supply chain with HBM3E has cost the company heavily. The missed opportunities with HBM3E could have far-reaching financial implications, potentially amounting to billions in lost revenue. However, Samsung’s shift to focus on HBM4 chips offers a potential path to redemption. The company’s accelerated timeline for HBM4 production readiness is a strategic pivot, giving it a chance to regain momentum in the race.
Looking beyond the immediate supply chain struggles, the entire memory landscape is set for transformation. As HBM4 chips come online, we can expect a new wave of AI hardware that will significantly improve computational power. For NVIDIA, the integration of HBM4 memory into future GPUs will likely boost the performance of its AI workloads, making its Rubin Ultra series even more powerful. However, much of this will depend on the ability of memory suppliers like SK Hynix and Samsung to meet demand and deliver reliable, high-performance products at scale.
In the end, the battle between SK Hynix and Samsung isn’t just about who can supply NVIDIA; it’s about defining the future of AI computing. With HBM4 on the horizon, both companies are positioning themselves for the next chapter in the high-performance memory market, and it will be fascinating to watch how this race evolves in the coming months.
Fact Checker Results:
- NVIDIA Rubin Ultra: Expected to feature HBM4e memory, launching in the second half of 2027.
2. SK
3.
References:
Reported By: https://www.sammobile.com/news/sk-hynix-delivers-more-pain-to-samsung-at-nvidias-tech-conference/
Extra Source Hub:
https://stackoverflow.com
Wikipedia
Undercode AI
Image Source:
Pexels
Undercode AI DI v2
