Memory giants like SK Hynix, Micron, and Samsung are thriving thanks to AI's hunger for high-bandwidth memory. Yet, this boom might lead to bust cycles.
Right now, if you're in the memory business, you're probably smiling. AI's datacenter explosion has sent demand for high-bandwidth memory, DDR5, and NAND flash memory through the roof. SK Hynix and Micron have seen their revenues triple over the past year, while Samsung's have doubled. But history warns us: what's up must come down.
Demand and its Discontents #
The current AI-driven feast has gobbled up the available memory supply, driving prices skyward. Everything from budget smartphones to AI infrastructure feels the pinch. In Buenos Aires, AI tools aren't a luxury. They're tap into. Yet, for ordinary folks trying to buy a new phone, it means higher costs.
To tackle this, memory giants are investing massively in new capacity. South Korea's putting $576 billion into the pot, with SK Hynix and Samsung leading the charge. Meanwhile, Micron promises up to $3 billion to strengthen the U.S. semiconductor supply chain, and it's expanding in Singapore, Taiwan, and Japan. But building these fabs is no walk in the park.
The Long Road to More Memory #
Setting up a new semiconductor fab is a marathon, not a sprint. First, you need financing, location selection, and permits. Then, there's the construction of power and water systems, and finally, the installation of specialized equipment. This process takes years, even without a hitch. Any new capacity from SK, Samsung, or Micron won't become reality for at least three years.
For now, memory prices are stuck in the stratosphere. The IDC suggests we might not see relief until 2028. That's a boon for memory makers but a headache for AI startups and model developers facing high infrastructure costs.
The Future: Boom or Bust? #
AI's insatiable demand has flipped the memory market script. Prices that should've dropped in 2025 and 2026 have risen instead. But what if AI's demand doesn't meet expectations? Memory vendors could face a bust cycle like never before, a risk they've always had to play against.
As we look ahead, the real question is: Can memory vendors scale up before AI startups run out of their VC-backed runway? If they can't, memory prices will no longer be the only reason you can't afford a new laptop.
Get AI news in your inbox
Daily digest of what matters in AI.