AI Infrastructure · Investment Research

Dissecting @aleabitoreddit’s 6,700+ Tweets: The Physical Bottlenecks Behind a +412% AI Infra Portfolio

During the historic AI bull run, retail investors often chased foundation model benchmark releases, consumer AI wrappers, and flashy keynote slides. Meanwhile, on X (formerly Twitter), a low-profile hardware investor known as @aleabitoreddit executed an uncompromising, engineering-first playbook centered on physical supply chain bottlenecks—posting a self-reported YTD +412.72% mark in late February 2026 and a YTD +3,152.77% mark by mid-May, while noting that over 70% of the portfolio was still unrealized.

Over more than a year, their timeline accumulated over 6,700 tweets and real-time channel checks, capturing first-principles insights across Indium Phosphide (InP) substrates, Co-Packaged Optics (CPO), Neocloud compute arbitrage, and humanoid robotic actuators. To ensure this high-density research is preserved beyond ephemeral feeds, we organized the dataset into a 1,200-note local Markdown vault, ready for one-click import into NoteLoom without installation.

The collection spans more than a year of hardware research; original post images are embedded as remote links so they add zero local storage weight (text and backlinks work offline). Furthermore, the standalone threads keep their original English phrasing (alongside curated research write-ups), and every reply to someone else is preceded by a “replying to @user” line that links the original post.

1. The Four Non-Consensus Rules of Hardware Investing

Across thousands of posts, four principles come up again and again (the grouping is ours; each quoted claim traces back to a post in the vault):

  • Physical Bottlenecks Over Commercial Narratives: Software narratives can be exaggerated, but Maxwell’s equations, thermodynamics, optical attenuation, and silicon substrate physics do not bend. In any technological arms race, outsized profits concentrate at choke points with multi-year manufacturing lead times.
  • Nvidia as the Ultimate Leading Indicator: Instead of trading rumors, monitor Nvidia’s ($NVDA) architecture roadmap (Blackwell, Rubin) and bill-of-materials (BOM) transitions. When Nvidia pivots toward optical interconnects or direct-to-chip liquid cooling, adjacent hardware sectors enter mandatory multi-quarter expansion cycles.
  • Holding Beats Churning: In May 2026 they said over 70% of the portfolio was still unrealized gains, and that after the 2016–2018 crypto market, ±20% swings in names like $AXTI or $AAOI feel like nothing.
  • Call the Next Theme, Then Concentrate: They describe the core skill as knowing which theme comes next, picking the winner, and concentrating heavily; they said most of their portfolio sits in photonics names like $SIVE and $AAOI.

2. Six Choke Points in the AI Value Chain

In the curated vault, all research threads are organized across six primary physical bottlenecks:

Sub-Sector Physical Choke Point Key Tickers Underlying Driver
Optics & CPO Copper reach limits, thermal dissipation walls, InP wafer supply deficits $LITE, $COHR, $SIVE, $AAOI, $AXTI, $POET Full transition from 800G to 1.6T transceivers, commercial adoption of CPO, and severe CW laser shortages
Neoclouds & Power Hyperscaler internal GPU hoarding, multi-year utility interconnect delays (MW capacity) $NBIS, $CRWV, $APLD, $IREN, $BE Securing energized multi-megawatt facilities and net cash positions during explosive inference demand
Memory & Interconnect HBM wafer allocation cannibalizing standard DRAM; horizontal bandwidth deficits in clusters $MU, $MRVL, $CRDO, $ESMT Broad memory super-cycle coupled with customized ASIC acceleration and high-speed PAM4 DSP expansion
Advanced Packaging CoWoS capacity ceilings, large-panel ABF substrate warpage, glass core planar tolerances $TSM, $AMKR, $ASML, Ajinomoto (ABF) Multi-die chiplet architectures driving demand for fan-out panel-level packaging (FOPLP) and high-NA lithography
AI Hyperscalers Ultra-cluster interconnect failure rates, multi-billion-dollar Capex durability and monetization ROI $NVDA, $MSFT, $GOOGL, Anthropic, OpenAI Global compute foundational backbone, funneling hardware cash flows into enterprise adoption and model training
Embodied AI Robotics Sub-micron mechanical tolerances, torque-to-weight ratios, dual-arm harmonic fatigue limits Harmonic Drive (6324.T), LeaderDrive, Sanhua, Moons' Compute escaping digital boundaries into physical actuators; precision gears replicating the multi-bagger curve of optics

3. Why a Local-First Markdown Vault Beats Web Browsing

While social feeds and web bookmarks are convenient for casual reading, institutional research across thousands of notes collapses in a browser tab:

  • Ephemeral Feeds and Dead Links: Tweets get deleted, accounts change handles, and platforms tweak algorithmic search, destroying months of saved references.
  • Lack of Cross-Document Linkage: Reading an article about $LITE on a webpage rarely reveals its dependency on Indium Phosphide suppliers ($AXTI) or previous channel check mentions.
  • Absence of True Data Ownership: Serious research belongs on your own disk, where you can instantly grep, query with local LLMs, and maintain a permanent personal archive.

By transforming this archive of over 6,700 tweets into a fully linked Markdown graph, every company, technical term, and historical thread connects natively via [[wikilinks]], forming an offline, searchable AI hardware intelligence network.

4. One-Click Import into Your Local Disk via NoteLoom

To make this intelligence base instantly usable without configuration, NoteLoom enables a seamless one-click local import leveraging the modern Chromium File System Access API.

Click the button below to launch NoteLoom in your desktop browser and write all 1,200+ notes directly into your chosen local folder. Zero cloud sync, zero sign-ups, and full offline access to the text (images load from the original posts, so they need a connection):

Import Full Vault into NoteLoom Locally (Instant Launch) →

📦 Prefer Manual Download for Obsidian or VS Code?

If you regularly use Obsidian, VS Code, or Logseq, you can download the standalone .zip package directly. Unpack it into any local folder to immediately explore the entire linked vault:

Download Complete Vault Archive (.zip) ↓

One-click import runs smoothly on desktop versions of Chrome, Edge, and Arc.

5. Vault Structure & Index

Once imported, the vault presents a clean, two-layer institutional hierarchy:

  • Curated Synthesis Layer (Core Research Dossiers):
    • 00_Dashboard_AI_Hardware_Bottlenecks.md: Master supply chain map with Mermaid dependency diagrams and Top 20 ticker hit rates;
    • 01_Core_Investment_Philosophy.md: Recurring principles such as bottleneck rotation, valuation vs. narrative, and asymmetric upside with verbatim source citations;
    • 02_Semiconductor_and_Compute_Glossary.md: Core semiconductor and hardware definitions (InP, CPO, FOPLP, ABF, Harmonic Drives, Neoclouds);
    • 10_Sector_Deep_Dives/: Deep-dives into Optics/CPO, Neoclouds, Memory, Packaging, Hyperscalers, and Robotics;
    • 20_Core_Tickers/: Company tracking dossiers ($NVDA, $LITE/$COHR, $SIVE, $AAOI, $AXTI, $MRVL, $NBIS, $MU, $POET) + Valuation & Catalyst Matrix;
    • 30_Iconic_Thread_Digests/: Institutional research editions of high-engagement viral thread analyses;
    • 40_Deep_Dives_and_Filings/: Mature DRAM turnaround logic, 13F filing analyses, and supply chain channel checks.
  • Raw Archive Layer (Verbatim Source Posts):
    • 90_Raw_Tweets/Threads/: 1,153 thread pages (272 standalone threads + 881 single posts with reply contexts) grouped into monthly folders, each featuring a 00_Monthly_Index.md sorted by likes and tagged by ticker;
    • 90_Raw_Tweets/Monthly_Archives/: Chronological monthly digests preserving all standalone tweets and conversation contexts;
    • Post Images: Original post images embedded as remote links, adding zero local storage weight.

Related Reading

FAQ

What is inside this AI Infrastructure Investment Knowledge Base?
This vault compiles and structures over 6,700 tweets and research threads from more than a year of @aleabitoreddit’s timeline, refined into more than 1,200 structured Markdown notes with over 2,600 post images embedded as links (text and links are fully local). It features a 2-layer architecture: a top layer of curated Chinese research dossiers (AI infrastructure dashboard, glossary, sector deep-dives, company tracking notes with a valuation matrix, thread breakdowns, and 13F/channel check reports), and an underlying raw archive of original English post pages and monthly timeline digests.
Why import the vault locally into NoteLoom instead of reading it on a webpage?
Browsing thousands of fragmented tweets on static web pages loses the true power of relational research. This vault is interconnected with native bidirectional wikilinks and backlinks. When reading about optical transceivers, every related mention of Indium Phosphide (InP) substrate capacity, 1.6T production ramps, and Coherent vs. Lumentum earnings calls is instantly connected. NoteLoom uses the modern Chromium File System Access API to write all 1,200+ notes directly to your local computer in seconds—zero cloud sync, complete data ownership, and instant offline full-text search. (Images are embedded as links to the original posts, so they add no local weight but do require a connection; text and backlinks work offline.)
Does the one-click import upload my personal files or existing notes?
Never. NoteLoom is built on a strict Local-First architecture. The import process is purely a one-way static resource download: your browser fetches the public export manifest from the static CDN and uses the File System Access API to write the files into the folder you designate on your hard drive. No user data, keystrokes, or local notes ever leave your machine.
Can I use this knowledge base in Obsidian or other Markdown tools?
Yes, 100%. All notes in the vault follow standard GitHub Flavored Markdown (GFM) and universal [[wikilinks]] syntax, free of any proprietary plugins. You can download the complete .zip archive using the link on this page and open it directly in Obsidian, VS Code, Logseq, or any plain text editor.
What is the core premise of the "Physical Bottleneck" investment thesis?
The core rule is: "Physical bottlenecks triumph over commercial narratives." In the generative AI gold rush, software features and LLM wrappers face rapid commoditization, but fundamental physics—thermal dissipation limits, optical loss in silicon, copper skin effects, and micron-level mechanical tolerances in robotic joints—remain non-negotiable constraints. Companies controlling the scarce manufacturing capacity for these physical choke points capture the most durable margins and pricing power.

Published · Updated