SK Hynix just held a forum and dropped some real signals about where memory is headed. Three things stood out:
First, conventional DRAM margins have been recovering faster than HBM margins lately. That's creating short-term pressure on SK Hynix since they've gone all-in on HBM. But they're not flinching — they're staying focused on HBM and next-gen memory tech instead of chasing near-term profitability. The bet is on building memory solutions for future AI systems, not optimizing for today's spreadsheet.
Second, customer requests for 3D stacked DRAM have jumped significantly since the start of the year. This isn't just R&D curiosity anymore — it's turning into actual demand. Memory architecture is moving vertical, and it's happening faster than most people expected.
Third, HBM thermal management is hitting a wall. You can't solve heat problems at the package level alone anymore. The next phase requires system-level optimization across HBM, GPUs, interposers, substrates, power delivery, and cooling. That means advanced packaging, liquid cooling, backside power delivery, and glass substrates are all moving from nice-to-have to must-have.
This is the kind of shift that creates supply chain winners and losers. If you're tracking AI infrastructure plays, pay attention to who's solving the thermal and stacking problems — not just who's making the chips.
First, conventional DRAM margins have been recovering faster than HBM margins lately. That's creating short-term pressure on SK Hynix since they've gone all-in on HBM. But they're not flinching — they're staying focused on HBM and next-gen memory tech instead of chasing near-term profitability. The bet is on building memory solutions for future AI systems, not optimizing for today's spreadsheet.
Second, customer requests for 3D stacked DRAM have jumped significantly since the start of the year. This isn't just R&D curiosity anymore — it's turning into actual demand. Memory architecture is moving vertical, and it's happening faster than most people expected.
Third, HBM thermal management is hitting a wall. You can't solve heat problems at the package level alone anymore. The next phase requires system-level optimization across HBM, GPUs, interposers, substrates, power delivery, and cooling. That means advanced packaging, liquid cooling, backside power delivery, and glass substrates are all moving from nice-to-have to must-have.
This is the kind of shift that creates supply chain winners and losers. If you're tracking AI infrastructure plays, pay attention to who's solving the thermal and stacking problems — not just who's making the chips.