Samsung & SK Hynix Bet $590B on AI Memory Demand

South Korea's two largest chipmakers, Samsung and SK Hynix, announced a combined $590 billion investment in new chip manufacturing capacity on June 29, backed by the South Korean government under President Lee Jae Myung. The plan includes four new factories in the country's southwest valued at 800 trillion won, a new advanced packaging center for 81 trillion won, and 30 trillion won earmarked for next-generation chip research over fifteen years.

The investment is a direct bet on AI. Samsung and SK Hynix together supply roughly 80 percent of the world's high-bandwidth memory — the specialized chips that stack inside every AI server alongside its processors. Each H100 GPU and each custom AI accelerator requires stacks of this memory to function. When AI compute demand expands, high-bandwidth memory demand expands with it, and the current AI buildout has pushed that demand well past what existing supply can satisfy.

What analysts project for memory prices

Jefferies equity research projects memory prices will rise 40 to 50 percent in the third quarter of 2026, with a further 30 to 40 percent increase in the fourth quarter and another 40 to 45 percent through 2027. That is not a misprint — it is three consecutive years of price increases, driven by AI demand outstripping supply. The relief from this wave of factory construction is expected around 2028, when approximately 15 to 20 percent of the new production capacity comes online. For at least two more years, the cost of any product containing memory will reflect this pressure.

Consumers are already experiencing the early effects. Apple raised the price of the 16-inch MacBook Pro by $300, bumped the 11-inch iPad Air from $599 to $749, and added $30 to the HomePod Mini. The company cited component costs. The memory inside an iPhone is not the same high-bandwidth chip that powers a server GPU — but the manufacturing capacity for both types is constrained by the same underlying bottlenecks. Resources committed to building high-bandwidth memory for AI servers are not available for consumer memory, and the buildout is absorbing an enormous and growing share of the industry's expansion plans.

Why this is a structural commitment, not a cyclical bet

The scale of this commitment is worth pausing on. $590 billion is larger than the annual GDP of most countries. A fifteen-year research program funded at 30 trillion won is not a hedge — it is an expression of conviction that AI-driven demand for memory will remain at this level and grow for the foreseeable future. Samsung and SK Hynix already dominate the market with roughly 80 percent combined share; an investment of this magnitude is designed to ensure no competitor can close that gap in a generation.

The announcement also carries a broader signal. Companies committing hundreds of billions of dollars to chip factories that will take years to come online are expressing a belief that the demand will still be there when those factories open. This is the AI infrastructure bet expressed not as a quarterly earnings call narrative but as a fifteen-year capital program. It is the clearest measure yet of how the industry views the permanence of the AI buildout.