Ex-Palantir Analyst On The Inner Workings Of Intelligence & Markets In A World Of Socialism

Summary

• Alex Good (Wharton → Citi FX → Palantir → Balyasny → founder of Perpetua, now the Post Fiat L1) built his edge running the actual ads for stocks he traded: EA could acquire a $50-game buyer for $4 while Booking spent “five out of their $6 of profits” on ads yet traded at a higher multiple — long EA / short Booking. The strategy “only got nonlinear” with meme stocks, when Tesla ad clicks all came from the search “Elon Musk” and he realized the alpha was no longer how cheap it is to sell a car, but how cheap it is to sell the stock.
• His “goldfish theory” is the tradeable core: a dandelion-tea page converted 8% of visitors for years — never 6, never 10 — so predicting people’s affinity to buy assets is far more predictable than predicting the future. The model breaks on fragmentation and trust: STRC launching next to MSTR splits the vortex, “Elon dropping SpaceX is unambiguously bearish for Tesla stock,” a likely Anduril IPO would be bad for Palantir — and Bitcoin is advantaged in this framework because it has no competitor while every alt-L1 fights for the same pie.
• Bearish Palantir; warns 100x sales brings investor headwinds: its pitch — that LLMs need a Palantir ontology to avoid hallucinating on enterprise data — “I just know that’s not true,” and the need for ontologies has “decreased linearly” as models improve. Meanwhile OpenAI and Anthropic now field their own forward-deployed engineers: “there are three things in the store instead of one.”
• AI unemployment kicks in within ~3 months, he says — not because it’s happened (we’re at full employment, Accenture still has 800k staff) but because the market is forcing it: Figma down 85%, Accenture down 26% in a day, and software now trades at half the multiple of commodity companies. The trade: firms with credible AI turnaround margins, plus bottleneck victims like Nintendo (down ~60% y/y on memory costs) and Lindy-IP owners — post the Anthropic/Libgen settlement, “there’s a world where all future AI Marios pay Nintendo.”
• The Blackprint thesis: AI “has stopped being a technological phenomenon and started becoming a political phenomenon.” Nvidia’s 50,000 employees are worth more than the entire Russell 2000, which employs millions — “people won’t vote for a right-wing accelerationist system which doesn’t benefit them economically,” and laid-off Accenture types will be very effective Bernie/DSA organizers. Trump already regulating Anthropic normalizes blocking model releases on both wings.
• Contra Andrew Kang, he’s fully against the humanoid-robotics trade (Trump banned port automation; he expects governments not to rug blue-collar workers) and bearish AI-biotech on national-security tail risk — “all it takes is one guy running an unsanctioned experiment on a bat” and it’s banned. What survives, per the 1945 nuclear analogy: computers, digital economies, entertainment. “How do you pay for the data centers? You cook everyone’s brains.”
• Endgame: productivity is running 6% annualized, not the 10% Microsoft’s CEO promised, so AI doesn’t rescue WWII-level sovereign debt — “then it’s basically a sovereign margin call, and that’s why I’m in crypto.” Robinhood, Interactive Brokers, and Hyperliquid are the kingmakers as capital recycles into the attention economy. Avi’s close: the conversation made him more bullish Bitcoin.

Deep dive

1. The edge was running the ads yourself

• Good’s path was “economic necessity” all the way down: Citi FX, then Palantir big data, then Balyasny hired him for a dataset that got cut off by SWIFT problems at Stan Chart and HSBC — so he invented a strategy trading advertising stocks while literally running the ads. Risk kept calling about 15% of his book in Amazon; his answer: “everyone is saying the retail business is worth zero. I know it’s worth a cajillion dollars because I’m running all these ads and they’re breaking even.”
• The canonical pair trade: Booking spent ~70% of its margin on ads while EA could sell a $50 video game for $4 of acquisition cost — “company B is spending five out of their $6 of profits, and they’re trading at a way higher multiple.” Long EA, short Booking, with a live kill-switch: if Call of Duty ads got expensive that quarter, EA wasn’t cheap anymore.
• The turn: “The alpha I’m measuring with Tesla is not how cheap it is to sell a car. It’s how cheap it is to sell the stock” — every Tesla ad click came from the search “Elon Musk.” That’s how he got into XRP, Cardano, Binance affiliate programs, and ultimately Twitter itself, needing 10,000 followers to keep his affiliate deals: “I need to start saying things on the internet.”

2. Goldfish theory — attention is more predictable than the future

• The founding observation: an Amazon dandelion-tea page converted exactly 8% of visitors for years — never 6, never 10. “You don’t know in advance who is going to be the vortex of attention, but once you know that they are, it’s very predictable that they’re going to continue to be.” He’d rather estimate that 40% of people who see GameStop’s Pokémon-card news buy the stock than predict macro: “I don’t like predicting the future when it comes to trading.”
• Jonah’s pushback — where does the model break? Good’s answer: fragmenting liquidity and eroding trust. Launch STRC next to MSTR and you split attention into two comparable things (Saylor is “a great real-time example”); losing trust is “his star rating going down” — the tea’s conversion fell when reviews slipped from 4.5 to 3.8 stars. “Elon dropping SpaceX is unambiguously bearish for Tesla stock”; a likely Anduril IPO would be bad for Palantir.
• The crypto corollary: Bitcoin has no competitors — no other asset has successfully argued fixed-supply store-of-value — while Ethereum, Solana, and every other L1 fight for the same pie. “That’s why altcoins have such a hard time.”

3. Inside Palantir — and why he’s bearish now

• He was among the first users of Palantir Finance (built with Thiel and Bridgewater), turning credit-card and SWIFT data into macro signals. Every deployment had two sides: compliance (“this guy is selling barrels of oil at $130 and the price is $80 — it’s probably transfer pricing”) and the revenue cherry on top. The ontology concept came from targeting: “How do I know that I’m killing the right terrorist?… How do you even know that Osama bin Laden is Osama bin Laden?” — license plate, associates, bank accounts.
• The bear case: Palantir now claims LLMs need that ontology layer to work on enterprise data, and “I just know that’s not true — you can point an AI at your codebase and just be like, yo, figure this out.” As models improve, “the need for ontologies has decreased linearly”; the labs themselves say prompting matters less than ever.
• Plus attention fragmentation in his own framework: Karp publicly mocked Dario Amodei’s 10%-unemployment call (“you might have a high EQ, but you’re actually retarded”), while OpenAI and Anthropic (with Goldman) now deploy their own forward-deployed engineers. “Now there are three things in the store instead of one. If you’re trading at 100 times sales, you’re going to face investor headwinds.”

4. Data is the new drop shipping — but don’t sell it, monetize it

• With LLMs, “the ability to turn raw data into structured data has never been cheaper,” so raw data’s value is rising — “the new drop shipping is data acquisition; everyone is starting a data firm to sell to labs.”
• His contrarian twist: “if you actually think a data source is valuable, you shouldn’t sell it, you should monetize it” — generate the richest possible dataset from pseudonymous crypto speculators and monetize internally rather than selling to a lab. The origin story is his disappointment with Morad’s Popcat Telegram group: “it wasn’t funny… there were not people colluding usefully” — the thesis (digital cultures as identity) was right, “but you didn’t take it far enough.” His bar: a voluntary command-and-control architecture with an AI at the center that he wouldn’t feel silly joining.

5. The layoffs haven’t happened — which is exactly why they’re coming

• He grants the disconnect: Amodei and Altman preach existential job loss while “the employment report comes out and you’re like, we’re at full employment.” Figma is down 85% with headcount up; Accenture down 60% with 800,000 employees and no firings. “Have they lost their jobs yet? No, definitely not.”
• But the market is forcing the issue — Accenture’s 26% one-day drop “woke management up real fast,” and the market keeps crushing egregious over-hirers like Salesforce. His call: unemployment starts kicking within three months, because current models — “Fable and GLM 5.2 can actually deliver a good experience” — are coming for the laptop jobs first.
• The trades that fall out: software with credible AI-turnaround margins now sits at “bottom of the barrel multiples” — “commodity companies right now are oftentimes twice as expensive as software companies.” Bottleneck victims bounce if memory breaks (Nintendo couldn’t launch consoles because it couldn’t afford memory, stock down ~60% y/y; etched-style ASICs with the model baked in could kill the bottleneck). And after the Anthropic/LibGen settlement — you can extract “97% of Harry Potter” from ChatGPT — canonical, Lindy IP could get paid: “there’s a world where all future AI Marios pay Nintendo,” plus Games Workshop, Disney, Hasbro’s D&D, Star Wars, all at radically different valuations.

6. The Blackprint: AI is now a political phenomenon, and socialism is the response

• The core thesis: “people won’t vote for a right-wing accelerationist system which doesn’t benefit them economically.” The math: Nvidia’s ~50,000 employees are worth more single-handedly than the entire Russell 2000, which employs millions of voters. The market assumes “Trump the stand all the way… the right wing will keep winning for some reason. I’m like, no.”
• The laid-off Accenture employee “is going to be quite effective at coordinating grassroots Bernie Sanders votes — those are competent people.” Avi adds the DSA is already installing increasingly competent socialist candidates, with populism rising on the right too — horseshoe theory, different constituencies, same subsidies.
• The gloves are already off: “Trump went in and regulated Anthropic… started the process of the political system blocking major model releases.” That would have been “a Biden move” before; now it’s normalized on both wings. Watch the midterms as the checkpoint that validates the worldview.

7. What gets banned: robots and AI biotech, not chatbots

• Contra Andrew Kang’s generalized-robotics vision (construction, homes, assembly lines within 5 years): “Trump banned port automation — even port automation was a no-no.” Governments rugging blue-collar workers via humanoid robots, “given the history of self-driving, is laughable” — someone attacks a Waymo and they’re pulled off the road for a month; it’ll be worse with robots. Avi’s pushback — doesn’t that put us behind China? — gets a flat: “I don’t think that’s how you win elections. You win elections by telling people you’re going to keep your job.”
• He’s also against the consensus long-biotech trade (“what if Anthropic cures cancer?”): the national-security tail risk post-Wuhan means “all it takes is one guy running an unsanctioned experiment on a bat… the second you have one event, just one event, it’ll be banned.” The tell: “people are already worried about Fable debugging your codebase — which is why they pulled Fable. Imagine if Fable is creating a new peptide.”

8. Not the end of the bubble — a 1945 settlement, then a sovereign margin call

• He explicitly rejects the “AI bubble is ending” read: entertainment hasn’t kicked yet. A personalized AI video costs ~$55 today and isn’t good; at $2 “in two years, basically,” you get mass personalized video, compelling game NPCs, and better feed algorithms as tokens-per-second grows exponentially. His analogy: nuclear in 1945 — weapons banned, certain physics banned, but “computers are allowed, digital economies are allowed, entertainment is allowed… How do you pay for the data centers? You cook everyone’s brains.” AI makes society more like it already is.
• The macro sting: productivity is ~6% annualized this quarter and 2.8% year-on-year — “Microsoft’s CEO said we’d see 10%… that’s a lot different.” People aren’t generating more work, they’re generating more capital recycled into the attention economy, which is why Robinhood, Interactive Brokers, and Hyperliquid are kingmakers. He’s equally skeptical of SpaceX’s “ungodly multiple” premised on Mars or space data centers by 2029: “It’s easier to underwrite the government cracking down on things.”
• The close: if AI productivity doesn’t bail out WWII-level government spending, “it’s basically a sovereign margin call — and that’s why I’m in crypto.” Avi, self-described Bitcoin bear coming in: “this entire talk has maybe actually made me a little bit more bullish on Bitcoin.”