Every fact carries its own source
Company roles and filing rows link to the public document behind them, with a date. Data refreshes on a schedule, not live on each click.
Research, not a buy list
Read the AI supply chain from the bottleneck down.
Which layers AI demand has to pass through, where supply is tight, which companies do that work — and the public source behind each claim. Free to read, no sign-in. Not a trade signal; you decide.
- Open a layerStart from the map, not a ticker.
- Read a sourceEach claim points at a public document.
- You decideNo buy list. No price target.
- Free map and ticker research — no account wall on public pages.
- Each company claim links a public source with a checked-on date.
- Counter-view and “what this does not tell you” on research threads and layers.
- en / 中文 / 한국어 URLs — same HTML for people and crawlers.
AI Industry Map preview
Real layers from the worker cache — tap one to open research. This is the map, not the signed-in console.
Find the bottleneck before you look at the stock
This week's research thread, read from the worker cache: the Compute layer. It rotates weekly across layers whose sources are complete — not by price or activity.
Why this layer matters
The visible centre of the chain, and the layer most people stop at. It is also the layer that cannot run alone: an accelerator needs memory bandwidth, packaging, interconnect, optics, storage, power and cooling around it. Any one of those lagging and the silicon idles.
Where supply is tight
Design leadership concentrates here, but the near-term ceiling is usually set elsewhere — packaging, HBM supply, or power — rather than by accelerator design.
XingAI's structural observation about suppliers and lead times — not a view on any security.
Companies whose main business sits here
This layer holds merchant chip vendors, custom-chip designers and CPU architecture licensors; they compete in different ways.
Company
Advanced Micro Devices AMD
Role
Supplies Instinct data-center GPUs used for AI training and inference.
Evidence
Source: AMD Instinct GPUs · checked 2026-10-01Company
Arm Holdings ARM
Role
Licenses CPU architecture and Neoverse designs that partners build into data-center and cloud AI chips.
Evidence
Source: Arm for cloud AI · checked 2026-10-01Company
Broadcom AVGO
Role
Designs customized system-on-chip ASICs for customers' own systems, including AI accelerators.
Evidence
Source: Broadcom custom silicon · checked 2026-10-01Company
Cerebras Systems CBRS
Role
Builds wafer-scale AI systems, currently marketed for fast inference.
Evidence
Source: Cerebras CS-4 · checked 2026-10-01
2 more on the layer page.
The other side
Concentration in accelerator design may not last. Large cloud operators design their own chips, inference can run on cheaper or more specialised parts, and software that makes models portable across vendors lowers the cost of switching. Each of these spreads compute spending across more designs instead of concentrating it.
What would change the picture
If packaging, HBM and power stop being the limit, accelerator design itself becomes the ceiling and this layer matters more. If custom chips take a growing share of new deployments, or inference moves to lower-cost hardware, design leadership here matters less.
What this does not tell you
Layer placement and these notes are XingAI's classification; each source supports only the company role beside it. Price, valuation and timing are not covered here. You decide.
Full layer evidenceResearch thread date (UTC): 2026-10-20 — not a live quote tape or trading session date
Watch our weekly show
Wall Street Views Weekly, every Sunday: the week's biggest Wall Street calls — who said what, the data behind it, and the other side of the argument. Research and education only, same as everything here.
No citation, no claim.Trust & limits
Our labels are marked as ours
Layer placement and bottleneck notes are XingAI's taxonomy. A filer disclosed a position; it did not endorse our map.
The limits of 13F filings
Form 13F-HR is reported 45+ days in arrears, covers long US equity only, excludes shorts, cash, non-US holdings and most derivatives — and a reported position may already have been sold.
Not advice
Not a registered adviser, not a broker, and not a signal to buy or sell. Verify filings and prices before you act.
More research pages
US ETFs, side by side
Thirty widely followed funds in twelve groups. Yield, fee, and size — research only, not a buy list.
Why I Own It
A small investor's own notes on TSLA, AMZN and the rest — the products he uses every day, and why that is the reason he holds them. Research only.
Method
How data refreshes, why 13F filings lag, how layers are labeled, and alerts you define.
Disclaimer
Research and education only. Not advice. You decide whether to act.
Open the signed-in console — same research cache, plus your watchlist and alerts you define. Educational context, not a personal recommendation. · Sign in
FAQ
- Is this investment advice?
- No. Content is informational and educational. Nothing here is a solicitation to buy or sell a security.
- What can I read without an account?
- The AI Industry Map at invest.xingai.app/ai-map, plus method, disclaimer, terms, and privacy.
- Which languages are supported?
- English, 中文, and 한국어 in the interface. Personal PDF reports are Chinese.