Eric Ruleman

we wanted self-driving cars, but instead we got self-driving code

@RJCcapital do it

i am done shitposting time to lock in so I can finally hit 9 figures by end of the year

@brookeleblanc cute

It is a good hair day, Paul!

@gabriel1 same

i'd pay 90% tax for permanent sun

@jameygannon beautiful

shaken up la colombe latte out of a guinness glass 10/10 full apt reveal coming soon 🙈

@GENIC0N printing press pulled a fast one

turns out mass literacy was chiefly for the benefit of the machines

@bronzeageshawty incredible

Lifting in the morning, hot yoga in the evening, and then a long walk after. All powered by fruit, coffee, milk, and my ridiculous supplement stack.

@michaelsikand huuuge

NGL this a bigger flex than my watch or my car

@RJCcapital this is my quant

mega green on red day. only1rjc

@creatine_cycle finally the content we need

here is the full list of our SF restaurant picks: chinese: Sichuan Tasty Restaurant steak: Brazen Head dumplings: Dumpling House (Castro) burritos: Taqueria Buen Sabor tacos: Taqueria Vallarta on Folsom and 24th Italian: Seven Hills Izakaya: Fenikkusu or Kibatsu omakase: An (Japantown) and Doma Sushi (Bernal) thai: Prikhom vietnamese: Hai Ky Mi Gia (Tenderloin) nepalese: Timur (Sunset District)

trying out the new agentic @RobinhoodApp account - very easy to setup via Claude Code excited to see who performs better — me or the AI

@firesidealpha @GavinSBaker @tbpn great write-up!

Always a must-listen. Gavin Baker (@GavinSBaker of Atreides) broke down the SpaceX IPO, orbital compute, retail, China, and where AI value accrues, on @tbpn (~0:35 to 1:10 min mark). Here were the biggest takeaways: 1. The orbital-compute math is starting to pencil out. Gavin's version: a gigawatt costs about 60 billion to build on Earth, and 25 billion of that is the power and cooling you do not need in space. Once Starship is reusable he puts a launch at about 5 billion, so a gigawatt in orbit lands near 30 billion against 60 on the ground. The half he expects to keep inflating is the power and cooling, which only widens the gap over time. 2. The near-term SpaceX story is terrestrial compute, not Mars. Gavin is genuinely excited for asteroid mining and a city on the moon, but says the tangible drivers over the next year are how fast SpaceX can bring gigawatts online and how richly it monetizes them. He points to roughly 50 billion a gigawatt on the Google deal and, per the Altimeter work, 2 to 3 times neocloud pricing, and notes that in Jensen's words SpaceX energizes data centers faster than anyone. 3. SpaceX looks confident it can bring compute online fast. Gavin cites the Altimeter figure that SpaceX has ordered 20% of Nvidia's Rubin chips, with Groq's LPUs integrating over the next six to nine months. He reads the Anthropic deal as having an out and possibly running short-term, reflecting the training-versus-inference tension, and argues that being a token factory is more than enough for the next five to ten years, so SpaceX has no need to rebuild a full AWS unless it chooses to. 4. Cursor is the other variable he is watching. He flags Cursor's Composer 2.5 becoming roughly Pareto-dominant after a few weeks of reinforcement learning and fine-tuning, asks what happens when that approach meets a bigger base model, and finds it telling that Cursor is now in half the Fortune 500. 5. Retail keeps beating the pros who dismiss it. When people use "retail" as a pejorative, Gavin flips it back on them with "stupid is as stupid does," meaning you judge by results, not labels. The retail-favorite indices are up a lot in both 2024 and 2025, and by his read retail is outperforming the overwhelming majority of professional managers across private equity, venture, and public equity. He calls it the most powerful force in the market in his career, while drawing a hard line that this is nothing like the year 2000 bubble. 6. The lockup-overhang fear is overstated. Over 10,000 SpaceX employees bought on the IPO, and Gavin notes that anyone on the cap table has had a chance to sell every six months for the last decade. The people who wanted liquidity already took it, in his read, which is why he thinks the supply and demand picture from here is more interesting than the standard lockup analysis suggests. 7. Public markets are more patient than venture believes. Looking at how the market treated Tesla and Amazon through their heavy build-out years, Gavin argues the public market has a much greater tolerance for investment and a much longer time horizon than a lot of people in the venture ecosystem give it credit for. 8. His "token path" framework, and where CDNs sit in it. Value accrues to whatever sits in the flow of AI tokens. Cloudflare and Akamai are partly there, and he cites Akamai's roughly 1.8 billion dollar Anthropic deal plus a premium for low-latency delivery, since the lesson from Cerebras is that people pay for speed. But he notes CDNs touch less than 1% of all tokens consumed, maybe under 10 basis points, because most inference happens internally. Asked what is fully in the path, he joked: beachfront property, airplanes, and really nice cars. 9. The bottleneck trade is ending. For the last year, in Gavin's telling, the game was "having Claude run what's the next bottleneck." He thinks that trade is nearly done, pointing to Ajinomoto, the Japanese chip-input supplier the bottleneck crowd had bid up, which then chose not to raise prices. The next game, in his framing, is what holds enduring franchise value on the other side of the bottlenecks. 10. The market doubts Meta can monetize its own assets. Gavin increasingly values these companies on enterprise value to net property, plant, and equipment, a "high asset value, low obsolescence" lens. Meta's multiple there says the market has immense skepticism about its ability to monetize its asset base, and he thinks that skepticism is warranted given that the main thing management has offered so far is the idea of personal superintelligence. 11. The industry is behaving like the end game has arrived. Gavin reads the way players are moving as a sign the end game may be here sooner than expected. His take on OpenAI cutting Codex pricing is that coding tokens are so valuable for getting into the recursive self-improvement loop that everyone is racing for them, and he says Zuckerberg feels that acutely. 12. China's distillation edge is real but fragile. Chinese labs are world-class at industrial-scale distillation, and Gavin relays that it took only about 160,000 reasoning traces from o1 and o3 to reconstruct the original DeepSeek, run through endpoints across every available API. His warning is that all of it goes away the moment American labs stop releasing models at the frontier. 13. Sovereign AI will not reach the frontier. Every country will want a sovereign strategy for national defense, but Gavin's read is that for everyone outside the United States and China it means reinforcement learning on your language, culture, and values, run on the best open-source model inside your own data centers. Frontier-grade sovereign AI is the one outcome he does not see happening. 14. SpaceX may be the most important company of his lifetime. Gavin frames it as a once-in-a-career investment, possibly the most important company ever, and recommends seeing a launch in person, where "a lot of people cry." He notes he owned 15% of Nvidia and 10% of Tesla when each was sub-2-billion in market cap, and puts SpaceX in that same rare category.

@tenobrus very useful to hear the thought process for people who aren't always caught up on everything

i don't think this really needs much explanation but it's super out of character for me to be so terse so lemme ramble on anyway - both openai and anthropic are actively working on and hiring for custom chip programs - these *may* only be for inference, but i would find this unlikely, and we know eg TPUs are viable for both. even if they are inference may dominate soon - improved capabilities accelerating chip design and development seems *very* likely. if it's not already happening then they'll train for it - successful RSI will drive this much much faster - anthropic at least has proven themselves very willing to prevent external use of their models to compete with them - there's a pretty likely world where as RSI speeds up they use internal only models to massively accelerate chip dev and don't allow any external companies to use it for the same - nvidia does not manufacture chips themselves, their moat is their IP - they would plausibly remain a natural leader if they continued to have access to frontier intelligence, but as discussed this doesn't seem certain - so pretty quickly we find ourselves in a world where anthropic has no further demand for nvidia chips. this is obviously super desirable to them, nvidia charges a massive premium - but the actual manufacturing is much further outside their wheelhouse. TSMC or intel are basically the only plausible candidates

@Duderichy yes

my dream job is to sit in a room and work on hard problems and barely talk to anyone walk around a lot and go through a lot of notepads and paper

@MoneyPrinter0x @BigNoseCapital great reply

tbh that sector is mostly for copers/copium flow who missed memory supercycle, semis supercycle, gpu supercycle, compute supercycle, etc in other words, its useful to use it to track on mtf cross-sectorially what the ppl last to the party is doing. usually a leading factor/one of the leading indicators on 8th-9th inning environments such as ours. obviously this is under bayesian conditioning that we are in 8th-9th innings re: ai capex supercycle current wave, so its gd to enter this framework while being situationally aware of the base denominator event driven & its respective invals

interesting to play arm, qcom, neoclouds this coming week gd to monitor photonics for further fwd-looking regime changes

@mrexits so important

I think about Henry Kravis's ($12B) parting words to Ravi Gupta as he left to go to Instacart almost every day

@RJCcapital hot

when you tell her you’ve been long $SNDK since February 2025

@itsalasdairmann great overview 👍🏻

@pleometric @deepfates that would be cool!

@deepfates Me, I need to clean it up but I can open source it this week

Who was making brainrot Minecraft parkour videos for arxiv ML papers? somebody had a pipeline I think

@magetrades @RJCcapital accelerate

@eric_ruleman @RJCcapital Pause

if you think this rip is something you not ready for what’s coming

@magetrades congrats, great stocks

Grateful to the markets for an extraordinary year so far 🙏 Proud to be a shareholder in the AI infrastructure buildout. Concentrated in $NBIS $AAOI $MU $DRAM with $NBIS as my highest conviction hold. We’re early.