$Sandisk(SNDK.US) and HBF thoughts:
An analyst question from the Sandisk earnings call: "High bandwidth flash, I'm wondering if you're seeing any price premium on that...versus conventional NAND."Sandisk's response: "It's a little early to talk about pricing on some of those nodes just yet."This answer has left me scratching my head...HBF is Sandisk's near-term growth vector for entry into AI inference workloads. Since HBF will command higher ASPs than standard NAND flash, Sandisk's HBF economics need to be forecasted to gauge what this means for future earnings.In turn, HBF pricing needs to be compared to HBM since every HBF order is an allocation decision made against HBM — same construction, same adv packaging capacity, same customer, same BOM line, same qualification cycle etc etc.However, the part that left me confused is this.It's usually the case in most industries that when you get these types of non-answers regarding new product pricing — it's a signal of a premium pricing model vs. incumbent companies and "competing" products. There are many examples you can Google separately.So in simple terms, is this a signal that Sandisk's HBF is to be priced ~similarly to HBM?This wouldn't make much sense since it's widely assumed that HBF is worse than HBM, by design. For example, HBF is lower latency and has slow and finite write endurance vs. unlimited for HBM.Therefore, should HBF not be priced relatively lower than HBM to account for these shortcomings? Ultimately, if Sandisk's HBF does get priced highly, what's to say that Sandisk's demand would remain sticky longer term?These are just musings, but considering that the SK Hynix and Sandisk HBF specs were released earlier this week, I hope that Sandisk provides more clarity/clues in their Investor Day next week.
Share your thoughtsWhat's on your mind?
0 / 2,000
- G
Samsung reported an explosive 250 fold increase on 2026Q2 operating profit. Revenue jumps 130% YOY and increases 28% quarterly to US$119 billion. HBM delivers the strongest sales by amounting to >60% of the revenue. The stocks rise 4% after this report but quickly drop to closed at 1% lower. It seems that this wonderful report still fails to stop the rout in the Korean stock market. Analysts expect this to continue until most of the investors who use leverage to invest are being flushed out. Is Korean memory stock uninvest able or investor just waiting to do bottom fishing?$SK Hynix(SKHY.US)
Long image
Trade Showcase: Trade, Show & Earn Rewards! - E
Needham: WFE & Memory
WFE Requirements for Vera Rubin PODs> 1 Million GPUs Impact: Building out greenfield fab capacity to support 1 million Rubin GPUs (roughly 5 GW of data center capacity) drives $14 billion in WFE, split roughly evenly between DRAM/HBM and logic.> WFE Intensity Floor: The total WFE intensity for a Vera Rubin POD sits around 9%, which is expected to act as the intensity floor moving forward.> Growth in Spend: Compared to an individual Rubin GPU package, a full Vera Rubin POD sees a 225% total WFE increase, driven heavily by surges in advanced logic (up 256%) and DRAM/HBM (up 164%).Silicon Content & Ratio Shifts> Total Die Count: Nearly 20,000 Nvidia dies are packed into a single Vera Rubin POD.> Wafer Area Shift: The rise of low-latency inference and agentic AI has shifted the DRAM-to-logic ratio (wafer area) from 7:1 down to 4:1.> Supply Dynamics: While traditional DRAM remains in short supply, the market is seeing advanced logic supply tighten correspondingly.> POD Subtotals: Across its 40 total racks (spanning NVL72, Groq 3 LPX, Vera CPU, Bluefield-4 STX, and Spectrum-6 SPX), a single Rubin POD requires roughly 1,384 total 300mm wafers, dominated heavily by DRAM (958.6 wafers) and advanced logic (248.9 wafers).$NVIDIA(NVDA.US) $Micron Tech(MU.US) $DRAM $EWY - E
Needham: "DRAM/HBM will remain a real bottleneck for AI."
> DRAM/HBM Bottleneck: DRAM and High Bandwidth Memory (HBM) are projected to remain significant bottlenecks for AI hardware scaling.> Rapid Bit Growth: Total DRAM bit growth (millions of GB) is expected to expand dramatically through 2028, driven heavily by increasing HBM adoption.> Surging HBM Share: HBM bits as a percentage of total DRAM are forecasted to grow steeply, scaling past 15% by 2028.> Wafer Capacity Shift: DRAM wafer capacity dedicated to HBM (measured in kWSPM) is projected to accelerate sharply, with expectations that nearly 50% of total DRAM wafers will go into HBM by 2028.$Micron Tech(MU.US) $DRAM $EWY - O
SK Hynix pricing this Thursday, listing Friday. i hold the 2x SK Hynix ETF as my most direct play on the 7/10 listing catalyst. small position because leverage into an event is a coin flip, but the setup is too clean to sit out entirely. Samsung's $58B print today makes me feel better about the HBM demand thesis backing it.
- O
SK Hynix doing the biggest foreign IPO in US history right when memory pulled back tells you everything. HBM king raising $29B to build more capacity is not what a glut looks like. i hold a small slice of the 2x SK Hynix ETF into the July 10 debut, keeping it tiny because leverage into an event is a coin flip. too clean a catalyst to sit out fully though.
- M
Another major factor is strong structural growth in demand, and this one is clearly satisfied.
The most straightforward reason is Nvidia’s need to keep upgrading its “token factory” hardware to deliver higher token throughput. This drives extremely fast generational upgrades in both HBM bandwidth and HBM capacity (size), leading to exponential demand growth.
As concluded in the earlier AI semiconductor analysis, token throughput equals HBM size multiplied by HBM bandwidth, and this roughly doubles with each new generation. On average, HBM size per GPU is growing more than 40% every year.
This demand curve is rising so steeply that it is very difficult for the supply side to keep up. Even with 14% growth in wafer production and 9% improvement in memory density, the overall $Roundhill T-Rex 2X Long DRAM Dly TrgtETF(RAM.US)DRAM supply growth simply cannot match the pace of AI-driven HBM demand.
In the hardware architecture, the attention stage’s KV cache requires both extremely high bandwidth and very large memory capacity. This gives HBM a unique and irreplaceable position. Even if HBM prices rise three to five times, the extra token throughput gained by investing in more HBM is still far more cost-effective than spending the money anywhere else.
Other memory technologies, such as SRAM, HBF, CXL, and PIM, are currently unable to compete directly with HBM in the critical KV cache and attention workloads. There is no viable replacement in sight for at least the next five years, and possibly much longer.
@Bridge Buzz SG
Trade Showcase: Trade, Show & Earn Rewards! - L
A trillion dollar market cap on $Micron Tech(MU.US) still feels surreal to me. I get the supercycle but I trimmed a third into the run, NAND and HBM cycles always turn eventually. Anyone here actually adding at these levels or are we all just holding and praying 🙏
- H
$Micron Tech(MU.US) | Micron Technology - Fiscal Q3 2026 Post-Earnings Call Summary
𝐂𝐅𝐎 (Mark Murphy)➤ "We're really pleased with the financial trajectory of the business."➤ "The last two quarters we've generated as much cash flow as any period in the company's history."➤ "We expect that cash flow growth will increase in the fourth quarter."➤ "We feel good about the durability of the performance of the business given the secular growth demand drivers."➤ "We intend to increase our capital return beginning after the second anniversary of our CHIPS agreement."➤ "We expect market tightness to continue beyond 2027."➤ "The HBM TAM now easily exceeds $100 billion in 2027."➤ "Startup costs will begin to increase meaningfully in the fourth quarter and continue through the first half of next year."➤ "FY2027 CapEx will increase substantially, with more than half directed toward construction."𝐂𝐡𝐢𝐞𝐟 𝐁𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐎𝐟𝐟𝐢𝐜𝐞𝐫 (Sumit Sadana)➤ "Demand for HBM products continues to be well above our ability to supply."➤ "Even beyond 2027 into 2028, customer demand is far in excess of what we can support."➤ "The same supply-demand imbalance is also true for non-HBM DRAM."➤ "Shipment growth is increasingly determined by supply rather than demand."➤ "Strategic Customer Agreements are designed as take-or-pay agreements."➤ "Customers cannot walk away from these agreements."➤ "The current agreements include over $22 billion in customer cash and financial commitments."➤ "We expect LPDRAM to continue increasing as a percentage of data-center memory."➤ "Both DRAM and NAND remain highly supply constrained."➤ "The combination of AI demand, structural supply challenges and Strategic Customer Agreements is completely transformative for our business." - N
Trump just posted on X that he has reached an agreement with Iran to "highest level Nuclear inspections long into the future" and that Strait of Hormuz is now open without Naval blockade, with about 19 million barrels of oil flowing out yesterday. I don't know how much truth there is in his words, but it seems that he is trying to calm the stock market. I personally is no longer as emotionally affected by his words anymore as we all know from time to time again that what was agreed upon may change within days, if not hours.
☕️ [Task Coins Giveaway] Daily Market Talk — Micron Tops $1,200 Before EarningsIt's all about AI today. Micron ripped nearly 7% to a record above $1,200 on a long-term memory deal with Anthropic, two days before its earnings. SpaceX went the other way, cracking 16% back below it...
