Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
SpaceX S-1, Anthropic Revenue Booms, OpenAI Cracks Erdős Problem | Diet TBPN
TL;DR
SpaceX’s S-1 reframes the company as an AI giant, not just a rocket company — The filing shows 2025 AI capital spend running at roughly 3x space spend, with a $28.5 trillion TAM slide and an $80 billion+ IPO target that could top Saudi Aramco’s $26 billion debut.
Anthropic’s growth looks absurdly fast and suddenly profitable — The Wall Street Journal reports Q2 revenue could hit $10.9 billion, up 130% quarter-over-quarter, with first operating profit, while Tom Brown said Anthropic is scaling GB200 capacity on SpaceX’s Colossus 2 throughout June.
SpaceX may have stumbled into a monster ‘neocloud’ business through Anthropic — The hosts say Anthropic is ramping toward roughly $15 billion a year of spend, a huge jump against SpaceX’s $18.67 billion 2024 revenue, and likely at premium pricing for compute.
OpenAI appears to have cleared a real math milestone, not a benchmark gimmick — An internal general-purpose model reportedly solved Erdős problem #90 with a genuinely new proof, not scaffolding or paper retrieval, and mathematicians including Terrence Tao called it a meaningful breakthrough.
The anti-AI ‘this will never make money’ crowd is losing the plot — The segment contrasts Gary Marcus calling GenAI a scam in 2026 with Anthropic’s profit trajectory, OpenAI’s valuation gains, and evidence that AI is already replacing $1,000 workflows with results costing a fraction as much.
Nvidia’s earnings got overshadowed by AI news saturation — Even with strong results and an $80 billion share buyback authorization, the hosts treat Nvidia as almost secondary because the week’s bigger story is that every major tech narrative now collapses into AI infrastructure and demand.
The Breakdown
SpaceX’s IPO finally gets real
The show opens on the Wall Street Journal’s biggest story: SpaceX has published its investor prospectus and kicked off what could be an $80 billion-plus IPO as soon as June 12. The hosts linger on how wild the company’s arc is — from “almost went out of business” startup to a 22,000-person giant with businesses spanning rockets, Starlink, and now AI.
The S-1 reads like sci-fi — and that’s the point
They clearly had fun with the filing, calling it one of the most enjoyable S-1s in years, complete with cinematic photos and a “best TAM slide ever.” The standout claim: a $28.5 trillion addressable market, including $26.5 trillion in AI, which sparks some skepticism on pieces like X-driven advertising but also reinforces that SpaceX is pitching itself as a civilization-scale company.
Elon’s newest play: turn SpaceX into an AI infrastructure monster
The hosts zero in on the real surprise in the filing: AI capex is reportedly 3x space capex, prompting one quip that SpaceX is now “an AI company with some rockets.” They connect that to a broader coordinated narrative shift — space data centers, the xAI/Colossus buildout, the Cursor rumor, and now a détente with Anthropic — as evidence that Musk is once again “making plays” faster than everyone else.
Anthropic’s numbers hit the ‘mind-blowing’ zone
Then comes the Wall Street Journal scoop: Anthropic’s Q2 revenue is expected to reach $10.9 billion, up 130% from the prior quarter, with operating profit. Tom Brown’s note about expanding with SpaceX’s Colossus 2 turns the story from abstract growth into a concrete infrastructure partnership, and the hosts hammer home how shocking it is that a company spending over $1 billion a month on compute can also be the best rebuttal to the “AI will never be profitable” camp.
Why AI skepticism now feels stuck in 2024
They spend a few minutes roasting the habit of lumping AI in with NFTs, the metaverse, and other hype cycles that never delivered a universally jaw-dropping product moment. Their point is simple: unlike those fads, AI already gives ordinary users startlingly useful experiences on demand, which makes blanket “it’s all a scam” takes — especially in 2026 — look increasingly detached from reality.
OpenAI solves an Erdős problem, and the math people actually care
The most technical segment is Tyler’s walkthrough of Erdős problem #90, a combinatorial geometry question about the maximum number of unit-distance pairs among n points in the plane. He explains why this result matters: OpenAI’s internal general-purpose model didn’t just retrieve an old paper or use a special scaffold, it produced a novel 18-page proof showing the original conjecture fails for infinitely many n — a result serious mathematicians are treating as a real step change.
Nvidia, random internet detours, and a goofy finish
Near the end, they note that Nvidia posted strong results and an $80 billion buyback, but the story almost feels buried under the sheer volume of AI news. The final stretch turns more playful — Messi’s drink brand Moss shutting down, Vanguard’s terrible UI as an accidental investor feature, a fake Catan clip about building a data center, Matthew Ball joining Xbox, and the White House putting $1 billion into quantum companies — before the hosts sign off in full Friday-energy mode.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
