Just finished the latest All-In Podcast episode (280). A few big names discussed the SpaceX IPO playbook, the IPO progress of OpenAI/Anthropic, the real returns behind AI Token “burning money,” and the wild data around the launch of Trump’s account—there’s a ton of information. Here’s a整理一下 summary.

SpaceX’s IPO this time set a template for future mega-companies going public
With a valuation of $1.75 trillion, SpaceX raised $75 billion. After listing, it briefly surged to $200 per share, and is currently stable around $150. That implies roughly a $2 trillion market cap and $35 billion in forward revenue. It’s now the world’s seventh-largest listed company. The consensus is that SpaceX’s phased unlocking combined with early index inclusion effectively provides a textbook-level model for these future ultra-large unicorn IPOs.

Anthropic has reportedly secretly filed for an IPO. There are claims its annualized revenue is pushing toward $100 billion. If it truly lists, its market cap could reach $3 trillion. On OpenAI’s side, revenue has rebounded to about $70 billion. GPT-6 is said to be released within 30 days, but due to a more complex company structure overhaul—and because it previously burned money at a fast pace—its IPO timeline may be slower than Anthropic’s. Brad said their firm, as a large institution, would place large orders for both companies’ IPOs even at a $3 trillion valuation. The reason: these firms’ revenue compound growth rates over the coming years can still exceed 30%.

AI Token consumption is exploding, but the real returns at the enterprise level are being questioned
The discussion was pretty sharp. Chamath mentioned that their CTO’s report said Token spending doubles every 45 days, but the downstream productivity gains are only about 5%. Even harsher: in the S&P 500 (excluding Nvidia), of the per-share earnings growth, the part truly driven by AI is only 0% to 2%. If companies can’t show that Token spending actually produces investment returns, this kind of spending frenzy may face a round of correction.

That said, a few companies have offered smarter response examples. Uber already has 99% of engineers using AI to help write code, and it even set up “agent teams” that dive into departments like HR, legal, and marketing—calculating costs and efficiency in a more fine-grained way. DoorDash’s CTO, meanwhile, came up with a “model routing” strategy: push simple, low-level tasks to cheaper open-source models, and only send the hardest core work to Anthropic’s top-tier model—successfully bringing costs down.

The route battle between closed-source frontier models and open-source/sovereign AI is getting increasingly intense
Brad insists the advantage of closed-source top models still exists: economic value is still highly concentrated in OpenAI and Anthropic as the two leading labs, and for enterprises, paying for a slightly more expensive top model to replace a $200/hour senior engineer or consultant—spending an extra dozen dollars—is rational. But after Chamath joined the UN AI committee, what he saw is another line: countries like the UAE, Saudi Arabia, and Japan don’t want to hand over the lifeblood of their technologies completely to closed-source US models. They’re racing to build autonomous, controllable open-source “sovereign AI” stacks. At the same time, Zuckerberg isn’t idle: Meta just released the Muse Spark 1.1 model, targeting the same quality but at only 1% of the cost—directly firing a price war to break down the closed-source models’ high-price barrier using open source and low pricing.

China is also said to be considering limiting top models from going overseas
There are reports that China’s regulators are meeting behind closed doors with companies like Qwen, ByteDance, and Zhipu—considering restricting overseas access permissions for both top open-source and closed-source models. The goal is to prevent technology leakage and misuse by the US side. Sacks’ interpretation is that this is a very typical “catch-up phase relies on open source to build an ecosystem; once it catches up or gets close to the frontier, immediately pivot to closed source” strategy—similar to the route shift OpenAI itself took back then. On the US side (including the White House and Treasury), the core consensus is “do everything possible to maintain AI leadership over China.” Since some China-side models were found to contain大量蒸馏 of outputs from US models, the US is also rolling out anti-distillation countermeasures simultaneously.

Trump’s account officially goes live, and the data frenzy is a bit absurd
Under the American Investment Act, Trump’s account was officially launched on July 4, the day of the 250th anniversary of the nation. The official app immediately shot to the #1 spot on the US App Store downloads list. The core mechanism: every child born in the US automatically receives a lifelong tax-exempt investment account opened by the government, with an initial injection of $1,000. The funds go directly into the S&P 500 index and incur no management fees for life. In the 24 hours before launch, 1.5 million accounts were opened, and deposits exceeded $1 billion.

The tax benefits are also aggressively designed. Each year, friends and family can put up to $5,000 per child and enjoy tax-exempt compound interest for 18 years. Even more ruthless: employers can inject up to $2,500 tax-free per employee’s child each year, and this portion is deducted directly from the employer’s and employee’s taxable income—stronger than the traditional 401(k) and IRA. After the child turns 18, the money can roll over seamlessly into a Roth IRA. In the CPA circles, this is broadly called a “historic-level wealth transfer and tax-avoidance tool.” If you start saving from age 0, by ages 18 to 28 the money can grow into well over a million dollars just through compounding.

Charitable donations are also over the top. Brad personally donated $100 million. The Dell couple anchored with $6 billion to fund children from 25 million low- and middle-income families. SpaceX’s CEO donated SpaceX stock worth $350 million (effectively letting each low-income child directly hold SpaceX equity). Micron also donated $250 million to match employee children. Most people believe this setup fully bypasses the old inefficiency path of traditional NGOs, with layers of middlemen taking cuts—because the money goes directly into personal accounts. It’s expected that over the next 10 years, more than 100 million accounts will be opened, with $2 to $4 trillion of assets injected—pushing the US stock market ownership rate from about 50% to 75% or more, to a certain degree turning the entire population into direct beneficiaries of the capital markets.