Agent

Here's a rundown of what I can do and the research I have access to.

What I am

I'm a research assistant that answers only from the curated research libraries connected to this workspace — I don't rely on my own background knowledge, and I cite every claim back to a specific source in the corpus. If the answer isn't in the library, I'll say so rather than guess.

The research I work with

1. Creative Strategies knowledge base — public articles and research reports from Creative Strategies, including Ben Bajarin's research notes on semiconductors, AI, and consumer tech. Examples of what's in there:

• Earnings deep-dives on AMD, Micron, and others (e.g., AMD's data-center GPU forecast raised to $4.5B+ in 2024, later to $5B+) 1
• Long-form research like The Semiconductor Gigacycle, which tracks how the AI hardware TAM — CPU, GPU, ASIC, networking — is projected to exceed $1 trillion by 2030, with dedicated AI accelerators alone reaching roughly $300–350B by 2029–30 13

2. Diligence Stack Reports corpus — source-linked institutional research notes, market primers, model knowledge, and uploaded report material. Categories include AI Networking, Market Primer, Model Knowledge, SaaS, and Vendor Notes. Examples:

• The AI Infrastructure Buildout: A Comprehensive Framework for the Datacenter and Power Cycle — covers the optical transceiver supercycle (38M 800G units and 14M 1.6T units projected in 2026), back-end Ethernet fabric growth to ~$11B in 2026, and 9–12 month AI networking lead times 11
• AI Networking Silicon: The Market Is Looking at the Wrong Layer — the thesis that fabric-layer silicon, not optics, is the durable control point in AI infrastructure, mapping scale-up (NVIDIA), scale-out (Broadcom), and scale-across (Cisco/Arista) control points 12

3. Published editorial notes — dated house commentary in five types: earnings reads, news reads, model updates, thesis checks, and signal notes. The most recent batch covers earnings for AMD 2, Astera Labs 3, Apple 5, Arm 6, Microsoft 7, MediaTek 8, Qualcomm 9, Meta 10, Sandisk 1, and SpaceX 4.

What I can do

• Search across both libraries in parallel and pull the exact evidence behind any hit, with source links.
• Read full reports when a snippet isn't enough, including long PDFs page by page.
• Hand you original files — if you want a PowerPoint deck, Excel workbook, or PDF artifact itself, I can fetch the exact original download link.
• Show you source figures — charts, tables, and page images from the research, embedded inline where they help.
• Build comparison charts from data synthesized across multiple sources, always plotting only numbers the sources actually support.
• Give you the editorial history on a company or theme — for example, the earnings-read trail for AMD, Microsoft, or Apple.

What I can't do

• I can't invent or extrapolate: every number I give is traceable to a fetched source, and I flag stale or superseded material when dates matter.
• I can't see raw images or pages directly unless an image is attached, though I can work from derived text and provide the original file when available.

Want me to demonstrate? Try asking something like “What does the research say about the AI networking TAM through 2030?” or “Summarize AMD's earnings trajectory across the research notes.”