All-In Latest Highlights: Anthropic IPO vs OpenAI, Unveiling AI's Real ROI, China's Model Export Restrictions, and Universal Shareholding

Compiled by: Deep Tide TechFlow

Guests: Chamath Palihapitiya (Founder of Social Capital), Brad Gerstner (Founder and CEO of Altimeter Capital), David Sacks (Partner at Craft Ventures)

Host: Jason Calacanis, All-In Podcast

Podcast Source: All-In Podcast

Original Title: OpenAI vs Anthropic IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts

Broadcast Date: July 11, 2026

Key Points Summary

In this episode of All-In, Episode 280, Friedberg is on vacation, and Brad Gerstner is filling in. The show begins with a trillion-dollar IPO race: SpaceX has successfully gone public with a valuation of $1.75 trillion, and Anthropic secretly submitted its application on June 1, followed closely by OpenAI. Gavin Baker predicts that Anthropic's revenue could exceed $100 billion this year, with a potential IPO valuation of $3 trillion. Brad confidently states that Altimeter will aggressively buy into the IPOs of both companies.

However, Chamath poured cold water on the excitement. He noted that his company's token costs double every 45 days, while downstream productivity increases by at most 5%. He asked Claude 5 a question: How much EPS growth has AI brought to the S&P 500? The answer was 50%. But excluding the portion from Nvidia selling chips to Amazon, the actual EPS growth for the S&P 493 is only 9%, most of which comes from pricing power above inflation and buybacks, with the real AI ROI between 0% and 2%. Chamath's judgment is: If they can go public now, they should, while these numbers haven't yet seeped into the market's awareness.

The latter part of the discussion shifted to China. Reuters reported that the CCP is considering restricting overseas access to China's top AI models, classifying AI research leaks as national security crimes. Sacks has discussed this topic with people in Washington, including the White House and the Treasury Department, and his judgment is that China's strategy is similar to Sam Altman's approach years ago: open source while catching up, then close source after catching up. He also revealed that GLM-5.2 contains distilled watermarks from cutting-edge U.S. models, and the U.S. government will likely take action against distillation. At the end of the show, Brad spent nearly an hour discussing Trump Accounts, a plan that gives every newborn in America $1,000 and invests in the S&P 500, with 1.5 million accounts opened within 24 hours of the app's launch, attracting over $1 billion in deposits.

Highlights of Opinions

On IPO Timing

• Chamath: "If they can go public now, they should, while these numbers haven't yet seeped into the awareness. Because I think that's the window where you can sell at a high price and raise a lot of money."
• Brad: "Today, Altimeter will buy into these two IPOs at scale and volume."
• Brad: "Anthropic's annual revenue could exceed $100 billion, while SpaceX's projected revenue is only $35 billion. Based on SpaceX's success, this will be a phenomenal IPO."

On AI ROI

• Chamath: "My token costs double every 45 days, while downstream productivity might increase by at most 5%. My costs are doubling, and my returns are basically flat."
• Chamath: "The EPS growth for the S&P 493 is 9%, most of which comes from pricing power above inflation, and another 3% from buybacks. The real AI ROI is between 0% and 2%."
• Brad: "We've never seen revenue growth like this because we've never seen such a large TAM. Intelligence is the largest addressable market in human history."

On Open Source vs Closed Source

• Sacks: "The spirit of enterprises is willing, but their capabilities are weak. They want to transition from closed-source models but can't."
• Sacks: "The share of open source in enterprise spending is actually declining, from 19% last year to 11% this year."
• Brad: "Whether you use a $3 cheap model or a $15 cutting-edge model to replace a $200-an-hour consultant, the price difference is negligible."

On China's Shift from Open Source

• Sacks: "China's strategy is: you open source while catching up, and once you catch up, you close source. Sam Altman did this three years ago."
• Sacks: "GLM-5.2 contains distilled watermarks from Mythos. The U.S. government will take action against distillation, and that's what should be done."
• Chamath: "The best thing for the U.S. would be for China to also have a doomsday cult community emerge."

On Trump Accounts

• Brad: "If you receive $1,000 at birth, and someone matches some, and you save $10 a week, by age 18, you'll have $50,000. All invested in the S&P 500."
• Sacks: "If Trump accounts were fully funded from the start, based on the market returns of the past 30 years, by age 28, that child would be a millionaire."
• Jason: "This could replace Social Security. It replaces the giving pledge."

Main Text

Chapter One: The Trillion-Dollar IPO Race: SpaceX Sets the Example, OpenAI and Anthropic Prepare to Debut

Jason: Let's start with the IPO updates. A trillion-dollar IPO sprint is underway, with SpaceX already listed, and trading prices are close to the issue price. The pricing is perfect, and theoretically, there are two more to come: OpenAI and Anthropic. SpaceX's stock price once surged to $200 but has now fallen back to $150, right at the issue price. Its current market value is $2 trillion, making it the seventh-largest company globally. Anthropic secretly submitted its application on June 1, and Polymarket gives a 65% probability of going public this year. Gavin Baker said two weeks ago that he believes Anthropic's revenue will exceed $100 billion and become profitable by the end of this year, with a potential valuation of $3 trillion if they go public now. Chamath, you previously said that Elon going public first was a good move; what are the chances that these two will come out this year or in the first quarter of next year?

Chamath believes both companies are excellent businesses, but the core issue is where the market clearing price is. This depends more on the market's appetite for new stock offerings and at what price they can be digested.

OpenAI and Anthropic are at different stages. The last information disclosed by OpenAI shows that cash consumption is still high due to a diversified business model that relies more on the consumer side. Brad previously mentioned that Anthropic might have unexpectedly turned a profit. Chamath shared a detail: he asked his CTO about token spending, and the response was, "Currently doubling every 45 days." He followed up by asking how much downstream productivity has increased, and the CTO said, "At most 5%." Costs are doubling, and returns are basically flat. The CTO explained that to achieve the next iteration, a lot more tokens need to be consumed because the effects are starting to diminish.

Chamath's judgment is: if they can go public now, they should, while these numbers haven't yet seeped into the market's awareness. This is likely the window for raising large amounts of money at high prices.

Brad, as an investor in both companies, offered a more optimistic assessment. SpaceX's IPO is a textbook case: raising $75 billion, with a valuation of $1.75 trillion and projected revenue of about $35 billion, with the stock price already up 25%. Anthropic's revenue is said to potentially exceed $100 billion this year, and if true, next year's GAAP revenue could far exceed this figure. Based on SpaceX's successful precedent, Brad believes this will be a phenomenal IPO. SpaceX has done pioneering work in IPO volume, pricing, liquidity, index inclusion, and lock-up arrangements, and both Anthropic and OpenAI are learning from it.

Regarding the controversy over index inclusion, Brad explained that the previous rules made sense because most newly listed companies are younger, have less revenue, and weaker profitability. However, SpaceX is too large and important not to be included in the index. Exchanges and index companies have made adjustments to avoid stuffing it in at the peak, thus preventing the common issue of a 30% pullback post-IPO affecting passive investors.

Brad also revealed the latest developments at OpenAI: revenue has rebounded to about $70 billion this year, and GPT-6 may be released within 30 days. Although it is only twice the revenue of SpaceX and less than the rumored $100 billion of Anthropic, as one of the two leading labs, going public with this growth rate exceeding a trillion is reasonable. He does not believe there is a race between the two; they will both act when the timing is right. OpenAI's corporate restructuring is more complex, so it may go public after Anthropic.

Chapter Two: Token Costs Double Every 45 Days, Is AI ROI Close to Zero?

Jason: We have been discussing the ROI issue of token spending for the past few weeks. CTOs and CEOs in the industry have started to respond publicly on X. Uber's CTO Pinen shared their approach: 99% of engineers are using AI tools, with over 70% of pull requests coming from local or cloud agents, and engineers have built 200 agentic skills. They have deployed engineers to various departments as "frontline deployment engineers" to work with department heads to streamline processes. Brad, what do you think of Uber's approach?

Brad believes Chamath is right, but the issue is just the timeframe. There is indeed a lot of money being spent in experimental buckets that may not have direct ROI. However, it is still too early for companies to adopt AI. The addressable market is unprecedentedly large, encompassing every company on Earth. Revenue distribution is also not concentrated, with millions of customers making rational decisions independently every day.

Brad made a bold prediction: If Anthropic's revenue exceeds $100 billion by the end of the year, their revenue could multiply by 3 to 5 times next year. An increase from $100 billion to $300 billion, a $200 billion incremental revenue, is unimaginable in Silicon Valley's history.

Chamath's skepticism centers on the sustainability of ROI. He posed two questions to Claude 5. The first: How much EPS growth has AI brought to the S&P 500? The answer was 50%. However, he found that this figure included Nvidia's chip sales to Amazon. So he asked the second question: What is the EPS growth for the S&P 493 (excluding Mag7)? The answer was 9%. Breaking it down, most of it came from pricing power above inflation, with another 3% from buybacks. The actual ROI attributable to AI is between 0% and 2%.

Chamath believes that while the enterprise side looks glamorous, the problem is that smart investors like Brad and Gavin will eventually ask companies: What is your ROI? Where is the actual EPS improvement? If the answer is "I'm not sure," and you lack sustained pricing power, the enterprise side will become vulnerable. The consumer side, on the other hand, has become a safe haven because you have tens of millions of buyers, with much smaller price points, and the two orders of magnitude difference in the number of buyers exempts you from ROI scrutiny.

Jason added another perspective: The uniqueness of this technology is that it touches everyone in the organization. When Excel came out, the accounting department was excited, but HR and marketing felt little impact. AI is different; in an organization of a thousand people, everyone is using it, spending $200 a month each, doubling to $400, which only increases their $150,000 annual salary by 3 to 4%. The key question is: Has it improved this person's efficiency by 3 to 5 times? If so, that explains why token spending is skyrocketing.

Chapter Three: Open Source vs Closed Source: Revenue Concentrates at the Frontier, But Enterprises Want to Escape {#article-toc-32510-7}

Jason: Sacks, CTOs are starting to discuss smart routing on X, first sending tasks to open-source models, and if that fails, falling back to Claude. What do you think of this trend? If you were an investor and the CFO of a frontier model started asking, "Can it be cheaper?" how would you view the growth of frontier models?

Sacks believes enterprise CTOs indeed want to shift token consumption to cheaper models. They are watching token costs soar and are looking for ways to hit the brakes or at least control it. Coupled with last week's discussion on AI sovereignty, enterprises are concerned about handing over core alpha to a frontier lab that could become a competitor in the future.

Sacks's core judgment is: Enterprises want to shift from closed-source models, but most lack the technical capability to do so. The spirit is willing, but the flesh is weak.

Coinbase and DoorDash have succeeded; they built token routing middleware that sends frontier tasks to frontier models and non-frontier tasks to regular models. But most enterprises lack this capability. This is why the wallet share of closed-source models is actually increasing. The share of open-source in enterprise spending has dropped from 19% last year to 11% this year. Of course, this doesn't mean usage is declining; it may simply be that using open-source models incurs only hosting fees, not payments to labs, making it hard to track.

Sacks also cited the founder of Decagon's viewpoint: When you know exactly what you want to do, using small and cheap open-source models is correct, but you need data and post-training. If you still don't know what to do, you want the most powerful general intelligence. Mature use cases use open-source, while immature use cases use frontier models.

Jason mentioned a discovery by Databricks founder Ali: With the same model, changing the harness (task orchestration framework) can cut costs in half. GLM-5.2 performs exceptionally well with a specific harness, directly halving the task volume. Jason himself has experienced this: he built a trend discovery agent that runs hourly, and after optimization, token consumption dropped by 80%. When tokens became cheaper, he changed the agent from running daily to hourly and split a single agent into three parallel tasks. He woke up to find 14 tasks completed, feeling completely different.

Brad's view on this is: The core debate is whether intelligence will converge. Eighteen months ago, when the DeepSeek moment occurred, the market dropped by 40%. Many believed frontier models were finished, and open-source would kill them. But 18 months later, the opposite is true. Jesse Zang's tweet pointed out that the wallet share of frontier labs is actually increasing, even though token usage is rising on both sides.

Brad proposed a counterintuitive hypothesis: Perhaps intelligence will not converge at all. If superintelligence becomes self-recursive, the smarter the model, the more money it earns; the more money it earns, the more computing power it buys; the more computing power it buys, the better models it builds. The distance may not shrink in the next 2 to 3 years, but rather widen.

Jason also mentioned that he interviewed Lovable's CEO Anton, whose product has been online for about 30 months, with revenue rising from zero to $600 million. He also asked 11Labs CEO Matti: You are a major customer of frontier models, spending tens of millions of dollars each year; are you worried about data leaks and competition? Both said they are developing their own models. These are eight-figure and nine-figure major clients, and if they all start building vertical models, frontier labs will feel the pressure. But Chamath countered: 11Labs wants to create the best voice agent in the world; if the best voice capabilities come from frontier labs, can he afford to use a suboptimal self-built model in a competitive market?

Chapter Four: Zuck Launches a Price War: Same Quality, One Percent Cost {#article-toc-32510-8}

Jason: Meta released Spark 1.1 this week, a powerful agentic encoding model at a very low price. Zuck has been unusually active on X, tweeting more than ever before. He is basically saying: I offer you the same quality, but at only one percent of the cost. Brad, what do you think of Zuck's strategy?

Brad believes Meta previously made mistakes in their open-source strategy, but now Zuck has clearly chosen the direction of a price war. Meta also released a new model API, not just creating models but also providing tokens. Competition is good for America.

Brad used an analogy to explain why frontier models won't be easily replaced: If your AI agent is replacing a $200-an-hour consultant, whether you use a $3 cheap model or a $15 frontier model, that price difference is not significant. The key is whether the $15 model can complete the task without error. If the task crashes halfway through, you lose both tokens and time.

Chamath has a different view. He believes that just like when the iPhone first came out, people kept upgrading because the new price was worth it. But one day, people will say, "The old phone is good enough." When he tried Claude 5, he found that some research directions were restricted and unanswered. Everyone will reach the critical point of "good enough" at different times.

Chamath also shared his experience in the UN AI Committee. He participated in the UN AI Committee co-chaired by Benioff, along with Benioff, Jensen, and Brad Smith. His observation is that there is not a single country in the world that is not formulating its own sovereign AI strategy, and no country is willing to use America's closed-source models as the answer. Many countries would rather take an open-source model, like Nvidia's, and build a complete set of infrastructure themselves.

Examples of sovereign AI include: the UAE's Falcon model, Saudi Arabia's Arabic LLM, and Japan's $6 billion investment in the Neoterra alliance, which directly jumps to physical AI and robotics. Chamath believes that when models reach 95% to 99% of frontier levels, many countries will say, "Good enough." On the other hand, some companies do not have enough profit growth to support such spending and lack the courage to make large-scale cost reductions. Just like the famous letter he wrote to Zuck, Zuck was ultimately pushed by pressure to execute. Most companies will just let problems accumulate.

Chapter Five: China Considers Restricting AI Model Exports: Closing the Door After Catching Up {#article-toc-32510-9}

Jason: Reuters reports that the CCP is considering restricting overseas access to China's top AI models. Two regulatory agencies have interviewed Alibaba, ByteDance, and Z.AI (the one making GLM-5.2), discussing restrictions on overseas access to top open-source and closed-source models. They are listing AI research leaks as national security crimes and want to control who can invest in Chinese AI labs. Sacks, last week I posed the reverse question: Should the US ban Chinese models? Now it's reversed; China says it wants to restrict. What do you think of this chess game?

Sacks believes this news may be somewhat exaggerated. China's top model is ByteDance's, which has always been closed-source. Alibaba's Qwen was previously open-source but may now be turning closed-source. Z.AI's GLM-5.2 was open-source before but is also transitioning to closed-source.

Sacks's judgment is: The strategy is clear; you open-source while catching up, and once you get close to the frontier, you close-source. Sam Altman did exactly the same thing with OpenAI three years ago, transitioning from non-profit to profit, from open-source to closed-source.

The benefit of open-source is attracting the developer community, and in the AI field, it also gives you a reinforcement learning data flywheel. But once you catch up, closed-source can capture all the value.

Sacks discussed this topic with the White House and the Treasury in Washington this week. He said there is absolute consensus among all regulatory disputes: to maintain a lead over China at all costs. From the president down, everyone is asking, "How far ahead are we?" and "What do we need to do to stay ahead?" The idea of letting American frontier labs retreat while allowing Chinese open-source models to flow freely does not exist in Washington. He also revealed that GLM-5.2 contains Mythos's distilled watermark, and the US government is likely to take action against distillation.

Sacks believes that China's actions will actually have little impact on the US. The US has the capability to create open-source models; Nvidia is doing it, and Reflection is also doing it. He has talked to frontier labs about why they don't do open-source, and the response was, "There isn't much demand; if there were, we would do it." For China, restricting exports may hurt itself more.

Chamath made a joke: the best thing for the U.S. would be for China to also have a doomsday cult community, worrying all day about AI job losses and existential risks. If China's laboratories also start to be constrained by regulations, that would be the biggest boon for the U.S.

Chapter Six: Trump Accounts: Opening an S&P 500 Account for Every American Child at Birth {#article-toc-32510-10}

Jason: Brad went to Washington this week. The Trump Accounts app has become the most downloaded app globally. Congratulations, Brad, this is the fruit of your four years of hard work. Tell us what happened.

Brad introduced this as a four-year journey. Last year, the Invest America Act was signed into law as part of a bill, and the app officially launched on July 4 this year. Every American newborn receives $1,000 deposited into a private investment account, all invested in the S&P 500. The account is free for life. Within 24 hours of launch, 1.5 million accounts were opened, attracting over $1 billion in deposits. They held the first-ever joint bell-ringing ceremony in history between the NYSE and NASDAQ at the Oval Office, attended by hundreds of CEOs. The president proposed to automatically create accounts for 50 to 70 million minors under 18.

Sacks analyzed the brilliance of this mechanism from a financial planning perspective. Each year, $5,000 can be deposited into the child's account (from relatives and friends), and employers can contribute $2,500 tax-free. Tax-free compound interest is enjoyed until the age of 18. After turning 18, up to 25% can be withdrawn for buying a house, starting a business, or attending college, with the remaining amount rolling into an IRA. If they wait until the child is no longer a dependent (for example, just graduated and in a 0% tax bracket) to convert the IRA to a Roth IRA, they can turn the money into a lifetime tax-free investment with almost no tax cost.

Sacks calculated: if the Trump account is fully funded from the start, based on the market return rate over the past 30 years, by age 28, this child would be a millionaire. If they have $200,000 to $300,000 at 18, it could compound to over $10 million by age 60.

There were also a series of significant charitable announcements. Michael and Susan Dell donated over $6 billion, providing $250 for each child from 25 million low- to moderate-income families. SpaceX President Gwen Shotwell donated $350 million worth of SpaceX stock, directed to children in low-income communities. Micron donated $250 million, with up to $1,000 for each employee's child. Brad himself donated $100 million, covering all children in Indiana.

Brad said they told the president they expect to raise $100 billion within 12 months. This will become the largest direct charitable platform in U.S. history, with no intermediaries, going directly into children's accounts, which cannot be withdrawn until they turn 18. On this trajectory, there will be over 100 million private investment accounts in the next decade, and over the next 15 years, $2 to $4 trillion could flow into accounts that previously had nothing.

Jason summarized from a broader perspective. He said this project could replace social security and the giving pledge. Currently, only 50% of Americans own stocks, but if Trump accounts are successfully promoted, it could rise to 70% to 75%. Australia is one of the happiest countries in the world because their superannuation system mandates that everyone saves 12% to 14% of their income into an account similar to a 401k. What Trump accounts do is something similar, but at a more fundamental level.

Jason also specifically thanked Joe Gebbia (co-founder of Airbnb) for joining the government to oversee the software design for this project. He said the U.S. government has produced outstanding consumer-grade software, which is rare in history. Brad added that the team includes Michael Dell, Vlad Tenev (CEO of Robinhood), Joe Gebbia, and Luke Pettit from the Treasury Department, aiming to create not just the best government product but one of the best consumer-grade products.

Disclaimer: This content is provided for general branding and informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online events, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets or to use any services. Crypto assets are highly volatile and may result in loss. WEEX services and online events may not be available in all regions and are subject to applicable laws, regulations, and eligibility requirements. You are responsible for ensuring that your use of WEEX services complies with local laws and for carefully assessing the risks before participating in any crypto-related activities.