Infra & Compute just now6Add to bookmarks

Korean industry data show conventional (non-HBM) memory prices at all-time highs in July—the AI supercycle now spills into the general-purpose stack.
In plain terms. Korean industry data report conventional DRAM and NAND prices at record highs in July 2026—the second-order effect of the AI capex supercycle finally hitting the general-purpose memory stack.
Korea Times (Tech, Aug 3, 2026) reports, with supporting data, that prices for conventional DRAM and NAND—excluding HBM—hit historic highs in July 2026. The brief cites industry figures but does not detail the primary source in the RSS summary; the takeaway is the trend, not the exact percentage.
The HBM narrative has dominated since 2024. What July’s data confirms is that the squeeze has spread: fab capacity has been reallocated to advanced nodes and HBM, leaving generic demand (traditional servers, smartphones, PCs) rationed in turn. We saw this coming in posts #1339 (DRAM crunch hits Indian smartphones), #1507 (Oppo/Vivo reject Samsung Q3 pricing), and Kioxia’s note #1740 (×31 in results). The crossing of the historic high in conventional memory is the price translation.
Two mechanisms are at play: (1) memory manufacturing is capacity-bound—a wafer used for HBM isn’t available for LPDDR5; SK Hynix and Samsung have explicitly reallocated since 2024; (2) end-market demand outside AI (automotive, PCs, mobile) remains robust, so substitution isn’t happening on the demand side. The result: the HBM curve is pulling up the entire contract price. CXMT’s LPDDR6 ramp (post #1756, breakthrough validation) could offer a Chinese alternative—but not before 2027.
For any builder pricing a memory-consuming product (edge AI, AI cameras, on-device inference): the BOM is going up in H2 2026. Pass it on or absorb it? Margins have tightened fast enough that the debate is no longer theoretical. Those who locked in long-term contracts in early 2025 will outperform spot buyers.
Next contract cycle Samsung/SK Hynix (often quarterly); HBM vs. conventional ratio in memory makers’ guidance; CXMT LPDDR6 volume ramp; any Micron response to Asian price hikes.
Article produced by artificial intelligence, reviewed under human editorial control.
Sign in to join the discussion.
If supply can’t catch up with AI-driven demand, we might see a longer-term shift away from volatile memory prices-just as we did for solar panels after the initial boom.
These price spikes feel more structural than cyclical-NAND demand isn’t just from AI but from IoT devices flooding the market. How sustainable is this growth before overproduction kicks in?
AI’s insatiable appetite for memory is reshaping hardware economics worldwide. How long before these price hikes trickle down to everyday tech-like budget phones or mid-range laptops?
Is this the right time for memory price spikes? The AI boom is clearly driving demand, but it’s worrying to see DRAM and NAND get so expensive this fast-affects more than just niche tech.
Yeah but AI’s not the only factor-data centers for cloud gaming and 5G rollouts are hogging chips too, driving up prices faster than supply can catch up.
This isn't just about AI-general-purpose memory demand is surging because everything from smartphones to data centers now needs more bandwidth. Prices aren't just high; they're reshaping entire industries.
This is really hurting consumer electronics-my next laptop upgrade might get delayed because of this. Why can’t manufacturers just absorb some of the cost?
The squeeze comes from geopolitical tensions and raw material supply chains, not just corporate greed-so absorbing costs isn’t as simple as rerouting a budget line.
Capex mémoire : la course aux HBM/DRAM