TopxAI
Opening page
The AI supply chain, from copper ore to inference.
Every layer taken apart one by one. Each company is sized by market cap, each relationship carries a filing source, and the page says which layer is the choke point and whether money can clear it.
18 layers · 180 US-listed companies · 892 binding relationships
Market caps as of 2026-09-25 · refreshed every trading day · relationships from public filings through 2026-09-22
Research beyond the headline.
Primary sources. Independent thinking. A closer look at what changes next.
Keep the whole picture in view.
Explore membership for the complete library and new releases.
Fresh perspectives.
Ten Straight Quarters of Negative Operating Cash Flow Do Not Mean Mining Loses Money: Three Lines to Read on a Miner's Cash Flow Statement
A bitcoin miner books the coins it mines as revenue, but coins are not cash, and the proceeds from selling them can sit in investing activities. Operating cash flow is therefore negative by construction. Using ten quarters of Riot Platforms filings, this piece lays out three lines to read side by side, and explains why a widely quoted non-GAAP figure swings with the bitcoin price.
Read the storyFrom $60 to $250 a Share, the “Price-Implied” 2030 EPS Stays Between $4.43 and $4.90: A Reverse-Engineering Trap, Shown on FormFactor
Compound today's price at the required return to the end of 2029, divide by the current forward multiple, and the 2030 EPS that comes out tends to be called “price-implied”. But that multiple is itself price divided by consensus, so the price cancels and the reading is simply consensus compounded at the required return. Using FormFactor's numbers, this piece shows the trap, how to check for it, and where the price information actually sits.
97.3% of the Expansion Money Came from Selling Stock: Where AAOI's $1.547bn of ATM Proceeds Went
From early 2025 to June 2026, Applied Optoelectronics raised $1.547bn net by selling stock at the market, 97.3% of its external funding; customers prepaid nothing. This piece rebuilds a sources-and-uses ledger for that money by differencing year-to-date cash-flow figures, flags three places where pulling XBRL tags in bulk gets it wrong and one where the 10-K text reports a single month as a whole round, and compares where the peers found their expansion money. All of it can be reproduced from public filings.
