June 2026 Random Ramblings

Summary

• Andrew Walker’s provocation of the month: with SpaceX announcing a ~$60 billion acquisition of Cursor — in which FTX/SBF held ~5%, later sold back to Cursor at cost by the bankruptcy court — was SBF secretly the greatest VC of all time? Walker says SBF had an early investment in Anthropic, thinks he had investments in SpaceX and Robinhood, and possibly one or two more; “a grand slam in VC is a thousand X, not four X… He hit three.” His tentative thought: SBF may have “frauded his way into the VC network effect” — spraying money bought deal access and brand effects that legitimate firms rely on, with SoftBank’s Masa the only other example he can think of.
• Walker couldn’t think of another fraud or blow-up holding assets that might have “returned the entire equity.” Enron had pipelines (which Kinder Morgan, he believes, left to build on) and spun out EOG — Enron Oil and Gas, a ~$75 billion EV company — 2–3 years before bankruptcy, but nothing on the scale of FTX’s venture book; GGP’s GFC bankruptcy was a liquidity event with real mall value, not a fraud.
• On AI he’s climbing out of his “valley of despair” toward AI-as-force-multiplier: “if you’ve got vision, you’ve got creativity, I think AI is really going to reward that.” His analogy: the best singer in a 1500s European village could probably find work; today’s singer must be “top 0.000001%” — yet he leans optimistic that “there’s just so much human creativity, everybody’s going to be able to carve a niche out for themselves.” Non-coders needn’t assume they must code: “Why do you need to learn to code to use AI? Why can’t AI do the coding?”
• The tradeable corollary: as AI slop floods everything, deep subject-matter expertise may command a growing premium. A college student can now produce a basic research report in 30 minutes that reads “really effing good” to a layperson, but the Knicks-challenge example shows what an expert might add — AI might say “you’re 98% to win” while the expert saves the challenge for a higher-leverage spot; likewise Adam’s read of ABBVX’s safety signals (“look at the sample sizes, look at the base rates”) versus a generalist’s “black-box warning, boom.” Disclosure: Walker is long ABBVX.
• Second brand thesis: legacy brands may “suck back up more equity” in an AI world of second-hand hallucinations. The FT reported that a lot of KPMG’s report on corporate AI use was hallucinated, with Shell, Walker thinks, among the companies discussed; “big consultancies such as KPMG and EY are viewed as highly credible, so their reliance on false information increases the risk of second-hand hallucinations.” Extend it to CBS News, even People/TMZ for celebrity gossip — and Walker “could see a world” where ChatGPT pays trusted sources for citable access.
• Hot take on the common edge claim in investing: “my edge is time horizon” is often underperformance in disguise. Looking further out — say, 6 months or 1 year versus 3 years, or 3 years versus 5 — “is literally saying, hey, I drag my Excel file over a column or two.” The diamond-hands version presupposes the drawdown — “if that’s the presup, why do you need to buy at 10 and go to five? Couldn’t you buy at five?” The tell: “the unspoken part I generally see is we are underperforming” — hence his 3-year rule: a stock that’s done nothing for three years demands a look in the mirror.
• Polymarket’s MicroStrategy resolution — Bitcoin was sold during the week of May 31, the 8-K was filed June 1, and the “sell by May 31” market resolved NO — exposes what Walker calls a structural risk: persistent gray areas plus real-world reflexivity. In a hypothetical, he said Elon Musk could have skipped $5 million in Ohio or wherever it was and instead pushed the 2024 Trump market from 60 to 80 two weeks out to demoralize opponents; in thin markets $5,000 can make anyone “the Polymarket favorite.” And a reminder: if these markets are functioning, 55% “should mean… it’s almost a coin flip,” not a certainty.

Deep dive

1. Was SBF the greatest VC ever — or did fraud money buy the network effect?

• The trigger: SpaceX announced it’s buying Cursor for about $60 billion, and FTX/SBF owned roughly 5% — “would have been like $2 billion or something” — which the FTX bankruptcy court sold back to Cursor at cost. Walker says SBF had an early investment in Anthropic, thinks he had investments in SpaceX and Robinhood, and thinks there were one or two more: “a grand slam in VC is a thousand X, not four X… He hit three.” Walker’s first reaction: “why were you running the fraud? You’re running the greatest VC investment firm on Earth.”
• His mechanism for deflating the genius theory: VC returns run on network and brand. His own example — offered 10% of a startup for $1M from Kleiner Perkins or $1.1M from Andrew Walker, “you should go take the Kleiner Perkins deal,” because the brand recruits employees and customers. And a common trait of fraudsters from the fraud books: “they’re just willing to spray money around because it’s not their money” — 1MDB, Enron. So maybe SBF “bootstraps the VC network effect on steroids”: “you kind of frauded your way into the VC network effect.”
• Could a rich person replicate it? Masa at SoftBank is the only other example Walker can think of — he sprays capital, gets access to deals, and has lumpy returns; “every time they’re down and you think they count them out… the AI stuff ends up working out for them.”
• The historical search that came up empty: Walker couldn’t think of another fraud or blow-up that held assets that would have returned the entire equity. Enron had pipelines and power plants — Walker believes Kinder Morgan had the pipeline plan and left to build the business — and spun out EOG (Enron Oil and Gas, a roughly $75B EV company) two or three years pre-bankruptcy. GGP was “one of the biggest special situations home runs of the global financial crisis,” but that was a liquidity bankruptcy with valuable malls, not a fraud.

2. AI as force multiplier: the village-singer analogy

• Walker oscillates “between my valley of despair” and excitement, and insists you can’t judge AI as it stands — the strawberry test fails today, but “I have trouble believing that the AI five years from now is going to be failing the strawberry test.” The question is what AI does three years out.
• His non-coder anxiety has faded: “Why do you need to learn to code to use AI? Why can’t AI do the coding?” He expects many AI interactions to be humans saying “hey AI, here’s my dream, help me build it.”
• The analogy carrying his optimism: the best singer in a 1500s European village could probably find work; today you must be “top 0.000001%,” and today’s stars sing, dance, and do everything. AI may amplify creative fields in the same way — his example is the viral video of someone AI-inserting himself into Game of Thrones to slap Joffrey and save Ned Stark; “he could never have created that video on his own as recently as 12 months ago.” The honest worry — what about top 5%, who might have had a village career in the 1500s but get nothing today? — resolves into a hedge: “I kind of think there’s just so much human creativity, everybody’s going to be able to carve a niche out for themselves.”

3. In a world of AI slop, deep expertise earns a premium

• Walker is himself producing slop — his “One Idea Per Day” Substack is AI-generated — and notes the recursive worry that AI trained on slop “law of averages everything down.” A college student can now write a basic research report in 30 minutes that reads “really effing good” to a layperson while an expert instantly spots the repetition and errors.
• The Knicks challenge example (they just won the Finals — “go Knicks”): AI might flag a first-quarter foul on KAT as “you’re 98% to win in this challenge,” but a subject-matter expert might save the challenge for a high-leverage moment. Walker’s example involves a lesser role player, Jose Alvarado, whom he describes as a Knicks bench player; NBA teams have two challenges at most, with the second available only after a successful first challenge, and are also constrained by timeouts and game context.
• The investing version: ABBVX reported blowout efficacy with safety signals, and a generalist or AI might conclude “this is a disaster, black-box warning, boom” — while Adam’s subject-matter expertise let him look at “the sample sizes… the base rates… the timing.” Hedged as hedged: “maybe Adam’s right, maybe Adam’s wrong, I don’t know” — and Walker discloses a position.

4. Second-hand hallucinations and the re-rating of legacy brands

• The KPMG story from the Financial Times: a big report on AI use in business, published around October and discussing companies including — Walker thinks — Shell, had a lot of hallucinated material. That echoes lawyers filing AI-written briefs with invented citations. The line Walker keeps returning to, quoted directly: “Big consultancies such as KPMG and EY are viewed as highly credible. So, their reliance on false information increases the risk of second-hand hallucinations.”
• His verification parable is Musk-era Twitter: when anyone could buy the checkmark, people could change their name to Eli Lilly and tweet “we’re making all of our insulin free” — and the checkmark made it believable.
• The thesis: brand power, diminished since the big-three-networks era by cable and then the infinite internet, may reverse — “the brands start actually sucking back up more equity.” A KPMG-branded report he’d pay for and rely on; an anonymous one he can’t. How far does it stretch? Maybe even to People or TMZ — “if I hear Taylor and Kelce are having a baby, I want to know it’s true” — and maybe ChatGPT starts paying trusted sources for citable access, because “ChatGPT wants people to trust the sources.” He flags this as speculation: “I don’t know.”

5. Time-horizon “edge” is usually underperformance in disguise

• The dinner debate: an investor claimed “my edge is time horizon,” and Walker admits “the me of 10 years ago probably would have said” the same — “and I’ve just come to view that as silly.” Version one — if everyone plays six months or a year, he plays three years; if everyone plays three, he plays five — “is literally saying, hey, I drag my Excel file over a column or two.” He’s skeptical that smart, competitive investors are leaving that money on the table purely because of constraints from a boss, employer, investors, or fund.
• Version two — diamond hands through drawdowns, with roots in Buffett’s “we’d rather buy a lumpy 15% return than a smooth 12%” — presupposes its own refutation: “if that’s the presup, why do you need to buy at 10 and go to five? Couldn’t you buy at five?” Genuine forced selling is “a lot rarer than I think people think.” The real case he discusses is a failed drug trial: a company once valued at $10B, with $1B of cash, can trade at $750M as holders sell; many holders are “there to play for literally the next cure for cancer,” not the cash. Dividend cuts and index kicks, by contrast, generally involve businesses where “things ain’t great over there” or whose stock simply isn’t working.
• The tell that undermines the argument for him: “most of the time when I hear someone say I have a longer time horizon… the unspoken part is, I’ve underperformed for the past 3 years, the past 5 years.” He’s seen letters claiming seven straight years of underperformance in a “bubble” market — “how much are you going to outperform in the eighth year to justify those 7 years?”
• His own discipline: a 3-year rule — a stock that’s done nothing for three years forces the mirror question, “what is the market seeing that I’m missing?” Honest accounting: “every now and then it’s caused me to miss a huge winner that would have just done great in year four,” but more often it gets him out of stocks that would be flat for another three years or deteriorate further. The pattern he keeps finding: a 2019 entry still being defended in 2026 — “it feels like that’s a busted thesis and you’re just not willing to admit it to yourself.”

6. Polymarket’s resolution problem — and its untapped reflexivity

• The case: MicroStrategy sold Bitcoin during the week of May 31 but filed the 8-K on June 1; the “will MicroStrategy sell Bitcoin by May 31” market resolved NO, and a student who bet a lot on yes saw the market resolve against him. Walker defends the rule in principle — his Diet Dr. Pepper example: you can’t leave a market open in case his memoir confirms it 20 years later — but stress-tests it: “what if MicroStrategy had filed that on August 15th when they reported earnings two and a half months later?”
• The deeper problem is persistent gray areas: did the US–Iran market resolve at Trump’s tweet, at Friday’s memorandum-of-understanding signing, or 30 days later? “Even if you try to define things up front, a lot of times the way things play out… you won’t be able to define.” Combined with reflexivity, he calls it “a pretty big risk to the business model in my opinion.”
• On reflexivity, he’s surprised the wars haven’t started: a candidate could slam a thin market with $5,000 of yes bets and email donors “I am the Polymarket favorite.” In a bigger hypothetical, Elon Musk, instead of an incremental $5 million in Ohio or wherever it was, could have bet Trump from 60 to 80 two weeks before the 2024 election — “if you were a swing donor, why would you donate to Kamala Harris? She’s clearly going to lose.” And people misread the numbers anyway: if these markets are functioning, 55% “should mean there’s a 45%… it’s almost a coin flip,” not a certainty.