Key Takeaways
- Samsung and SK Hynix will invest $590 billion in chip production, including four new factories in South Korea
- Memory prices are projected to rise 40-50% in Q3 2026 and another 30-40% in Q4, with relief not expected until 2028
- The two companies control nearly 80% of the global high-bandwidth memory market that AI workloads depend on
Samsung and SK Hynix, backed by the South Korean government, will invest $590 billion to expand chip production as AI-driven demand sends memory prices into a steep climb. The scale of this commitment, roughly 911 trillion won, marks one of the largest coordinated semiconductor investments in history. And it still may not be enough to meet demand.
The bulk of the money, 800 trillion won, goes toward four new fabrication plants in southwestern South Korea. Another 81 trillion won will fund a chip packaging center, a critical bottleneck for high-bandwidth memory production. The remaining 30 trillion won spreads over 15 years for next-generation chip research.
How high will memory prices go?
The short answer: much higher, and soon. Jefferies Equity Research projects memory prices will jump 40 to 50 percent in Q3 2026, followed by another 30 to 40 percent increase in Q4. For 2027, analysts expect a further 40 to 45 percent climb. The investment surge from Samsung and SK Hynix won't provide relief until 2028, when 15 to 20 percent of new capacity finally comes online.
These numbers matter for anyone building AI products. High-bandwidth memory is the kind AI workloads depend on, and Samsung and SK Hynix together control close to 80 percent of the global HBM market. When these two companies can't keep up with demand, there's nowhere else to turn.
Why the investment is happening now
President Lee Jae Myung's administration is pushing regional economic development, and semiconductor manufacturing fits that agenda. But the timing is driven by hyperscaler demand. AI data centers are consuming memory at rates that have outpaced every industry forecast from 2024.
South Korea already dominates memory production. Samsung and SK Hynix collectively hold roughly 70 percent of the global DRAM market and over 50 percent of NAND flash. SK Hynix currently leads in HBM specifically, the specialized memory stacked in chips like Nvidia's H100 and H200. This investment aims to lock in that dominance for the next decade.
The downstream impact on consumer electronics
Rising memory prices don't stay contained to data centers. Apple has already hiked prices on Macs and MacBooks, citing component costs. Expect similar moves across smartphones, laptops, and any device that depends on flash storage or DRAM.
For AI builders, the calculus is straightforward but painful. Training runs will cost more. Inference hardware will cost more. Cloud compute pricing from AWS, Google Cloud, and Azure will reflect these input costs within quarters.
What this means for AI product roadmaps
If you're planning AI features that require significant GPU memory, the window for locking in pricing is closing. Teams building on-premise inference setups face a choice: buy hardware now at current prices or wait and pay 40 percent more by Q4. Neither option is comfortable.
The packaging center investment deserves attention. HBM manufacturing isn't limited by wafer production alone. Stacking memory dies and connecting them to logic chips is the actual constraint. The 81 trillion won packaging facility addresses this bottleneck directly.
Logicity's Take
This investment confirms what AI teams have suspected: the compute shortage is becoming a memory shortage. For product teams, the practical response is to optimize for memory efficiency now. Techniques like quantization, pruning, and efficient attention mechanisms aren't just nice-to-haves. They're becoming essential for controlling inference costs. Companies with memory-efficient architectures will have a structural advantage through 2028. The 80% market concentration also means any production disruption at Samsung or SK Hynix, whether from natural disaster, geopolitical tension, or equipment failure, would ripple through the entire AI industry within weeks.
Frequently Asked Questions
When will new chip capacity come online?
Jefferies Equity Research estimates 15 to 20 percent of new capacity will arrive in 2028. Until then, demand will continue outpacing supply.
How much will memory prices increase in 2026?
Analysts project 40 to 50 percent increases in Q3 2026, followed by another 30 to 40 percent in Q4. Additional increases of 40 to 45 percent are expected in 2027.
Why is high-bandwidth memory so important for AI?
HBM provides the memory bandwidth that large AI models require. It's stacked vertically and connected directly to GPUs, allowing faster data transfer than standard DRAM.
What share of the HBM market do Samsung and SK Hynix control?
Together, the two South Korean companies control close to 80 percent of the global high-bandwidth memory market.
Another look at how AI tools are reshaping tech infrastructure decisions
Need Help Implementing This?
Building AI products and need help navigating hardware constraints or optimizing for memory efficiency? Reach out to the Logicity team for introductions to infrastructure specialists and cloud architecture consultants.
Source: The Decoder / Matthias Bastian
Huma Shazia
Senior AI & Tech Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
