.jpeg)
In a major development for the global semiconductor and AI industries, Samsung Electronics and SK Hynix have signed letters of intent to supply advanced memory chips to OpenAI, in connection with its ambitious Stargate project.
This partnership signals a new chapter in how AI infrastructure is being built: not just through compute (GPUs, accelerators) but by ensuring the memory backbone that keeps everything humming. Below, we explore what’s known so far, the implications, and what to watch going forward.
What Is the Stargate Project?
Launched in January 2025, Stargate is a multibillion-dollar AI infrastructure initiative spearheaded by OpenAI (in collaboration with partners like SoftBank and Oracle).
Its goals include:
- Building a network of high-performance AI data centers across the U.S. and beyond
- Scaling compute capacity (in the order of gigawatts)
- Securing supply chains for the chips, memory, cooling, energy, and hardware that future AI workloads demand
Because modern AI models (especially large ones) require not just raw compute but vast memory bandwidth and low latency, memory (especially DRAM and high-bandwidth memory, or HBM) becomes a critical bottleneck.
OpenAI expects the chip demand for Stargate to reach 900,000 wafers per month, a figure that dwarfs many existing memory supply volumes.
What Samsung and SK Hynix Bring to the Table
Dominance in Memory
Samsung and SK Hynix already command a large share of the global memory market:
- They jointly hold about 70% of the DRAM market globally.
- For HBM (especially relevant for AI accelerators), their share is even higher — nearing 80% in advanced memory technologies.
Because of that dominance, scaling to meet Stargate’s demand gives OpenAI a more secure route to memory supply, rather than relying on smaller or more fragmented vendors.
Strategic Partnerships & Local Infrastructure
Beyond just supplying chips, the deal appears to extend deeper:
- OpenAI has expressed interest in building “Stargate Korea”, i.e. AI data center infrastructure in South Korea, in collaboration with Samsung, SK Hynix, and possibly SK Telecom.
- Samsung’s broader affiliates—such as Samsung SDS, Samsung Heavy Industries, and Samsung C&T—may partner in providing data center design, floating/offshore data centers, cooling solutions, and operations.
This isn’t just a chip supply contract; it’s a multi-layered collaboration across infrastructure, operations, and technology integration.
Why This Is a Big Deal
1. Memory Becomes a Strategic Bottleneck
In the AI era, memory is no longer a commodity — it’s strategic. High-bandwidth, low-latency memory is essential for feeding the compute (GPUs, AI accelerators) without bottlenecks. A shortage or disruption in memory supply can throttle the entire AI stack.
By securing memory supply from two of the world’s top providers, OpenAI is hedgeing against supply risk. This also places Samsung and SK Hynix squarely at the center of the AI supply chain.
2. Market Repercussions & Competitive Shifts
The announcement had immediate ripple effects in financial markets:
- SK Hynix’s shares jumped by as much as 12%, and Samsung’s by around 5%.
- Meanwhile, competitors in the memory space, such as Micron, saw declines (as investors anticipated a shift in chip procurement patterns)
This also underscores how the AI arms race is now reshaping semiconductor market dynamics.
3. Supply Volume & Industry Capacity Strain
To contextualize the scale:
- 900,000 wafers per month is enormous — it may consume a very significant share of global DRAM output. Some analysts suggest Stargate’s demand could reach up to 40% of global DRAM capacity under certain assumptions.
- Meeting that demand will require ramping up production, expanding fabs, and possibly pushing memory technology roadmaps forward faster than planned.
4. Geopolitics & National Strategy
This deal also highlights how semiconductors remain a geopolitical lever:
- South Korea is positioning itself as a global AI and semiconductor hub, elevating its strategic importance.
- The United States, through Stargate and other initiatives, is seeking to anchor AI infrastructure domestically and reduce dependencies.