AI Boom Fuels Chip Giants' $590B Expansion

Samsung and SK Hynix plan to spend $590 billion on new chip capacity. The move comes as AI data centers drive record demand for high-bandwidth memory.

The companies operate under direct support from the Korean government. Their joint plan includes four new fabrication plants, a dedicated packaging center, and fifteen years of additional research spending. Production lines will focus on high-bandwidth memory chips used inside AI accelerators.

Samsung and SK Hynix Detail $590 Billion Expansion

The investment breaks down into clear line items. Samsung and SK Hynix will spend 800 trillion Korean won on the four new fabrication plants. Another 81 trillion Korean won will fund a central packaging facility. Thirty trillion Korean won will support research into next-generation chips over the next fifteen years.

Both firms already hold nearly eighty percent of the global high-bandwidth memory market. New capacity will come online gradually, with only fifteen to twenty percent of added supply expected by 2028.

AI Data Centers Create Immediate Supply Pressure

AI training clusters require far more memory per server than traditional workloads. Each new AI accelerator now ships with multiple high-bandwidth memory stacks. That shift has produced sustained shortages since early 2025.

Jefferies analysts expect contract prices for memory chips to rise forty to fifty percent in the third quarter of 2026. A further thirty to forty percent increase is forecast for the fourth quarter. Prices are projected to keep climbing through 2027 before any meaningful relief appears.

Memory Price Increases Reach Consumer Devices

Higher memory costs have already moved into finished products. Apple raised prices on certain Mac and MacBook models in response to the memory squeeze. Other electronics makers are absorbing part of the increase but expect to pass more costs to buyers later this year.

Enterprise buyers face similar pressure. Cloud providers that rent GPU capacity have quietly raised rates on memory-intensive instances. Those increases affect companies that run large language models or real-time inference services.

Market Control Limits Near-Term Relief

Samsung and SK Hynix dominate production of the memory type that AI systems need most. New entrants lack the process technology and clean-room scale required to compete quickly. Industry observers therefore see limited ability for supply to catch demand before 2028.

The Korean government has framed the expansion as a national priority. Subsidies and permitting support aim to accelerate construction timelines. Even with that help, building and qualifying advanced fabrication plants takes years.

Risks Center on Execution and Demand Timing

Delays in plant construction or yield problems could extend the price surge. At the same time, any slowdown in AI capital spending would reduce memory orders. Analysts note that both companies have committed capital well ahead of confirmed customer contracts.

Some chip designers are testing alternative memory architectures that reduce reliance on high-bandwidth stacks. Success with those approaches would ease pressure on the current supply chain. Most observers still expect high-bandwidth memory to remain the dominant choice for large AI models through the end of the decade.

Watch Production Milestones and Contract Prices

Three signals will show whether the expansion is easing shortages. First, track quarterly contract prices reported by memory buyers. Second, watch for public announcements on when the new fabrication plants reach volume production. Third, monitor capital spending updates from major AI chip customers.

Continued price increases through 2027 would confirm that demand is still outpacing supply. Any earlier moderation would suggest the market is balancing faster than expected. Either outcome will shape hardware costs for AI developers and device makers alike.

remio helps teams capture and retrieve the internal context around hardware decisions. When memory prices shift or suppliers announce new capacity, the details stay searchable across meeting notes and vendor documents. This reduces repeated research when procurement plans change.