$DRAM $MU $SKHY $NVDA More AI Factories. NAVER, NVIDIA, and Brookfield announced plans to expand Korea’s sovereign AI factory infrastructure at NAVER’s GAK Sejong data center. The initial multi-tenant NVIDIA DSX AI factory will scale from 55 MW to 200 MW by 2028, with NAVER targeting a path to 1 GW overall. NVIDIA plans a $1 billion strategic investment in NAVER (subject to conditions), while Brookfield has a nonbinding term sheet to fund up to $9 billion. The expanded capacity will support Korea- and U.S.-based AI innovators with production-scale compute using the full NVIDIA DSX platform

$DRAM $MU $SKHY $AVGO Samsung Electronics signed a strategic MOU with Broadcom at the AI Summit in San Francisco to jointly develop next-generation HBM-based AI accelerators. The partnership covers memory and foundry cooperation valued at more than $200 billion through 2030. Samsung will supply HBM (including custom HBM) for Broadcom’s AI accelerators and produce Broadcom chips on 2nm-and-below processes, while also providing advanced packaging and pursuing system-technology co-optimization (STCO).

"I am blown away by your system!" "...been playing around with the indicator and it's incredible." 3 Elite seats opened up today. We don't ask anyone for testimonials. They just keep coming. See $MU example chart below. 300+ charts daily. Try it free:

$SKHY $MU $DRAM 2 gigawatts. That is SK Telecom declaring itself a hyperscaler. You have no idea how bullish that is. Context on 2 GW: Enterprise data center: 10 to 50 MW Hyperscale campus: 100 to 500 MW Elite AI campuses (Microsoft, Google, Meta, Oracle, Stargate): 1 GW and up Top tier globally, on next gen hardware and this is just the beginning. SK Hyper targets 5 GW from 2029, scaling toward 15 GW by 2035. A telecom carrier just became an AI infrastructure player at hyperscaler scale. New bidder for GPUs, ASICs, and above all memory. Fucking smart move by SK Group $SKHY. Mega Bullish 🔥

$MU $SKHY $DRAM For its entire history, LPDDR was phone memory. Low power, built for smartphones and tablets, the stuff inside every iPhone. Servers never touched it. They ran on DDR. That changes next year. LPDDR is being pulled into the data center at scale through a new form factor called SOCAMM2, and it drops a wave of Apple-sized buyers into the exact memory pool Apple depends on. For the first time, Apple is competing for its own memory type against customers the suppliers like better. The leverage Apple spent two decades accumulating over the memory trio is eroding, not because Apple got weaker, but because higher-margin buyers showed up for the same wafers. It is not one buyer. It is the whole server industry converging on mobile memory. When it was just Nvidia, you could call it a one-platform quirk. It is not just Nvidia anymore. Nvidia is the anchor. Its Vera Rubin platform is built around SOCAMM2, co-designed with Micron, needing LPDDR at data-center scale. Meta is the hyperscaler making the case. Micron and Meta co-authored a 2026 white paper validating LPDDR5X for hyperscale deployment on Meta's own production workloads. Their conclusion, in their words: LPDDR5X is no longer just an "edge" technology, but a vital and viable standard for hyperscale GPC and AI platforms. The supplier and the buyer, publishing joint proof together. AMD is next in line. It confirmed LPDDR5X SOCAMM2 support on its EPYC "Verano" server CPU in 2027, positioned as the host CPU for future Instinct GPU generations in its AI rack-scale platforms. AMD's own engineers now call memory, not compute, the next data-center bottleneck. Qualcomm is exploring it too. AMD and Qualcomm are now exploring SOCAMM adoption alongside Nvidia for agentic AI inference. And JEDEC is standardizing the whole thing as JESD328, which is the signal that this is a permanent industry platform, not a proprietary experiment. When the standards body codifies it, every server maker gets to buy it, and every one of them competes for the same LPDDR Apple needs. Four of the most important compute buyers on earth, converging on the memory type that used to be Apple's alone. The memory market's reliance on Mobile is nearly halved in both DRAM and NAND by 2030. This is the structural change happening in the industry right now. Mobile used to be the biggest demand driver, and it is slated to become smaller and smaller. Why they all want LPDDR? The pull is structural, not fashion. Two reasons, both proven in the Micron/Meta data. Power. LPDDR5X consumes about one third the power of DDR5, landing under 7% of total system power, with up to 75% lower DRAM power than DDR5. At rack scale, where power and cooling are fixed constraints, that decides TCO. Capacity for inference. This is the agentic AI hook. SOCAMM2 suits inference workloads with large context windows and persistent KV caches, and Micron's 256GB module enables up to 2TB of LPDDR5X per CPU socket. In the Micron/Meta tests, doubling LPDDR capacity killed disk spill and delivered 2x to 3x throughput, with a 38x slowdown when memory ran short. Capacity is the binding constraint on token throughput, and LPDDR is now how the industry scales it. 2TB of Apple's memory type, per socket, per server CPU. A flagship phone carries 12 to 16GB. Why this specifically erodes Apple's leverage Apple's power came from three things: the biggest volume in mobile memory, the willingness to prepay and lock long agreements, and the prestige of the design-in. It was the buyer the trio built their LPDDR roadmap around. SOCAMM2 hands all three to the AI buyers, in the same memory. Nvidia, Meta, AMD, and Qualcomm bring enormous volume, roadmaps the suppliers now organize around, and design-in halos bigger than Apple's. And every AI bit carries far better margin per wafer than a phone bit. When buyers of equal scale want the same constrained wafers, the ones paying more set the terms. Those buyers are no longer Apple. Apple has already conceded it out loud. Tim Cook described Apple as being in a supply chase mode for memory, currently constrained, with prices set to rise significantly. Sanjay once said certain customers, meaning Apple, drove pricing to a third of where it was. That customer is now standing in line behind the server industry. And spare me the Chinese memory FUD. The memory pool is global and shared. Chinese CSPs buy from the US too, and if less supply is available, they buy more US memory, not less. Tighter global supply cuts the same way for everyone. There is no side door out of this for Apple. So Fuck you Apple. This is the structural shift you never saw coming and you are stuck with it.

$DAVE $200 → $458. Posted this name a few times last year, including a potential Cup & Handle. Really long consolidation but it finally broke out. High-margin fintech neobank using AI-driven underwriting. Blue candle? 👀 300+ charts daily. Try it free:

$MU $SKHY $NVDA $TSM Open models are bullish for chips, ask @JensenHuang AI models improve at software speed. Silicon doesn't. Models improve at software speed. Each generation smarter than the last, compounding on itself. In other cases, an AI can distill its successor in weeks. There's no factory, no yield curve, no physical limit slowing it down. Silicon lives in a different universe. Leading-edge computer chips require anywhere from 500 to over 2,000 individual process steps. Every new generation of GPU and HBM takes years to qualify and to scale into reliable mass production. You can't distill a chip. You can't train a fab into existence. So intelligence races ahead while the hardware meant to run it falls further behind. Open models widen it further. Frontier capability lands in everyone's hands overnight, and suddenly the whole world is bidding for compute that hasn't been fabbed yet. You've seen it with Kimi K3, stopping subscriptions. The gap between what AI can do and what silicon can support is the defining constraint of this cycle. That's where I'm looking. AI is not cycle. It's exponential.

$PYPL Holy Cow. The Blue Candle called it. 38%+ gains. I knew it was cheap. What I did not know was that Stripe and Advent would lob a $53B buyout offer to light the fuse. Sometimes the chart sees value before the news. 300+ charts daily. Try it free:

Holy Shit $NVDA and $SKHY Group on Friday unveiled a more than $500 billion AI initiative spanning large-scale AI data centers and next-generation memory “SK Telecom plans to build a 2-gigawatt AI data center powered by Nvidia's Vera Rubin chips and SK Hynix's HBM4 high-bandwidth memory, with the first facility due to come online in 2027, Nvidia added.” SK Group is the large South Korean conglomerate (chaebol) that owns both companies. SK Hynix is its semiconductor subsidiary focused on memory chips (especially high-bandwidth memory like HBM), while SK Telecom is its telecommunications arm that builds and operates infrastructure such as the planned AI data centers. AI is NOT cyclical

According to the Korea Customs Service on the 21st, semiconductor exports hit $22.113 billion, up 180.6% from a year earlier. They accounted for 40.3% of total exports, 1.8 times the 21.9% share a year ago. $SKHY $MU $TSM

Your candles go red on a -5% down day. Ours printed BLUE. Next day? $PLTR +8.3% $MU +11.7% That is what happens when your model learns from data instead of reacting to price. We are Data Scientists. 2 Elite seats left. 300+ charts daily. Try it free:

$MU $10M in gains. Absolute madness. A separate $1K challenge account turned into $2M. One of the founding Elite members just shared this with the community. Legendary. 300+ charts daily. Try it free:

"In 2027, driven by strong demand from AI servers, DRAM bit demand growth will continue to outpace supply, extending the seller’s market and high-price environment." -Trendforce Structural Shift $MU $SKHY $DRAM

$MU $SKHY Memory Innovation Is Just Getting Started. 3D DRAM Incoming. "SK hynix is reportedly stepping up development of 3D $DRAM, a next-generation memory technology for on-device AI. … the company has reportedly begun recruiting 3D-Stacked DRAM-on-Logic Design Engineers through its U.S. subsidiary in San Jose, California. SK hynix is also said to have partnered with a specific customer to develop applications for the technology.” “SK hynix President and Chief Development Officer (CDO) Hyun Ahn said AI development is increasingly constrained by memory bottlenecks, requiring new memory architectures beyond incremental improvements. He said the company is working with TSMC to apply its advanced logic process to HBM4 base dies, with plans to expand the technology to High Bandwidth Flash (HBF) and 3D-Stacked DRAM-on-Logic to further integrate memory and logic for higher performance and power efficiency.” It's not just SK Hynix. I've talked to my buddies in the industry. Micron and Samsung are actively R&Ding on 3D DRAM. This product is going to be perfect for AI Inference.

$MU (15m) Blue prints: $932 → $949. Recovering, but hurdles ahead (resistances). Just opened up 3 Elite tier seats today. 300+ charts daily. Try it free:

$MU $SKHY $DRAM When the GPU company sells you on memory, the memory is the product. "15% more compute, 50% more HBM4 memory capacity and memory bandwidth... more capacity for longer context" $AMD CEO Lisa The compute advantage is 15%. The memory advantage is 50%.

If $AMD wins, $TSM Advanced Foundry Process and Advanced Packaging win. Also a win for $DRAM as it packs 50% more memory than competitors. “AMD unveils its sixth-generation EPYC 9006 server CPU and Instinct MI455X GPU, expanding its AI portfolio across agentic AI, cloud computing, enterprise, and high-performance computing (HPC). … the move is expected to drive demand for TSMC’s advanced packaging technologies, including CoWoS-L and SoIC.” “AMD notes that the EPYC Venice processor has entered volume production on TSMC’s advanced 2nm process technology, while the MI455X is built using TSMC 2nm and 3nm FinFET technologies.” “The MI455X integrates 320 billion transistors using multiple advanced packaging technologies. Four Accelerator Complex Dies (XCDs) are hybrid-bonded onto each Fabric and Cache Die (FCD), while the two FCDs are interconnected with 12 HBM4 stacks, two I/O dies, and each other using TSMC’s CoWoS-L packaging technology.” “Some high-end EPYC Venice models are reportedly expected to adopt TSMC’s CoWoS-L packaging and be manufactured at TSMC’s Kaohsiung Fab 22, with production expected to ramp in the second half of 2026 to support growing demand for agentic AI servers.” “CoWoS-L capacity remains tight. … some Chip-on-Wafer (CoW) orders are expected to be outsourced to ASE, while TSMC and OSAT providers continue expanding CoWoS capacity with visibility extending through 2027 and 2028.” “AMD’s MI450, MI455X, and future AI chips are expected to boost SoIC (System-on-Integrated-Chips) demand. Institutional investors … estimate that TSMC’s SoIC capacity will reach 15,000–20,000 wafers per month by the end of 2026, before doubling again by the end of 2027.” “AMD’s Instinct MI455X adopts HBM4 memory, increasing capacity by 50% from 288GB of HBM3E on the MI350 series to 432GB. Memory bandwidth also rises to 22.3 TB/s, up 2.8× from 8 TB/s on its predecessor.” “The MI455X delivers up to 40 PFLOPs of FP4 and 20 PFLOPs of FP8 performance—roughly doubling the compute capability of the MI350 series. By comparison, Nvidia’s Rubin GPU provides 50 PFLOPs of FP4 and 17.5 PFLOPs of FP8 performance.” (Nvidia Rubin is noted with 288GB HBM4 and up to 22 TB/s bandwidth.)

KOSPI $SKHY -6.5% Honestly? I think it's just consolidating around the Volume Shelf (~1,856k). But if you'd been trimming on Pink & Red candles, you'd be in a far better position right now. Rule of Thumb 3: Scale In & Out. 300+ charts daily. Try it free: