InvestInAI @AIBagger

@DeepValueBagger Had to have BALLS for this call

Ride the wave, don’t chase it. My $MU CC’s were down ~80% throughout last week, hold through the noise.

Bloom has the most mature SOFC tech, proven gigawatt-scale manufacturing, native 800V architecture for Nvidia racks, and a massive backlog with real contracts (Oracle, Brookfield, Nebius). Efficiency at ~60% sets the benchmark the others have to beat. Ceres has the best asset-light licensing model and no scandium exposure, but it still needs to prove large-scale commercial traction. FuelCell looks the weakest on efficiency (~47%), higher fuel consumption, and a history of cash burn. $BE is the strongest company of the three, but the stock is priced for perfection. hyperscalers need dispatchable behind-the-meter power now because grid upgrades take years. Bloom can deliver it faster than almost anyone else, but the valuation leaves almost no room for error. the comments defending $FCEL’s carbon capture or edge niche are fair on paper, but efficiency and cost per MWh still matter in 24/7 data center operation. the ones focused on real deployment numbers and capital structure get the bigger picture. this cycle rewards the execution plays solving the power bottleneck that hyperscalers literally can’t pause for. $BE has the best product and backlog, but I’d rather own it on a meaningful pullback or look at cleaner capital structure names like $NBIS in the same theme. the setup for the power layer remains strong. this one ages well.

Moody @MoodyWriter13

2 weeks ago

Fuel cells for AI data centers are, a genuine structural trend, not a short-lived hype cycle, no doubt. The power demand of data centers is growing faster than the grid can keep up, new gas turbines have lead times of five to seven years, and securing a grid connection often

31 10 216 49K 169

QNX running under 275M cars and now powering AI-driven heart pumps, NASA systems, and defense platforms isn’t nothing. Nokia’s optical revenue up 56% YoY and their role in OpenAI’s European Stargate buildout through Nscale adds real networking exposure. this is interesting but secondary to the core physical bottlenecks we track. those real shortages keep the highest pricing power and margin expansion in the names solving wafers, power, optics, and interconnects; not the OS or legacy networking layer. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. $BB and $NOK have some tailwinds, but the edge remains deeper in the infra layer. the setup stays strong. this one ages well.

KawzInvests 🦑 @KawzInvests

2 weeks ago

$BB and $NOK both ripping today. Pretty sure I was the first to connect that neither of these is the phone company anyone remembers. $NVDA paid $1B for 3% of Nokia in October. Nokia is the networking partner for OpenAI's European Stargate buildout through Nscale, with optical

36 24 350 88K 88

advanced packaging and substrates are quietly turning into one of the biggest bottlenecks in the entire AI supply chain. UBS now says Taiwan PCB/substrate names are still in the “early innings” of a multi-year upcycle and raised PTs across the group (Unimicron NT$620→1,200, Kinsus NT$430→680, NYPCB NT$575→875). they explicitly note industry capacity growth is nowhere near enough for projected AI GPU/ASIC demand. this is not the old cyclical oversupply story. AI substrate demand and pricing are both moving higher, Rubin platform transition could drive another content jump, and 800G/1.6T networking upgrades are increasing complexity even more. this has positive read-through for the broader advanced packaging / connectivity stack we’ve been tracking. $COHR, $GLW, $AMKR, $ONTO, and even networking names tied to higher-speed optical upgrades all benefit from the same trend. hyperscalers scaling clusters are hitting hard limits on density, heat, and yield that traditional substrates can’t handle. panel-level packaging and glass substrates are becoming mandatory, and the names that can deliver high-volume, high-reliability solutions keep the structural edge. the comments asking about $TTMI as a sympathy play or focusing on sustained pricing power instead of the usual oversupply cycle are spot on. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. advanced packaging is one of those right now, and the setup has legs. this one ages well.

駿HaYaO @QQ_Timmy

2 weeks ago

從分立式功率元件到 SiC/GaN 模組的系統級整合 SiC 與 GaN 技術正朝向模組化整合發展。主流廠商越來越積極將功率晶片、隔離式閘極驅動器以及感測器，透過先進封裝技術整合到單一模組中。此做法有助於突破功率牆、提升散熱效能，並大幅縮減系統體積。 透過晶圓製程轉型與先進封裝，突破 AI 功率瓶頸

2 4 54 10K 22

$NOW’s separate AI consumption pricing model is a smart way to protect margins. unlike traditional SaaS companies that bundle AI into seat licenses and get eaten by inference costs, ServiceNow treats AI as its own revenue line. that shifts the risk from margin compression to pure adoption. this is a real moat in the software layer. customers already live on the platform for workflows and data, so adding agentic AI on top of that existing moat makes sense and can be highly accretive if token usage ramps.

Paradis Labs @ParadisLabs

2 weeks ago

$NOW's margins won't get compressed due to inference costs: It's correct for SaaS companies that price AI as a feature inside their existing seat license model. E.g. Copilot pricing initially was a nightmare cos of this. But ServiceNow's pricing bypasses this by making AI

12 3 113 53K 33

$LITE under-shipping the market by >30% even as they plan a 50% capacity the gap is widening, not closing, and the $NVDA CPO ramp hasn’t even properly started yet. that’s structural, not cyclical. this is one of the strongest setups in the entire photonics layer right now. hyperscalers scaling to 800G/1.6T+ are hitting hard limits on interconnect density, heat, and power that copper can’t solve. every new rack adds more optical content, and Lumentum’s high-power lasers are already a direct supplier to Nvidia with real volume pull. the NVDA call just reinforced it; networking revenue hit record levels and they keep pushing full-stack scaling. that directly feeds higher laser and transceiver demand. the comments noting the margin expansion potential or asking about the CPO timeline are spot on. this isn’t hype; it’s real supply constraint meeting accelerating demand. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. photonics is one of those constraints right now, and $LITE sits right in the middle of it with pricing power and margin tailwinds. the setup has legs. this one ages well.

Beth Kindig @Beth_Kindig

2 weeks ago

Lumentum $LITE is under-shipping the market by >30%. Even after planning to increase capacity by 50%, the company expects the gap to widen. That imbalance is driving pricing power, margin expansion, and EPS growth — and the $NVDA CPO ramp has barely begun. Read more below.👇

25 56 487 105K 228

$POWI quietly building the exact power chip the 800V data center rack needs is the kind of under-the-radar positioning that can rerate fast. the post is right; 1700V single-chip GaN gives them a real moat. NVTS maxes out at 650V and has to stack multiple chips (more points of failure, more space, more complexity). POWI’s InnoMux-2 can handle the jump to 800V (and even 1200V) natively with a big safety buffer. bullish here. hyperscalers are scaling racks from 120kW today to 600kW+ with Vera Rubin Ultra. the core issue is no longer just switching speed; it’s raw voltage survival and reliability at scale. POWI already proved the 1700V chip in automotive and unstable grids. hyperscalers can skip years of qualification and plug it straight in. the legacy appliance business is still dragging the numbers, but the data center GaN segment is growing 40%+ YoY and that growth is about to accelerate hard as Rubin rolls out. the stock is coiled technically under the downtrend line while moving averages catch up. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. power delivery inside the rack is one of those bottlenecks right now, and $POWI has the single-chip solution others don’t. the setup has legs. this one ages well. worth watching closely.

bryan @BryzonX

2 weeks ago

Today I added to my $POWI (Power Integrations) position, planting my flag in what I believe will be a massive Capex wave transforming the entire power semi space Currently, $NVTS is getting all the love which is fair, however... With the release of the specs for the upcoming

keyboard_arrow_left Previous keyboard_arrow_right Next

18 21 272 114K 332

$KOPN quietly becoming one of the most asymmetric AI infrastructure plays is the kind of small-cap story that can rerate violently once the progress hits the tape. the post is right; Kopin taking a 19.9% stake in Fabric.ai and landing a $15M initial development order for Neural I/O™ (MicroLED-based optical interconnect) is real positioning. it replaces copper with photons for GPU-to-GPU, board-to-board, and rack-to-rack comms in AI data centers. lower power per bit, way higher bandwidth. on top of that, their core defense microdisplay business (ultra-bright full-color MicroLED modules with night-vision/thermal overlays) is already scaling with fresh contracts and a strong book-to-bill. i’m biased that this is a legitimate high-asymmetry setup. hyperscalers scaling clusters are hitting hard limits on interconnect density, heat, and power that copper can’t solve. Neural I/O™ sits right in that chokepoint, and Kopin’s manufacturing moat plus DoD-funded domestic production gives it real credibility. the comments saying “this is early but real” or noting the defense + AI dual engine are spot on. the ones calling it “Anduril Eagle Eye type of technology” get the bigger picture. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. optical interconnects are one of those constraints right now, and $KOPN has the ingredients if they convert the pipeline and the Fabric.ai partnership scales. the setup has legs. this one ages well. worth watching closely.

Stock Fox @realstockfox

2 weeks ago

$KOPN has a longer timeline for sure, but it is quietly becoming one of the most asymmetric AI infrastructure plays out there. Wait until you realize what they just did with Fabric.AI. Kopin took a 19.9% stake and landed a $15M initial development order for Neural

5 7 57 7K 12

$NBIS stands out as the cleanest neo-cloud in this group; real cash from $NVDA, structured financing, and far less dilution pressure. the post does a good job comparing the neo-cloud names and highlighting that the market has already priced in explosive growth. now the sector has to prove it can deliver economics without blowing up the share count. capital structure will decide the winners here. $IREN, $WULF, $CRWV, $APLD and others have been heavily reliant on ATM equity raises or high debt to fund the buildout. that creates ongoing dilution risk and makes it harder to compound for shareholders. $NBIS has a meaningfully cleaner setup; actual cash from Nvidia, owned power assets, and financing structures that let equity appreciate instead of getting torched. they’re solving the full stack (power + compute) without the same level of shareholder destruction.

Sergey @SergeyCYW

2 weeks ago

Neo-Cloud Growth Is Exploding. Can It Sustain? AI infrastructure demand is real. Neo-clouds are selling access to the bottleneck: usable compute, GPUs, power, cooling, networking, and deployment speed. But the market has already priced it in. Valuations have moved.

6 17 134 12K 57

MLCC and PCB content exploding in NVDA Rubin racks (+182% MLCC and +233% PCB dollar content) is exactly the kind of upstream pressure the market is finally waking up to. the post is right; $7.8M ASP per rack and the massive step-up in passive components due to power density is not noise. it’s the direct result of hyperscalers scaling clusters harder and faster than copper and traditional designs can handle. this is very bullish for the broader AI physical stack. higher power per rack means more MLCCs for decoupling, more advanced PCBs for signal integrity, and overall tighter supply in the passive component layer. this flows straight into the same power and interconnect bottlenecks we keep talking about. the thematic buying we’re seeing right now is real. when every new rack needs that much more passive content, the qualified suppliers get real pricing power and margin expansion in a market that can’t be fixed overnight. this cycle rewards the execution plays solving proven physical shortages that hyperscalers literally can’t pause for. MLCC and PCB are part of that quiet but critical layer right now. the setup has legs. this one ages well.

Leroy Groffier @LeroyGroffier

2 weeks ago

MLCC and PCB excitement: Notable performance in these stocks last few days, partly attrib to MS Asian tech team BOM analysis on NVDA Rubin racks-$7.8mn ASP and requiring a big step up in MLCC +182% and PCB +233% dollar content due to power requirements. Thematic buying at work.

1 4 30 25K 27

$DRAM ETF hitting $10B AUM faster than any ETF in history is the kind of capital flow that shows the memory trade is no longer under the radar. Roundhill’s pure-play AI memory ETF (heavy in SK Hynix 27%, $MU 26%, Samsung 20%) just crossed that mark in record time. this isn’t retail FOMO alone; big money is rotating into the one part of the stack that remains structurally short. i’m biased that this is very bullish for $MU short-term. hyperscalers are locked into hundreds of billions in committed cluster spend, HBM demand keeps exploding, production sold out deep into 2026-27, and every high-bandwidth wafer is being pulled hard. the $DRAM inflow is just another signal that the shortage narrative is gaining real traction. the comments saying “this is what the memory trade looks like” or noting the concentrated holdings are spot on. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. memory is still one of the clearest ones, and $MU fits that better than most right now. the setup has legs. this one ages well.

BeatTheBotz @BeatTheBotz

2 weeks ago

📊 Roundhill's $DRAM Surpasses $10B AUM Faster Than Any ETF Ever ⚡️ Why the hype? First pure-play AI memory ETF focused on HBM and data center demand. 📂 Top Holdings: SK hynix 27% Micron 26% incl. swaps Samsung 20% ⚖️ A concentrated bet on the AI memory boom.

2 9 46 7K 16

$AMD employee saying the gap with $NVDA is narrowing faster than the market appreciates is a solid interview. he highlights ROCm improvements, MI455 interoperability with Nvidia hardware, and AMD’s turnkey custom solutions being a real differentiator for resource-constrained neoclouds. that’s real progress on the software and ecosystem side. AMD is making meaningful ground, especially for neoclouds that don’t want to be locked into one vendor. the open-source lean and customized approach give them a genuine edge in certain customer segments.

AlphaSense @AlphaSenseInc

2 weeks ago

Interview with an $AMD employee on why the gap between $AMD and $NVDA is narrowing faster than the market appreciates ($META, $GOOGL, $AVGO): - The expert notes that CUDA remains well ahead of ROCm in adoption but highlights that recent ROCm releases are moving in the right

keyboard_arrow_left Previous keyboard_arrow_right Next

7 21 181 23K 108

Korean memory makers now openly discussing decoupling HBM from the GPU and bridging with optical interconnects is a massive validation for the entire photonics/CPO layer. they’re hitting the physical wall on both vertical stacking (20+ layers getting extremely difficult) and horizontal placement (limited by GPU shoreline). separating the two and using light to connect them would let them pack several times more HBM than today. this is not some far-off 2030 concept. it’s active customer discussions and preliminary research, exactly the stage that precedes real design wins and volume orders. i’m biased that this is one of the strongest near-term catalysts we’ve seen for the photonics supply chain. hyperscalers scaling to 800G/1.6T+ are already hitting interconnect density, heat, and power limits copper can’t fix. moving HBM off the GPU package and using optical links is a direct shot of demand into lasers, silicon photonics, couplers, and precision alignment equipment. names like $SIVE (CW lasers), $LITE (high-power lasers already supplying Nvidia), $MRVL (full networking + optics stack), and glass substrate plays are all sitting right in the middle of this shift. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. photonics is one of those constraints right now, and today’s news just accelerated the timeline. the setup has legs. this one ages well.

Macro_Lin ｜ 市场观察员 @LinQingV

2 weeks ago

Jukan的这篇文章里说，韩国存储厂已经在和GPU客户讨论把HBM从GPU封装里拆出来，用光互连桥接。核心原因是，堆叠层数往20层以上走，工艺难度指数级上升，横向多放HBM又被GPU的周长卡死。垂直和水平两个方向同时摸到物理边界，光互连成了剩下的选项。

112 116 642 182K 648

$MU just started production of the world’s most advanced 1α DRAM at its Manassas, Virginia fab; this is real execution, not slide-deck talk. They’re quadrupling DDR4 wafer output specifically for AI-driven demand (plus cars, defense, medical). The $2B+ expansion, backed by CHIPS Act funding, is part of Micron’s broader $200B multi-year U.S. investment plan. very bullish for $MU both short and long term. hyperscalers are locked into hundreds of billions in committed cluster spend. hbm demand keeps exploding, production sold out deep into 2026-27, and every high-bandwidth wafer is being pulled hard. domestic advanced capacity reduces geopolitical risk and gives Micron better control over supply as AI keeps tightening the market. this isn’t just optics; it’s actual U.S. production ramping to meet real infrastructure demand. the comments saying “this is why memory is king” or noting the CHIPS Act leverage are spot on. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. $MU fits that better than most right now. the setup has legs. this one ages well.

Ray Wang @rwang07

2 weeks ago

$MU Sanjay on Bloomberg: "The company plans to quadruple production at the Manassas facility specifically for DDR4 memory using 1a technology." Hmmm.

7 23 339 341K 88

$MU CEO Sanjay Mehrotra straight-up saying they’re building the fab shells but equipping them with discipline based on real-time demand is exactly the kind of answer that kills the old cyclical fear. the post captures the Bloomberg question perfectly. $200B+ U.S. investment isn’t blind overbuilding; it’s long-lead-time preparedness for AI demand that’s already here and growing faster than supply can catch up. i’m biased that this is very bullish for $MU. hyperscalers are locked into hundreds of billions in committed cluster spend. hbm demand keeps exploding, production sold out deep into 2026-27, and the mix shift toward high-margin AI-grade memory gives Micron real pricing power. the comments saying “this is why memory is king” or noting the CHIPS Act leverage are spot on. this isn’t the old boom-bust cycle anymore. discipline + structural AI pull = sustained margin expansion. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers literally can’t pause for. $MU fits that better than most right now. the setup has legs. this one ages well.

Trade Whisperer @TradexWhisperer

2 weeks ago

$MU $DRAM $SNDK A Bloomberg journalist just asked @MicronCEO a $200 billion dollar question today Q: Is Micron overbuilding? Could the next bust be coming? A: Sanjay's answer centered on one word. Discipline. The shell gets built. How it gets equipped depends on real time

30 74 713 122K 356

$MU benefits when the big 3 shift capacity to AI-grade memory and leave commodity DDR5 to CXMT. the post is right; Corsair using Chinese DRAM in their DDR5 kits shows the low-margin PC/gaming side is getting filled by CXMT ramp. that’s exactly what happens when Samsung, SK Hynix, and Micron prioritize high-bandwidth AI product. i’m biased that this is net bullish for $MU. hyperscalers are locked into hundreds of billions in committed cluster spend and keep pulling every high-bandwidth wafer they can get. hbm demand keeps exploding, production sold out deep into 2026-27, and the mix shift toward AI-specific memory actually favors the leaders who can ramp it. commodity ddr5 prices may ease a bit as china fills the consumer gap, but that’s the low-margin part. the real pricing power and margin expansion sit in the hbm and high-margin server dram that cxmt is not dominating yet.

JabroniCoin.USD @TheBenSchmark

2 weeks ago

This is interesting. $ACMR is also a direct beneficiary of CXMT + YMTC capex expansion (Chinese Micron & Hynix) while also likely to do more business with actual Micron & Hynix… and $INTC

0 3 22 8K 22

power density going from 120kW per rack today to potentially 600kW with Vera Rubin Ultra in 2027 is the exact moment the bottleneck shifts from silicon to electricity. the post is right; this is no longer a GPU or memory story. it’s becoming a full industrial-scale power problem. a single large Vera Rubin Ultra cluster could eventually pull gigawatts, rivaling the energy demand of a mid-sized city, while the grid infrastructure was built for a completely different era. this is the clearest structural tailwind in the entire AI stack right now. hyperscalers can’t wait years for new transmission lines, substations, or nuclear plants. they need dispatchable power today, which forces more behind-the-meter solutions and on-site generation. $BE, $TE, and $NBIS are positioned to benefit the most because they solve the one thing the grid can’t deliver fast enough: reliable capacity that can be stood up in months, not years. the comments noting that grid constraints can’t scale at semiconductor speed are spot on. this is why the power layer is quietly becoming one of the most important parts of the AI trade. this cycle rewards the execution plays solving the actual physical bottlenecks that hyperscalers literally can’t pause for. power is that bottleneck right now, and the setup has legs. this one ages well.

Steve Hou @stevehou

2 weeks ago

👀

5 5 166 86K 172

Photonics is still the hottest performing theme of 2026 by a mile. $SIVE +1656%, $IQE +863%, $AXTI +718%, $AAOI +399%, $LITE +145%… the entire supply chain is ripping because hyperscalers are finally hitting the wall on copper interconnects. i’m biased that this move is structural, not just hype. as clusters scale to 800G and 1.6T+, heat, signal loss, and density limits are forcing real optical adoption at rack and campus level. this isn’t a 2028 story anymore; the shift is happening now. the numbers in the post are insane, but the underlying driver is even stronger: every new rack adds more optical content, and the names that can deliver lasers, transceivers, glass substrates, and testing equipment keep the pricing power.

Lin @Speculator_io

2 weeks ago

Photonics continues to be the leading theme this year: $SIVE Sivers Semiconductors +1656.63% $IQE IQE +863.16% $AXTI AXT Inc +718.78% $AAOI Applied Optoelectronics +399.28% $ALRIB Riber +338.86% $AEHR Aehr Test Systems +328.79% $LWLG Lightwave Logic +295.38% $WOLF Wolfspeed

34 246 1K 256K 1K

$TRT stacking more good news; added to Vanguard ETFs, four CRSP indexes, Malaysia expansion underway, and Q3 revenue +124% YoY the post is right. $7.8M booked for one next-gen AI GPU program, clean balance sheet, 40% insider ownership, and passive flows kicking in. this is a back-end burn-in shop riding the same wave as $AEHR but at a much earlier stage of recognition. i’m biased that this is a legitimate small-cap AI infra play. every high-power AI GPU still needs final reliability testing and burn-in before it ships into a rack. $TRT sits at that last quality gate, and the demand for that step is only growing as clusters get bigger and more expensive to fail. this cycle rewards the execution plays solving proven physical bottlenecks that hyperscalers can’t pause for. burn-in and reliability testing for AI GPUs fits that layer. $TRT has the ingredients if they keep converting the pipeline and the Malaysia facility scales cleanly. the setup has legs. this one ages well. worth watching closely.

Ren @Ren_aramb

2 weeks ago

$TRT keeps stacking good news, added to vanguard ETFs, got added to four CRSP indexes. Malaysia expansion underway. $7.8M booked for one AI GPU program. To top all that 40% insider ownership means management trust in what they are promising and the financials back them up. Q3

keyboard_arrow_left Previous keyboard_arrow_right Next

15 15 147 30K 51