August 2026 Random Ramblings

Summary

• In the episode proper, Andrew Walker says he is short a little Strategy (ex-MicroStrategy) common and long a little of the prefs, and can’t get over the latest 8-K: the company sold $300M of stock and $100M of Bitcoin, using about $80M of the Bitcoin-sale proceeds to buy STRC preferreds at a slight discount and park cash — while still trading over NAV. Management framed evolving from a “one-way capital issuer” to a “multi-way capital issuer” as genius; Walker’s verdict: that’s “not even good, like normal capital allocation,” and “how is this company trading at a premium?”
• Credit where due: Strategy’s 2020 pivot was a genuinely good call — Walker estimates Bitcoin was around $20K then versus ~$65K today (peaking ~$120K), and the bet that equity investors wanted “a one-click button to buy Bitcoin” was right. But they “kind of piled in at the top,” are underwater on overall purchases, and Walker sees “a completely busted capital structure that needs to be reset.”
• The through-line: investors are good at buying and bad at selling, and Walker questions his own mechanical exit plans. Buy at 10, think it’s worth 20, trim at 15 — but anchored sell levels make it hard to press when “this is better today than it was yesterday, I need to buy a lot more,” which is where “a lot of the big money is made.”
• The Situational Awareness blow-up: a generational long-AI-winners/short-AI-losers call, up ~10x over roughly two years, undone by never rebalancing. In his stylized version, long semis/power and short software both at 100 — semis to 400, software to 20 — a naturally de-grossing trade, but “they kept pressing and pressing,” and when software went 20→30 and semis 400→370, the fund blew up. The Anthropic private stake, which Walker says they could not really lever, is why he thinks they didn’t zero the fund out.
• Thematic trades have no price, which makes selling them unusually hard — and Walker says the past decade was an exception to the historical rule that “if it’s in the news, it’s in the price.” Long software in 2016, AI in 2024, or GLP-1s worked at almost any entry; the unanswered question is “where is the top?” One micro-to-macro tell he’s pursuing: the “juicy comp grants” software executives gave themselves in March/April were, in his view, a kind of buy signal around the bottoms before a ~50% index rip, but he found the trade too hard to act on over terminal-zero risk.
• On the theory that crossover AI funds benefited from quasi-MNPI — seeing Anthropic/OpenAI numbers before buying into the DeepSeek scare in January 2025 or the March semi sell-off — Walker mostly demurs: “I think they had real conviction here.” But conviction has limits: on memory stocks, “there’s just simply no way you can justify the current valuations” — these are commodity plays and “the back end is going to look really ugly once these get overbuilt,” unless memory demand can never catch up.
• The potential opportunity: former Bitcoin-miner power shells, hit hard by the Situational unwind because they were among its biggest positions, now include names trading near the DCF of their contracts with CoreWeave or Meta. “If you can buy them for the DCFs of their contracts and get everything else for free… that’s a very interesting call option to me” — with a CoreWeave credit-risk caveat and heavy diligence needed on who bears power costs, capex, and GPU risk.

Deep dive

1. Strategy discovers capital allocation — and gets applauded for it

• The 8-K that set Walker off: Strategy sold $300M of stock and $100M of Bitcoin, using about $80M of the Bitcoin-sale proceeds to buy STRC preferreds at a slight open-market discount and park cash. So a company trading over NAV is diluting shareholders and selling Bitcoin to buy back prefs — “how is this company trading at a premium? … it’s so crazy to me.” In the episode proper, he discloses that he’s short a little Strategy and long a little of the prefs.
• His read of the earnings call: management touts evolving from a “one-way capital issuer” (issue stock to buy Bitcoin) into a “multi-way capital issuer” — “saying it like they’re some genius” when “all they’re talking about is good capital allocation. I mean not even good, like normal capital allocation.”
• The fair credit: Walker estimates that Bitcoin was about $20K when Strategy made the 2020 pivot, versus $65K today and a ~$120K peak, and the underlying theme — equity investors wanted crypto exposure they couldn’t get pre-ETF — was right. Walker was skeptical then; “they were betting that people wanted a one-click button to buy Bitcoin and they were right.” But they “kind of piled in at the top,” sit underwater on overall purchases, and the structure is “completely busted… needs to be reset.”

2. The sell trap: mechanical exits versus new information

• The research Walker keeps returning to: investors “are very good at the buying. They’re very bad at the selling.” His own template — buy at 10, believe 20, start trimming at 15, heavier at 17-18, out at 20 — now looks suspiciously like Strategy’s old mechanical approach: “am I falling into that sell trap… I’m not putting as much thought into the sale.”
• The cost of anchoring: “a lot of the big money is made where you buy a stock at 10 and it goes to 20 and you say this is better today than it was yesterday, I need to buy a lot more” — but a pre-set sell ladder makes it “very difficult to flip your mind” when the clearing event lands mid-exit. He offers “no great answers,” just the diagnosis.

3. Situational Awareness: a generational call with no rebalance

• The trade: long AI winners, short AI losers, put on roughly two years ago, up ~10x. Walker’s take on why it didn’t go to zero: the big Anthropic private stake “that’s up a ton, and that they couldn’t really lever is actually why they didn’t zero the fund out.”
• The blow-up mechanics, simplified: long semis and power, short software, both at 100 — semis to 400, software to 20. That’s “a naturally de-grossing transaction unless you keep putting on, but they kept pressing and pressing and pressing,” so when software went 20→30 and semis 400→370, “the whole fund blew up.” It was a risk-management failure, including a failure to rebalance or press sell after the winner ran.
• The human aside: the trader’s wedding was the same weekend as the implosion. “It’s easy to say, ‘Oh, go have fun at your wedding’… I can’t imagine going to that wedding and having that hang over you.”

4. Thematics have no price — and lately haven’t needed one

• The structural problem: a thematic call — long AI, long GLP-1s — carries no valuation anchor. Historically, known themes were priced (“hey, we want to go long health care in 2006 because the population is aging” — already in the price). But Walker says the past decade was different: software-eats-the-world in 2016 could work at almost any price for years, through the 2022 growth reset and then ChatGPT’s arrival; AI in 2024 “basically didn’t matter the price.” The unanswered question: “where is the top? When does it matter?”
• Walker’s attempt to expand his micro process toward macro: in March/April, software companies across the board handed executives “really juicy comp grants” — “that was kind of your buy signal,” and the software index is up ~50% since those bottoms. He saw the panic but found the trade too hard to act on: “there’s the chance of all of these being terminal zeros.”

5. The post-blow-up opportunity: power shells near contract DCF

• On the theory that crossover AI funds benefited from quasi-MNPI — not criminal MNPI, but access to company numbers — Walker relays claims that they saw Anthropic/OpenAI numbers, bought into the DeepSeek scare in January 2025, and piled into the March semi sell-off. He describes the person in question as a former OpenAI researcher who is, “I suppose,” now married to Anthropic’s chief of staff. Walker says investors making this case are “asking a barber if you need a haircut” because they are talking their own book; he allows “a little bit of truth” but thinks they had real conviction. The numbers probably helped build that conviction, rather than replacing it.
• Where conviction can’t stretch: memory stocks. “There’s just simply no way you can justify the current valuations” — take whatever supernormal profits you want for 2-3 years, “these are commodity plays and the back end is going to look really ugly once these get overbuilt,” unless memory demand can never again catch up.
• The potential idea: former Bitcoin miners turned AI data-center plays — Bitcoin mining “might be the worst business ever invented” — were among Situational’s biggest positions and got hit hard, bouncing but not as much as Walker thinks they should have. A lot of these names, he says, are not trading for much more than the DCF of their contracts with CoreWeave or Meta: “if you can buy them for the DCFs of their contracts and get everything else for free… that’s a very interesting call option to me,” with a CoreWeave credit-risk caveat.
• His open diligence list: contracts vary enormously — how much capex remains, what terminal value to assume, what the shell is responsible for, whether the customer or data center pays for power, and whether it’s a triple-net lease or the shell also buys the GPUs, “a much higher risk higher return business.”