# HBM demand — X 热门讨论 (2026-10-08 13:00 UTC)
## @ProblemSniper (ProblemSniper) · 10-08 01:42 · ♥62 ↻1 💬5 Samsung just printed a number nobody's ever seen. 🇰🇷
Preliminary Q3 operating profit: ₩107.4 TRILLION (~$80.2B).
First quarter above ₩100T in the company's history. Up +782% from a year ago — nearly 9x. And it beat the ₩106.1T estimate. Revenue was ₩195T, up +127% YoY — huge, but actually under forecasts. The story isn't the top line. It's the margin.
The engine is memory. AI demand for HBM with tight DRAM/NAND supply is keeping chip prices elevated, and that gap isn't closing before 2027. Fourth straight record quarter. Q2 was ₩89.5T. This is the AI memory supercycle showing up in one print.
Preliminary numbers — full results later this month. What does this tell you about $MU and the U.S. memory names? https://x.com/ProblemSniper/status/2108010125495566458
## @HunterAllen4 (THE GAP FATHER) · 10-08 05:37 · ♥45 ↻1 💬14 $WDC
DON’T BLINK.
$550 comes quick. I will Quadruple down. I have shares.
WDC is sitting around $405 after falling nearly 49% from the June 18 intraday high of $799.87. That is a massive drawdown, but the business underneath it has not collapsed.
FY2026 revenue reached $12.9B, +36% YoY, Q4 revenue was $3.75B, +44%, EPS came in at $3.56, and Q1 FY2027 is guided to roughly $4.1B revenue, 55–56% gross margin and $4.00 EPS.
The market is repricing the shortage premium after the Toshiba capacity headline and TDK rumor not suddenly discovering that HDD demand disappeared.
The 44TB HAMR roadmap is a major piece of the thesis, but I want to be clear about what it actually means. Western Digital is targeting volume shipments of its 44TB HAMR platform in 1H 2027, using an 11-platter architecture at roughly 4TB per platter.
This is about cost per TB, rack density and watts per TB, allowing hyperscalers to store more data without proportionally increasing physical infrastructure.
WDC is already shipping 40TB ePMR UltraSMR, targeting more than half of nearline exabytes from that platform by Q3 2027, with a roadmap toward 60TB in 2028 and eventually 100TB-class drives.
And this is where the bigger demand story gets interesting. WDC doesn’t need 44TB to magically create new customers. Long-term agreements already extend into 2028–2029, with negotiations underway for 2029–2031.
Unconstrained demand is still expected to grow 25%+ in exabytes, while blended price per TB was up high-teens YoY. Seagate may be 12–18 months ahead on HAMR commercialization, but WDC is closing that gap while the industry is still dealing with a structural storage shortage.
The AI angle is also bigger than just training clusters. Roughly 80% of hyperscale data still sits on HDDs, and WDC gets roughly 88–89% of revenue from cloud.
Training datasets, checkpoints, logs, video, agent activity, inference data and enterprise archives all create enormous amounts of information that eventually has to live somewhere. If AI keeps accelerating data creation, HDDs remain one of the cheapest ways to store that explosion at scale.
I would not build the thesis around recursive self-improvement arriving by year-end. That’s speculation, not something included in WDC’s FY2027 guidance.
And even if AI systems suddenly began improving their own research loops, the first bottlenecks would likely be accelerators, HBM, networking and power.
WDC sits one layer underneath that as the storage backbone. RSI would potentially accelerate the amount of data being generated and retained it isn’t required for the existing HDD shortage thesis to work.
So I’m watching $395 closely. Hold that area and the post-Toshiba scare starts looking much more like a reset than a thesis break.
The real catalyst is the upcoming earnings report: $4.1B revenue, 55%+ margins, price/TB, exabyte shipments and commentary around the LTAs.
If those remain intact, I think the market can quickly start looking back toward the mid-$500s, with the Street’s higher targets showing just how much upside exists if the storage cycle stays tight.
The 50% drawdown is real. The 44TB catalyst is real. The demand is real.
The question is whether the market just gave us another chance to RIP THAT SUCKER.
RIP SALAD PENDING. 🥗
Don’t blink. https://x.com/HunterAllen4/status/2108069126128427430
## @AventusNetwork (Aventus DAO) · 10-06 15:49 · ♥31 ↻5 💬6 AI Made Memory Impossible to Ignore. The Onchain Economy Will Do the Same for Compute. https://x.com/AventusNetwork/status/2107498556525261134
## @SemiconductorsX (Semiconductor Insider) · 10-08 04:16 · ♥31 ↻3 💬4 Goldman is putting Samsung HBM revenue at $74 billion next year.
Late-September note: it sees HBM sales rising from about $20 billion in 2026 to $74 billion in 2027, up 274%. The model uses a roughly 130% ASP increase, to about $3.6 per Gb, and shipment growth above 60%. HBM’s share of its DRAM revenue goes from 8% to 19%.
The same model has 2028 HBM revenue clearing $100 billion. HBM4 volumes scale with NVIDIA and ASIC demand. Wafers move off conventional DRAM to feed it.
That shift is the second effect. Less conventional DRAM out of the same lines is what keeps the rest of the memory price up. $DRAM If $3.6 per Gb is a 130% ASP lift and shipments are only up 60%, is the 2027 number a volume ramp or a price the customer has already agreed to pay? > 引用 @MojoTricks: Samsung HBM: $20bn → $74bn in one year. +274% in 2027, per Goldman Sachs. Mix goes from 8% to 19% of DRAM. 2028E clears $100bn. https://t.co/Cb2lpoZy6u https://x.com/SemiconductorsX/status/2108048840632996238