SpaceX's Financials Leaked: Is it Worth $2TN | Meta Debuts Muse Spark: Are They Back in the AI Race?

Key points

• Mythos, Anthropic's new cyber model, found thousands of zero-days autonomously and was held back from public release. Rory's analogy: it's not that older models can't find these — it's that Mythos does it at machine-gun pace agentically, across whole codebases, for about $20k of credits per run. Rifle vs machine gun from WWI.
• Immediate consequence: every AI-built app with weak auth (MyFitnessPal/Calai was breached 2 days post-launch — 3.2m records, no Firebase auth) becomes trivially exploitable. Expected transition phase of worse security before it gets better; security vendors who step up win, ones that don't die.
• Jason Lemkin's public break with Dario: 'I can't open the Strait of Hormuz myself. Enough already.' Framing him as the boy who cries wolf — endless apocalyptic marketing is fatiguing and net un-inspiring, even if sincere.
• Rory's counter: grandiosity is a rallying cry, not a prediction. Elon at IPO said we're going to Mars and then said 'we're going to the Moon' in the S-1; the vision sustained the mission. Dario's doomer rhetoric + mission clarity is why Anthropic has zero churn and a $30B revenue line.
• Trainium share: Anthropic trained parts of Mythos on Amazon's Trainium chips. Amazon's Trainium business is ~$20B ARR, triple-digit growth, largely internal (Amazon buying its own chips instead of Nvidia). Nvidia's fortress has a 10% dent — material, not existential.
• **The 60%-solution SaaS death spiral (Jason's thesis for this week):** every public SaaS vendor is shipping 'fine' in-product AI agents that are ~60% as good as standalone tools like Claude Code, Replit, Lovable. Problem: you cannot charge extra for a 60% solution. Customers will use it for free but won't pay $60k more for it.
• If you can't charge separately for your agent you don't get revenue reacceleration. Without reacceleration, SaaS companies move into the 'deep value play' bucket at ~8-11x forward P/E — the IBM trajectory. ServiceNow, Salesforce etc. are implicated.
• Public-markets illusion: you can't currently buy the growth story. The trade is 'sell SaaS, buy semis' because Anthropic/OpenAI aren't public yet. Once 5-6 AI-natives IPO, the relative-value math becomes real and some SaaS incumbents may re-rate to their cash-flow value plus optionality.
• Salesforce's $25B buyback is critiqued: financial engineering on a spreadsheet is worth little against the product problem, and the flexibility of that cash in a possible 2027-ish bust (when broke AI-natives need capital) would have been the smarter play.
• SpaceX at $2TN IPO: 'the most expensive IPO at scale of all time.' Elon discount rate is 0, Elon probability-of-failure rate is 0 to get there — the episode is sceptical.
• Meta's Muse Spark is Alex Wang's first model out of the new Superintelligence Labs. Discussion whether it signals Meta is meaningfully back. Short version on the show: clarity + product moat still favour Anthropic (enterprise) and OpenAI (consumer) for now.

Notable quotes

Themes

• Cyber's machine-gun era
• Dario fatigue and the inspiration deficit
• The 60% SaaS death spiral
• SpaceX at $2TN — what the Elon discount rate prices in