Good morning traders

šŸ—‚ļø On the desk today

šŸ“Š The Tape: what rates did to everything on Monday.

🌊 Tom Lee's Ethereum: what his company holds, what it cost, and why he can keep buying.

🧾 AI's $6 trillion bill: what Bain said, and what a viral post left out.

ā±ļø One thing worth doing today: write your exit down before you need it.

šŸ“Š The Tape

Gold fell more than 3% on Monday and silver more than 4%. When a government bond pays more than 5% to do nothing, a metal that pays nothing has to work harder to earn its place.

The week's big number is still ahead: September payrolls, Friday 2 October at 8:30am ET.

🌊 He's billions under water. Why is he still buying?

What the Ethereum cost, and what the same coins are worth now.

Most people think a buyer who's under water has to stop. Tom Lee's company says it hasn't stopped for one week in fifteen months.

Lee chairs Bitmine Immersion Technologies, a listed company whose business is now almost entirely buying Ethereum, holding it and staking it. What he owns personally isn't public. What the company owns is filed every Monday.

What it holds, from Bitmine's release on 28 September:

→ 6,001,302 ETH, priced at $2,698 each. That's 4.9% of all the Ethereum in existence.

→ 213 Bitcoin, $672 million in cash and securities, and two smaller stakes.

→ $17.2 billion in total.

What it cost. The last quarterly report, to 31 May, put the Ethereum at a cost of $19.05 billion for 5,416,945 coins, about $3,517 each. At the $2,698 in Monday's release, those same coins are worth about $14.6 billion. That's $4.4 billion under what they cost, 23% below the average price paid. Anything bought since June isn't in a filing yet.

Why he doesn't have to sell. Bitmine isn't buying with loans against the coins, where a falling price forces a sale. It's buying with new shares. In the nine months to May it sold 340.7 million new shares for $11.87 billion, and the share count went from 232.4 million to 579.7 million.

So there's no loan against the coins and no lender who can force a sale. The cost lands on shareholders instead, who own a smaller slice each time.

Why he's bullish, in his own words from the company's weekly releases: a crypto bull market he says started in late June, Ethereum as "the settlement rails for Wall Street tokenization", and "institutions are still underweight crypto".

He bought another 17,362 ETH last week. The company says it has bought every week since 30 June 2025.

ā

šŸ’¬ The Desk Take: Conviction isn't a signal. He has a structure that lets him wait years. A position 23% under your cost feels very different when it's your own money and nobody's issuing shares to fund your next buy.

🧾 "AI is $4.2 trillion short." That isn't what the report says.

One number, two headlines.

A post went round this week saying the maths behind the AI boom doesn't add up. The numbers in it are real. The headline isn't Bain's.

Bain's technology report came out on 29 September. It says paying for the computing power being built needs $6 trillion of revenue a year by 2031. What AI already sells could reach $1.2 trillion to $1.8 trillion. The remaining $4.2 trillion has to come from new kinds of product: self-driving cars and trucks, robots in factories, drug discovery.

The post calls that a hole. Bain calls it a gap too, one it expects new products to fill, and its own headline is "Global AI market could hit $6 trillion annually by 2031".

Same number. Opposite headline.

The part I'd watch is smaller and duller. I call it paper years. A company that buys a server spreads the cost over the years it expects to use it. Stretch the years and each year's cost shrinks, so profit rises without one extra sale. Microsoft moved its servers from four years to six. Alphabet uses six. Meta uses 5.5. Amazon went the other way last year, six down to five, and said AI hardware was moving too fast.

Michael Burry puts the gap at $176 billion of understated cost between 2026 and 2028, as reported by Fortune and others. That's his estimate, not the companies'.

ā

šŸ’¬ The Desk Take: A 2031 forecast can't be traded. The next earnings report can be read. Check two lines: is spending growing faster than the cash coming in, and did the server life change. Search the report for "useful life".

ā±ļø One thing worth doing today

Pick the position you like most and write down what would have to happen for you to sell it.

Not a feeling, a fact. A close below a line, a number in an earnings report, a date. Tom Lee can afford not to have one. You can't.

If you're not sure where your own process leaks, the Trading IQ quiz is ten questions on trend, entries, risk and discipline, and it emails you a report.

šŸ“¬ The full version

This week's Gate Board on Substack has the rest: where the three gates sit before Friday's jobs number, a 50 second video of the scanner I use to find stocks to buy, and the record. It's free to read.

If you want your exits written down next to your entries, the RB Trading Journal has a plan field on every position, 20% off with code RBT20.

If you hold coins for the long run, keep them off an exchange. A hardware wallet does that, and Ledger is the one on my links page.

Everything in one place: start.rbtrading.site

šŸ”œ Next issue

Friday, at the New York close. What the jobs number did, and one company taken apart.

Education, not financial advice. Company figures are from each company's own releases and filings. Some links are affiliate links, and I may earn a commission at no extra cost to you.