Samsung and SK Hynix plan $590 billion chip investment as AI demand sends memory prices soaring

Why Samsung and SK Hynix Are Betting $590 Billion on Memory: AI’s Insatiable Hunger for Chips

In a move that reshapes the global tech landscape, Samsung and SK Hynix have announced plans to invest a combined $590 billion into chip manufacturing. The reason? AI demand has sent memory prices soaring, and both companies are racing to secure their place in the new intelligence-driven economy. This isn’t just another business deal—it’s a signal that the AI revolution is now being built on physical silicon, not just code.

What’s Driving the $590 Billion Bet?

Simply put: AI needs memory—lots of it. Every time a large language model is trained or a real-time inference is made, vast amounts of data must be stored and retrieved at lightning speed. The type of memory used in these tasks—High-Bandwidth Memory (HBM) and advanced DRAM—has become the bottleneck for AI performance. As companies around the world rush to deploy AI at scale, demand for these chips has exploded, and prices have followed.

Memory prices have been soaring for months. The pandemic-era chip glut is long gone, replaced by a shortage of the specialised chips that power AI accelerators. Samsung and SK Hynix, the two dominant players in memory, are responding by pouring unprecedented capital into expanding factories, developing next-generation processes, and locking in long-term contracts with major cloud providers and AI hardware manufacturers.

While the exact timeline of the $590 billion investment isn’t fully detailed, industry observers expect the bulk to be spent over the next five to ten years. This is a long-term play, one that acknowledges AI is not a passing trend but a permanent shift in computing.

The Memory-AI Connection: Why It Matters

Most people think of GPUs when they think of AI hardware. But without memory, even the most powerful GPU is useless. HBM stacks multiple DRAM chips vertically and connects them through a high-speed interface, giving AI processors the bandwidth they need to feed billions of parameters quickly. Samsung and SK Hynix are the world’s top producers of HBM, and their investment will directly increase the supply and lower the cost of these critical components.

This investment also goes beyond HBM. Traditional DRAM and NAND flash are used in data centres, edge devices, and smartphones that run AI models locally. As AI becomes embedded in every application—from search engines to medical imaging—the demand for all types of memory will keep rising.

For businesses, this means the cost of AI infrastructure is tied to memory chip supply. When memory prices are high, AI becomes more expensive to deploy. The new investment from South Korea’s top chipmakers aims to add enough capacity to eventually bring prices down, making AI more accessible to smaller companies and startups.

What This Means for the Future of AI

The $590 billion commitment signals that AI is moving from experimental projects into full-scale industrialisation. Here’s how this will shape the next decade:

1. Faster AI Training and Lower Costs

More memory production means more supply, which typically leads to lower prices per gigabyte. As HBM and DRAM costs fall, training large models becomes cheaper. Companies that previously couldn’t afford to train models containing hundreds of billions of parameters may now have the opportunity. This could accelerate innovation in fields like drug discovery, climate modelling, and autonomous systems.

2. Edge AI Becomes Mainstream

AI on your phone, laptop, or car needs memory that is both fast and power-efficient. Samsung and SK Hynix are investing in new memory technologies like LPDDR6 and custom HBM variants designed for edge devices. With cheaper, more capable memory, we’ll see more AI features running locally without needing to call the cloud. That means faster response times and better privacy.

3. Geopolitical Reinforcement for South Korea

South Korea already dominates memory chip production, and this investment cements its role as the manufacturing backbone of the AI world. As the United States and Europe push for chip independence, South Korea’s massive spending ensures it remains indispensable. For global AI supply chains, this means continued dependence on a small number of players—but with enough capacity to meet demand.

4. New Business Models Based on Memory as a Service

Some analysts predict that cloud providers will start offering memory-optimised instances specifically for AI workloads. The investment could enable memory to be sold in more flexible ways, such as “memory on demand” for training bursts. Businesses that plan their AI projects around memory availability will have a competitive edge.

Practical Implications for Businesses

If you’re running an AI-driven company or planning to adopt AI, here’s what you need to know:

Budget for memory volatility: Memory prices can swing wildly. The investment will add capacity, but it takes years to build fabs. In the short term, prices may stay high. Consider locking in long-term agreements with memory suppliers if possible.
Rethink your model architecture: Some model designs are memory-hungry while others are more efficient. As memory becomes a major cost factor, choose architectures that balance performance with memory usage. Techniques like model pruning and quantization will become more attractive.
Monitor South Korean chip makers: Earnings reports and capacity announcements from Samsung and SK Hynix are now leading indicators for AI infrastructure costs. Watch these closely when planning your own AI budget.
Plan for multitier memory: Future AI systems will use different memory types (HBM, DRAM, SSD) for different tasks. Understanding these tiers will help you optimise your own hardware decisions.

Societal Ripple Effects

This $590 billion investment is not just about technology; it’s about jobs, energy, and the shape of the future workforce. Building these fabs requires tens of thousands of skilled workers, from engineers to construction crews. The cities around Samsung and SK Hynix’s campuses will see economic booms. At the same time, memory fabs are energy-intensive—the power needed to make billions of chips is enormous. Sustainability will become a pressing issue for the industry, and both companies have begun investing in renewable energy and water recycling.

On the flip side, cheaper memory could democratise AI. Smaller companies in developing countries might finally be able to afford the hardware needed for custom AI models. This could reduce the current concentration of AI power in a few big tech firms. But it also means more AI applications in every corner of life, raising ethical questions about surveillance, bias, and job displacement.

What Could Go Wrong?

Massive investments carry risks. If AI demand cools unexpectedly—due to regulation, economic downturn, or a shift in technology—these memory fabs could end up overproducing, leading to a price crash and losses for the investors. Additionally, both Samsung and SK Hynix face competition from US memory maker Micron and from Chinese memory startups that are subsidised by Beijing. Trade tensions could disrupt supply chains or change the export rules for advanced memory.

There’s also the technological risk: if a new type of memory (like resistive RAM or storage-class memory) makes HBM obsolete, billions in investment could be stranded. However, given the long transition times in the chip industry, both companies are likely diversifying their technology portfolios to hedge against that possibility.

The Big Picture: Memory as the New Oil

Just as oil powered the industrial revolution and microchips powered the digital revolution, memory chips are now fueling the AI revolution. The $590 billion investment from Samsung and SK Hynix is a clear acknowledgment that the scarcest resource in AI today is not algorithms or data—it’s the physical capacity to store and move information at the speed of thought.

For years, the conversation about AI has focused on software innovations: better neural networks, bigger datasets, new training tricks. But beneath the hood, hardware constraints have always been the real gatekeepers. This investment signals that the hardware race is just as important as the software race. Countries and companies that control the memory supply will have outsized influence over how AI develops and who benefits from it.

Over the next decade, expect to see memory becoming a strategic asset akin to rare earth minerals. Governments will likely offer subsidies to secure domestic memory production, and international alliances may form around memory supply chains. The era of cheap, abundant memory is not guaranteed—it will be built through these massive capital outlays.

Conclusion: A Pivotal Moment for AI

The decision by Samsung and SK Hynix to commit $590 billion to chip manufacturing is one of the largest industrial investments in history. It sends an unmistakable message: AI is not a hype cycle, but a structural shift in how computing works. Memory, once seen as a commodity, is now a critical strategic domain.

For businesses, the takeaway is clear: plan for a world where memory affects your AI costs directly. For society, this investment promises cheaper and more powerful AI, but also raises questions about concentration of supply and environmental impact. And for the future of AI, it means the next generation of models will have the hardware backbone they need to become truly ubiquitous.

Whether you’re a data scientist, a CEO, or simply a user of AI, the chips being forged in South Korea today will shape the intelligence of tomorrow.

TLDR: Samsung and SK Hynix are investing $590 billion in memory chip production because AI demand has driven memory prices sky-high. This massive bet will increase the supply of high-bandwidth memory and DRAM, lowering costs for AI training and deployment over time. It strengthens South Korea’s grip on the chip supply chain, accelerates edge AI, and forces businesses to think strategically about memory as a key cost factor. The investment signals that AI’s hardware backbone is as important as its software, and those who control memory will shape the industry’s future.