Anthropic's $2T IPO, Zuckerberg's open-source manifesto, and Nvidia's $500B GPU-finance play
TL;DR
Anthropic is reportedly targeting a $2 trillion IPO in October on a $100–120 billion annualized revenue run rate, a 10x year-over-year ramp, sparking a debate over whether open-source models will erode its pricing power. The crew dissects Zuckerberg's 6,500-word manifesto arguing AI is too dangerous to centralize, directly opposing Anthropic's effective-altruist stance. Nvidia's $500B GPU-financing partnership with Goldman and BlackRock treats compute as an asset-backed class, with Anthropic's quarterly earnings called the industry's pace car. They also weigh Amazon's DSP subcontractor lawsuits in New York and New Jersey, and xAI's Grok 4.6 and Grokbot catching the frontier.
Chapters
-
0:00 Policy and regulation
Rahm Emanuel and the Democratic civil war
The crew welcomes guest Gavin Baker, still riding the high of SpaceX's record $1.75T IPO, and David Sacks back from Washington. They recap the All-In Interview episode with Rahm Emanuel, who came in 'full guns blazing' and hammered the Democratic Party's socialist wing. Sacks notes Emanuel — the Obama/Clinton-era throwback, former Chicago mayor and ambassador to Japan — has no chance among the DSA Democrats, drawing a parallel to the GOP's internal civil war. Friedberg praises Emanuel's credentials and his hard line that democratic socialist policies are a road to nowhere, hoping moderates in both parties can still win.
-
2:36 AI and tech
Anthropic's $2T IPO and revenue ramp
Breaking news: the FT reports Anthropic is targeting a $2 trillion IPO in October, which would shatter SpaceX's $1.75T record, with Polymarket at 80% odds of a listing this year. Anthropic is on pace to end 2026 at a $100–120 billion annualized run rate, roughly 10x year-over-year growth for the third straight year, with revenue up 10x in the last year alone. Gavin Baker, who says he is not a shareholder, notes Elon Musk grew comfortable renting Colossus compute to Dario Amodei after his 'evil detector' cleared him. Baker is skeptical of the $2T figure, attributing leaks to bankers jockeying for lead-left, and expects smart pricing to absorb the lockup. Sacks questions whether 10x growth can continue given TAM, compute, and energy constraints, noting it would imply $1 trillion ARR next year.
-
12:22 AI and tech
Can AI demand outrun physical limits?
Gavin Baker walks through the demand and supply math for Anthropic reaching $1 trillion in revenue. Demand: a $25–65 trillion knowledge-work TAM before robotics, with labor substitution so far reading as accelerating growth — there are more coder job openings today than a year ago. Supply: data-center buildout is 'atoms, not bits,' constrained by turbine blades, energy, and regulation, but American capitalism responds with 24-hour shifts, repurposed jet-engine turbines, and capacity expansion at Caterpillar, Cummins, GE Vernova, and Siemens. The crew attacks the media narrative that data centers drain water and raise electricity costs as a hoax, while Texas Governor Abbott's new energy audit and the moral panic itself act as brakes against an overbuild.
-
24:28 AI and tech
Zuck's manifesto: open versus centralized AI
Jason predicts Anthropic's growth will slow to a triple next year as corporate America adopts 90%-cheaper open-source models like GLM2, following the open-source adoption pattern of the past two decades. Gavin Baker endorses open source as guaranteeing 'diversity of AIs.' The crew dissects Zuckerberg's 6,500-word manifesto, which argues superintelligence is invention, not automation, that everyone should get an agent, and that concentrating AI power is historically dangerous. Sacks contrasts the EA/Anthropic view — that AI is too dangerous to distribute and needs enlightened control, per Thomas Sowell's 'vision of the anointed' — with the Zuckerberg/Musk/Jensen view that AI is too dangerous to centralize.
-
39:35 AI and tech
Decentralized AI debate and going direct
Jason credits Zuckerberg's job-displacement framing — individual capability gains may match automation — but flags hypocrisy: Meta sued countless startups for scraping its social graph while claiming the right to learn from all human knowledge. Sacks argues Dario's regulatory dream of an FAA-style approval for AI releases would evaporate Anthropic's six-month pricing lead, though a premium frontier tier (Apple vs Android) can survive. The crew celebrates Zuckerberg's 'Jeremy Strong era' and the going-direct movement on X, with Gavin Baker warning that CEOs without their own voice risk being defined by others. Sacks forecasts Anthropic exits 2027 at $400–500B ARR, under Jason's $500B line; Gavin Baker predicts 4–5x growth to around 500 billion.
-
58:19 Markets
Nvidia's $500B GPU financing plan
Nvidia partners with Goldman Sachs, BlackRock, KKR, Apollo, and others to raise $500 billion for AI compute, treating GPUs as a financable income-producing asset class — acting as matchmaker between borrowers and lenders, offering residual-value guarantees, and taking revenue shares above floors. Gavin Baker says elite asset managers validate GPU compute as a financeable asset, and the architectural variety of Chinese open-source models (Qwen, Kimi, DeepSeek, GLM) supports flexible GPU demand. Sachs raises 'dark GPU' overbuild risk, but political headwinds and Anthropic/OpenAI as the 'dead man switch' temper the buildout. Baker likens Nvidia to the Federal Reserve of AI and calls Anthropic's quarterly earnings the industry's pace car for demand signals.
-
1:14:25 Policy and regulation
Amazon DSP lawsuits over subcontractor model
New Jersey's attorney general and New York City's mayor are pushing to force Amazon to hire its delivery drivers directly, arguing the Delivery Service Partner (DSP) subcontracting model shields Amazon from liability, shifts worker benefits onto taxpayers, and pressures drivers with tragic results. Sacks defends freedom of contract, saying the DSP model enables rapid scaling and entrepreneurship, and warns the change could cost New York households $664 per year — about $5.20 per package — and threaten thousands of jobs. Calacanis pushes back hard, arguing the DSP structure was created to benefit Amazon, and urges Andy Jassy to proactively hire drivers at 25 cents more per delivery, as Howard Schultz did at Starbucks, to self-correct capitalism before democratic socialists win the political argument.
-
1:27:24 AI and tech
Grok 4.6 and Grokbot catch the frontier
Gavin Baker reviews xAI's releases: Grok 4.6 and Grokbot. Benchmark graphs show Grok 4.6 priced at or near the frontier — higher quality at lower cost than the gold-standard frontier model on Cursor and Databricks evals, with DHH praising it publicly. Baker credits Elon Musk's six-month catch-up to buying Cursor and reshuffling xAI management with 'killers' from SpaceX; a larger, more capable Grok 4.7 is expected within weeks. Sacks frames Musk's massive compute clusters as a call option on frontier-model status with a put option of selling compute to frontier labs at near-spot prices under a 90-day cancellation clause. Baker notes investors overlook Grok when valuing SpaceX, despite Anthropic's existence proof of going from $1B to $50B revenue fast.
-
1:35:09 Deals and companies
Silverlake-Workday bid and software revival
News breaks during the show that Silverlake may acquire Workday, sending the stock up 17%. Gavin Baker calls it a significant moment for software, which has been in a 12–18 month death spiral; he argues the rise of open source is a godsend for the American software industry and likely why private equity is returning — Workday fell from roughly $300 to $100 as a handful of frontier labs consolidated power at the model layer. The crew compares the dynamic to Bending Spoons, which slashes headcount and runs AI-first. The episode wraps with Gavin Baker tied with Brad Gerstner for fifth-bestie and plans for a steak dinner.
Key takeaways
- Anthropic is on pace to end 2026 at a $100–120 billion annualized revenue run rate — roughly 10x year-over-year for the third straight year, per FT reporting and the hosts.
- Sacks and Gavin Baker argue AI is 'too dangerous to centralize,' the direct opposite of Anthropic's effective-altruist framing that it is too dangerous to distribute.
- Zuckerberg's manifesto claims concentrating AI power has historically never led to safe or positive outcomes, and reframes doomerism as a self-serving argument for control.
- Nvidia's $500B GPU financing plan with Goldman and BlackRock treats GPUs like aircraft — asset-backed and financeable with residual-value guarantees rather than pure balance-sheet capex.
- Open-source models like GLM2 deliver comparable capability at roughly 90% lower cost than Claude Opus, which Calacanis says will be hard for corporate America to ignore.
- Amazon's DSP subcontractor model shifts liability and worker benefits onto taxpayers, and Calacanis predicts New York and New Jersey will win their lawsuits against it.
- Grok 4.6 is priced at the frontier on Cursor and Databricks evals, and Gavin Baker credits Musk's six-month catch-up to the Cursor acquisition and a SpaceX-led management reshuffle.
Quotes
This technology is too dangerous to centralize.
You don't trust AI Jesus.
I want the right to have my own AI.
Mentioned companies and people
Companies
- Anthropic
- Nvidia
- SpaceX
- OpenAI
- Meta
- xAI
- Amazon
- Workday
- Silverlake
- Goldman Sachs
- BlackRock
- Cursor
- Databricks
- TSMC
- Uber
- Palantir
People
- Dario Amodei
- Mark Zuckerberg
- Elon Musk
- Jensen Huang
- Sam Altman
- Rahm Emanuel
- Andy Jassy
- Howard Schultz
- Gavin Baker
- David Sacks
- Brad Gerstner
Tickers
- NVDA
- AMZN
- META
- MSFT
- GOOGL
- WORK
- UBER
Topics
- AI IPO
- open-source AI
- AI safety
- GPU financing
- data-center buildout
- compute demand
- inference costs
- labor regulation
- AI model competition
- capital expenditure
Predictions
-
Anthropic will exit 2027 at $400–500 billion of annualized revenue.
by end of 2027
-
Zuckerberg will deliver the best open-source model within a year.
next 12 months
-
New York and New Jersey will win the lawsuits against Amazon's DSP model.
unspecified