Schiele

Do we call this strength - $SPIR?

The longer you delay building the life you actually want, the more likely you are to normalize a weaker substitute. Drift hardens into identity faster than people realize.

NYTT CASE 🧵 $SHT Smart High Tech 1) AI-case. Jag har följt detta bolag under 2 års tid men avvaktat med att investera tills nu. Bolaget tillverkar kylningskomponenter till högpresterande elektronik såsom GPU, CPU och AI-chipp. Otroligt relevant i vår tid med ai.

$NBIS is up 17% so far today I wouldn’t be surprised to see it hit $100 in the coming months

A striking sentence I read today: We don’t think power is the constraint for TSMC’s chip demand, at least for 2027, but ABF substrate and HBM supply are major gating factors. -Morgan Stanley

Daily reminder :

PHOTONICS / OPTICAL STACK — THE FULL VALUE CHAIN “The layer below the chips” — AI’s real bottleneck Layer 1 — Materials Foundation of everything (InP, fiber, wafers) $AXTI $GLW $COHR Layer 2 — SiPh Foundries Where photonic chips are fabricated $TSMC $GFS $TSEM Layer 3 — Lasers (Bottleneck) EML + CW lasers — tightest supply in the chain $LITE $COHR $AAOI $FN Layer 4 — PICs (Optical Chips) Integration layer replacing discrete components $POET $LWLG $MRVL $AVGO Layer 5 — Transceivers 400G → 800G → 1.6T data movement $CIEN $AAOI $LITE $COHR Layer 6 — Co-Packaged Optics (CPO) Next-gen architecture — optics + silicon combined $NVDA $AVGO $MRVL $COHR $LITE $ALAB Layer 7 — Optical Switching (OCS) Routing light without electrical conversion $CIEN $COHR $LITE Layer 8 — Networking Systems DCI, long-haul, metro infrastructure $CIEN $ANET $NOK Layer 9 — Test & EDA Hidden backbone enabling scale $KEYS $FORM $ANSS Layer 10 — Demand Drivers Hyperscalers pulling the entire chain $NVDA $META $GOOGL $MSFT $AMZN THE BIG PICTURE (2026–2027) - AI demand is exponential - Electrical interconnects are hitting limits - Optical = the only scalable solution Key Constraint: Laser supply (InP → EML/CW) → Bottleneck today = pricing power tomorrow This isn’t a single stock trade. It’s a full-stack infrastructure shift — like semis in the early 2000s. The winners won’t just be chipmakers… They’ll be the ones enabling light-speed data movement. Not financial advice. DYOR. Bookmark this, retweet for others

VICE PRESIDENT VANCE: "We have tried for 21 hours, but we are returning to the United States without an agreement. The Iranians would not accept our best offer."

Unimicron, Ibiden, AT&S, Kinsus, Nan Ya, Hynix, Micron, Samsung My favorites are Ibiden, Unimicron, Hynix & Samsung

$MU Signs of structural memory transformation during earnings today. Must Read 👇 Automotive: "The average car today has less than L2 ADAS capability, containing approximately 16GB of DRAM, while vehicles with L4 autonomy require over 300GB." Robots: "Humanoid robots will be AI-enabled and powered by a compute platform that rivals a high-end L4-capable automobile, requiring significant memory and storage. We expect this exciting new category to further underpin the long-term favorable dynamics shaping our industry." PC: "PCs with on-device agentic AI have recommended memory specs of at least 32GB, twice the average PC." Workstations: "The fast-growing category of personal AI workstations, such as NVIDIA DGX Spark and AMD Ryzen AI Halo, come in 128GB configurations, ideal for running large language models on device." Mobile: "The mix of flagship smartphones shipping with 12GB or more of DRAM increased to nearly 80% in Q4, up from under 20% a year ago." Data Center LP DRAM: "We sampled the industry's first 256GB LP SOCAMM2 product, built on our 1γ node, enabling 2TB of capacity per CPU. Quadrupling content from just a year ago." Data Center Demand: "AI server demand (vs. traditional server) is driving DRAM and NAND data center bit TAM to exceed 50% of industry TAM for the first time in calendar 2026." The AI Shift is Here

$AEHR Strong earnings report. Anyone that looks at current revenue is an idiot and doesn't understand the story. - $37M quarterly bookings and a 3.5x book to bill - Backlog guidance is at high end of the range. - Received follow-on orders from their leading SiPho customer. - Follow-on orders of Sonoma Systems expected during FY27 from their largest hyperscaler customer for their AI processors ASICs. - They are scaling their co-man to produce 20 additional Sonoma Systems per month, implying ability to support $360M - $480M of Sonoma revenue ($1.5M-$2.0M ASP) All good news. Stock likely got a little ahead of its skies though and chasers might get caught here. I'll be looking for a re-entry this week. Will update in real time as always.

Relatively unknown CPO long idea: Browave (3163.TWO) at ~$2.4B market cap Decided to research more, and have decided to go long (small position) as the story seems pretty compelling. Just some very top level notes that are hopefully digestible. ----- Browave provides the fiber arrays and internal interconnects necessary for CPO to bridge the gap between AI chips and optical signals. ​Their Fiber Shuffle Box is a key CPO enabler that organizes massive complexity inside AI servers. With mass production scaling 10x units per month by early 2027. Partner with $GLW for high density assembly. ----- ​As AI data centers transition from 800G to 1.6T and look toward 3.2T, they face a physical crisis: Because CPO system and chip designs vary significantly (e.g. $NVDA vs. $AVGO vs. Hyperscaler custom ASICs), off-the-shelf connectors are failing. ​Traditional MPO architectures are too bulky for the 1U switch trays of the Vera Rubin and Feynman. Integrating optics inside the package makes them nearly impossible to service if a fiber breaks, unless a detachable mechanical interface is used. ​ > ​Browave is scaling CPO production from an initial 1,000 units/month in Q4 2024/2025 to a targeted 10,000 units/month by Q1 2027. This indicates a massive secured order book, likely tied to $NVDA's Spectrum-X Ethernet rollout scheduled for H2 2026. ​ > Browave has also launched MMC jumpers, which offer 3x the density of traditional MPO solutions. This lets $NVDA and others to fit 1,000+ fibers into a single 1U tray without blocking airflow. Browave anticipates that initial 2026 deployments will focus on scale-out switch architectures (rack to rack), with scale-up adoption (GPU-to-GPU) becoming the driver by 2028 A market that could be 10x larger than the current scale-out TAM. ​In terms of suppliers: Browave relies on Landmark for the InP and GaAs epiwafers used in the lasers they pig tail and assemble. ​ $SIVE also collaborating on ELS modules to separate lasers from 1000W GPUs. Risk disclosure: > Browave itself admits that meaningful scale-up CPO adoption (the 10x multiplier) won't materialize until 2028. This creates a potential investment valley in 2027. > LPO remains a viable, lower cost bridge for 1.6T. If hyperscalers choose LPO over CPO for another 24 months, Browave’s growth targets will be deferred. ----- ​Tldr: ​While logic companies fight over chips, Browave owns the "shuffle" that makes those chips useful. The recent 10x capacity surge target is the strongest fundamental indicator that a regime change in AI networking is underway. ​ Am Watching for the official announcement of the private placement subscriber in May/June 2026. This will determine if Browave is a "protected asset" in the $NVDA / $TSM supply chain. Note, I have not looked into financials yet as have been travelling, so investing purely on the story for now. But as mentioned, just a small speculative position for now. More focused on the likes of $TSEM, $AAOI, Samsung, $AEHR, $LITE / $COHR