TL;DW: Anthropic's $2T IPO, Zuck's Manifesto, Nvidia's $500B Bet & Grok's Comeback

TL;DR

“All their pricing power depends on them remaining in the lead by six months.”
— David Sacks, on Anthropic's fragile moat

“This technology is too dangerous to centralize.”
— Gavin Baker, on the open-source case

“Nvidia is kind of becoming the central bank of AI.”
— David Sacks, on the $500B financing scheme

“Never count Elon out. He's frequently late. He's never wrong.”
— on Grok catching the frontier in six months

Sacks and guest investor Gavin Baker run the whole AI-money stack in one sitting. Anthropic is reportedly chasing a ~$2T October IPO (bigger than SpaceX's record) on a ~$100–120B run rate and 10x-a-year growth — but its premium rests entirely on staying ~6 months ahead, and 90%-cheaper open-source models are closing in. Zuckerberg's manifesto grabbed the open-source/abundance crown Altman dropped, reframing the debate as centralize vs. decentralize (the panel is firmly decentralize). Nvidia's $500B scheme turns GPUs into a financeable asset class with Nvidia as guarantor-and-royalty-taker — "the central bank of AI." And Grok 4.6 quietly hit the frontier at a fraction of the price, turning the "frontier duopoly" into a three-way race. Bonus rounds: Amazon's driver-subcontractor suits as a socialism flashpoint, and Silver Lake's Workday bid signaling SaaS may have bottomed.

Key takeaways

Anthropic's IPO is the industry's pace car. Quarterly numbers will become the single most-watched demand signal in AI — if the leader wobbles, everyone behind it feels it.
The moat is 6 months wide. The premium for frontier tokens only holds while they lead; regulation (Dario's "FAA for AI") would've erased that lead — so losing the DC fight may have saved the IPO.
2027 revenue bet: both hosts land ~$400–500B exit ARR, not the internal "$1T next year" talk — compute and energy physics bite first.
Open source is the real pressure. GLM-5.2 / Chinese labs at ~90% lower cost; corporate America adopts open source the same way it did the last 20 years. Frontier stays valuable because it can orchestrate cheaper models beneath it.
Centralize vs. decentralize is the actual axis — not open vs. closed. "Too dangerous to centralize" is the winning frame; JCal flags Zuck's pro-scraping stance as hypocritical given Meta's own graph lawsuits.
Nvidia is financializing compute. Residual-value guarantees + revenue share above a floor → capital-light "cloud," possibly the Fed of AI. The one real risk is overbuild ("dark GPUs"), which political headwinds ironically insure against.
Grok is the surprise #3 — Cursor's team + SpaceX operators got Elon to the frontier on price in six months; investors under-rate it inside the SpaceX story.

Anthropic's $2T IPO

Polymarket: ~80% odds Anthropic IPOs before 2027

FT reports a ~$2T target for an October IPO — topping SpaceX's $1.75T record. Gavin's read: the leak likely comes from bankers who lost the lead-left slot and want to embarrass the winner; if it prices at $2T, testing-the-waters went great and it could trade to $3T. The whole bull case has one load-bearing beam — the ~6-month lead that justifies the token premium. Sacks' irony: it's fortunate Dario never got his "FAA for AI" regulatory dream, because approval delays would've commoditized the lead. Side note the panel couldn't resist: Dario reportedly musing Anthropic "might be the only private company in the world" earned a hard "pride cometh before the fall."

Zuck's manifesto

Zuckerberg's manifesto: "The Future is for Everyone"

Zuckerberg's ~6,500-word essay picked up the open-source + abundance banner Altman dropped going for-profit-private: open models, a free agent and tutor for everyone, "superintelligence is invention, not automation." The frame the hosts loved — the AI fight is centralize vs. decentralize, and history favors distribution (their right-to-bear-arms analogy: "my own AI"). The distillation line: "All AI models are derived from human knowledge"

JCal's catch: Zuck's "you can learn from anything you can observe" is hypocritical next to Meta suing hundreds of startups for indexing its social graph — but credit for going direct in his own voice on X.

Nvidia's $500B compute financing

Jensen's X post: "NVIDIA AI Factory Compute Is Becoming an Investable Asset Class"

Nvidia + Goldman/BlackRock/KKR/Apollo/Blackstone turn GPUs into an asset class you finance like aircraft or mortgages: borrow → buy Nvidia systems → rent compute → repay from cash flow. Nvidia plays matchmaker, adds a residual-value guarantee (bearing ~25% of the risk) and takes revenue share above a floor. Not circular (per the panel) — real asset-backed lending on real cash flows, with 9-year GPU lifespans cited. The genuine risk is an overbuild → "dark GPUs" echo of dark fiber; the anti-data-center political panic ironically guards against oversupply.

Grok's comeback

The panel's chart: "Grok 4.6 is Pareto Dominant"

Grok 4.6 hit the frontier at a fraction of the price — Gavin retires his "duopoly" call; it's a three-way race now, with independent-ish evals (Databricks, Mercor, DHH) positive and Grok 4.7 due in weeks. How Elon caught up in six months: bought Cursor's team, imported SpaceX operators. The tension worth watching — Elon is simultaneously Anthropic's compute landlord and its competitor, with a 90-day clause to claw compute back: a call option (become a frontier lab) plus a put option (sell compute near spot).

Lightning round

Amazon DSP lawsuits (NJ AG + Mamdani): subcontracted drivers as a liability shield and a socialism flashpoint. JCal: make them employees (~+$0.25/pkg), get ahead of it. Sacks: freedom of contract; forcing it adds ~~$5.20/package (~~$664/household/yr). Compromise everyone reached: a "Mamdani fee" line-item.
Silver Lake ~ buying Workday (+17%): first big PE software bid in 18 months — SaaS may have bottomed; open source commoditizing the model layer is read as a godsend for app software (the Bending Spoons "cut 80%, AI-first the rest" playbook).

This is my own summary and structure; quotes are short, attributed excerpts from the episode for commentary. Watch the full thing if you want the banter — this is the signal.