IPOs and SPACs are Back, Mag 7 Showdown, Zuck on Tilt, Apple's Fumble, GENIUS Act passes Senate
Speakers
Thomas Laffont
— Managing Director, Coatue Management
Chamath Palihapitiya
— CEO, Social Capital
Jason Calacanis
— Angel Investor / Founder, LAUNCH
David Friedberg
— CEO, The Production Board
David Sacks
— General Partner, Craft Ventures
Summary
The All-In hosts Jason Calacanis, Chamath Palihapitiya and David Friedberg are joined by Coatue's Thomas Laffont for a markets-heavy episode centred on the AI arms race. They debate Meta's 100 million dollar talent offers and the Scale AI deal, rank the Magnificent 7 on who wins AI over the next five years, dissect Apple's stagnation, and cover the reopening IPO and M&A window around CoreWeave, Circle and Chime. The second half argues the SaaS growth era is over, that dispersion is spreading from the Magnificent 7 into the S&P 493, that Amazon is a kingmaker for physical AI, and splits on whether Microsoft grows or shrinks. David Sacks joins at the end to explain the GENIUS Act stablecoin bill passing the Senate.
- Thomas Laffont argues Meta's 100 million dollar packages and the Scale AI deal are rational given how much of its market cap AI puts at risk.
- Chamath Palihapitiya says only companies that tightly couple models to their own silicon and infrastructure win, which drives his Tesla and Google picks.
- AI winners are ranked: Nvidia and Tesla for Laffont, Tesla and Google for Chamath, Google first for Friedberg, with a debate about Nvidia's China risk.
- Apple is criticised as a cash-harvesting business that cannot recruit AI talent, though Laffont and Friedberg argue it is too early to write it off.
- IPOs and M&A have reopened, with CoreWeave, Circle and Chime cited as the AI- and crypto-levered growth the market is starved for.
- SaaS median growth has halved since 2021 and enterprises are expected to rebuild software with AI tooling, pressuring per-seat and consumption pricing models.
- Chamath makes the case for shorting the S&P 493 and for owning a synthetic index of AWS, Azure and GCP; the group splits on Microsoft's next five years.
- David Sacks explains the GENIUS Act passing the Senate 68-18, bringing stablecoin issuers onshore with quarterly audits and full one-to-one reserves.
Ideas
Zuck's AI spending spree is rational insurance.
Meta is worth roughly 1.7 trillion, and if as much as half of that market cap is at risk from AI, then spending four or five percent of it on 100 million dollar talent packages and on 49 percent of Scale AI is highly rational insurance, because it raises the odds of winning the market even slightly. Laffont also reads the deal structure as urgency rather than only antitrust avoidance, since Alexandr Wang could show up at Meta the very next day.
Meta must buy compute and infrastructure secrets.
The labs that win compound secrets across the training layer, the model layer and tightly coupled compute: OpenAI with Azure, Google with TPUs, Anthropic with TPUs, DeepSeek and xAI with their own hardware coupling. Meta trains generically on off-the-shelf Nvidia and ships open source, which is why Llama quality is mediocre. Zuckerberg has now bought labeling and reasoning-data secrets with Scale and app and agent secrets with Nat Friedman and Daniel Gross, and he still has to buy infrastructure and compute secrets before Meta can fundamentally compete.
Magnificent 7 correlation breaks into winners, losers.
This may be the year of the greatest divergence inside the Magnificent 7: Meta up 18, Microsoft up 13, Nvidia up 8, Amazon down 3, Google down 8, Tesla down 20 and Apple down 21. After years in which the group traded as one correlated block, the market is now trying to sort out who is positioned to win AI and who is falling behind, so investors should expect continued dispersion inside the group rather than one single trade.
Tesla is misunderstood; vertical integration wins.
Every name in the group except Nvidia is baking and rolling its own silicon, and the winners will be those with a vertically integrated stack. After spending time at Tesla, Chamath says the business is once again misunderstood: it has the best vision models, it gains one of the best LLM and reasoning model families through xAI which can eventually run on Dojo, and that combined stack ends up inside all of the physical AI people interact with daily, whether a robot, a car or a robotaxi.
Apple became a cash cow, not innovator.
Apple is the only one of these companies with nothing public and apparently nothing private in silicon or models, and it is transitioning from a growth business into a cash cow in harvesting mode. Its longest-serving executives have been there 20 to 30 years, it is winning none of the AI talent that OpenAI, Meta and Google are fighting over, and its revenue chart shows a stalled iPhone with growth coming from replacement accessories, which is revenue optimization rather than strategy. Chamath says creative destruction of companies is acceptable, compares it with HP, Lotus, Intel and GE, and does not think Apple has a chance of doing anything exceptional from inside that culture.
All AI roads still lead to GPUs.
Asked who wins the AI prize over the next five years, Laffont's number one is Nvidia. He does not see the GPU being displaced: additional architectures will come on board and grow the market, but at the end of the day all roads still lead to the GPU for every one of these models.
Tesla's vertical integration from silicon to robots.
Laffont's number two and dark horse is Tesla, because it has the most potential for genuine vertical integration all the way from the silicon to the model to the hardware, and that integration may prove decisive not only in cars but in Optimus and humanoid robotics.
Google pivots price-per-click to price-per-token.
Google is Chamath's number two because it is close to a fully vertically integrated stack: the Gemini family landing model after model, its own TPU with an exceptional next generation, quantum work in the pipeline, and a funnel of billions of users through YouTube, Gmail, Workspace and search. Even if search declines, Google's economic north star metric can be pivoted from price per click to price per token across that same user base, which it is very well positioned to do once it rips the band-aid off.
Merge Tesla and xAI to win AI.
Calacanis picks Elon's companies for the AI prize: Colossus, plus Tesla's own hardware, software, FSD and Optimus, plus the real-time data of X and the xAI models, is a uniquely complete position, and after visiting xAI he describes it as a magnet for top talent. He argues the Tesla and xAI boards should combine the roughly trillion-dollar and hundred-billion-dollar companies so all that brainpower pushes in one direction instead of Elon context-switching between them; if that happens he thinks Elon wins outright.
Google's ad network wins even losing search.
Calacanis argues search share is the wrong question for Google. What matters is whether its ad network gets more effective, and with chat queries, Gmail content, Chrome browsing, Android usage and YouTube listening data, Google's ads should perform so much better that the ad business keeps growing, and even accelerates, while search share erodes.
Tesla owns humanoid robotics, premium already priced.
Friedberg calls Tesla the best place to invest if you want a shot at a massive new industry: on top of the baseline automobile business there is a low-probability, high-upside call option in humanoid robots, and he does not think any company on earth is better positioned to execute against that opportunity. His caveat is price, since that option already carries a healthy premium in the valuation, so on alpha and beta-adjusted returns he ranks Google ahead of Tesla and is unsure he would pay the premium today.
China chip breakthrough threatens Nvidia's moat.
The common view is that Nvidia is the most protected and durable of these businesses, but Friedberg sees a low-probability, very high-severity risk from China. A 1-nanometer manufacturing process was demonstrated there last month, a roughly 40 billion dollar state fund is going into full-stack domestic semiconductor manufacturing, and China is developing its own DUV and EUV systems, meaning the lithography IP moat is being crossed. The more the US isolates China through policy, the more it subsidizes an alternative to Nvidia's moat, and he expects to be surprised by Chinese process technology the way everyone was by DeepSeek.
Google is a portfolio of trillion-dollar options.
Friedberg's portfolio pick is Google: a diversified set of high-beta bets, any one of which could be a trillion-dollar outcome, ranging from Waymo to quantum computing to the biologics work Demis Hassabis is doing at Isomorphic, plus model depth well beyond LLMs such as the weather forecasting and graph-based models nobody else is working on. The core search business may be at risk, but only one of those options has to hit to make up for it, and Sundar Pichai is being reasonably aggressive about evolving search; on risk-adjusted returns he would put Google number one.
Too early to count Apple out.
Apple's defining advantage was the coupling of hardware and software, which won the mobile era, but in AI it controls neither the silicon nor the underlying models, which makes it look like the PC makers that did not own the operating system. Even so it still has a monopoly on users and three trillion of market cap to play with, and this management team already executed one very hard transition, taking the iPhone from over 90 percent of gross profit a decade ago to about 40 percent today, for which it gets too little credit. It is far too early to count Apple out, though Laffont wants it to be far more aggressive with acquisitions.
Apple can win ambient AI assistant.
Friedberg believes Apple does have a winning path and is already building it: an ambient AI assistant that lives across the whole device fleet rather than in one new gadget, so the agent follows the user between watch, phone, AirPods, car and computer and carries context between them. He owns around 30 Apple devices, which makes the transition easy for existing customers, and of all the companies discussed Apple is best suited to reach the consumer and design and engineer that experience; it does not need to own the full model stack to win it.
AI and crypto IPOs are compounding winners.
CoreWeave and Circle went public months apart and their charts are almost identical even on a dollar basis, one levered to AI and one to crypto, which says the market is paying up for open-ended growth tied to the big future themes while everything else is treated like the anemic S&P 493. With SaaS growth cut in half, Laffont expects investors to hunt for businesses that can compound at roughly 25 percent a year over the next five to ten years, and he names CoreWeave, Circle and Chime as the new cohort that fills that gap.
Short the S&P 493, own disruptors.
The average profit margin of the S&P 493 is about 12 percent and average growth is single digits, and any of those businesses can be decapitated by something built by a couple of kids in a garage using OpenAI or Grok. Chamath argues investors should be less long the past and at minimum hedge into the investable companies in the big themes of the future. For the first time in twenty years, despite being negative on stock picking, he thinks the moment is transformative enough that you can short the S&P and pick a handful of category killers, because the dispersion between the companies that rebuild with AI and those that do not will be one of the biggest money-making opportunities in decades.
SaaS growth halved; the index trade dies.
In 2021 the median SaaS company grew 17 percent and a quarter of the cohort grew above 25 percent; today the median has been cut in half to 9 percent and only 5 percent of the cohort grows above 25 percent. Sectors that were bought as reliable growth are now slowing, so simply owning the Bessemer SaaS index for the next decade no longer works. He adds that Anthropic alone added roughly 70 percent of the net new ARR of the entire public SaaS industry in Q1, showing where that growth has migrated.
AI rebuilds software; the jig's up.
Buyers have realized that yet another vertical software purchase adds bloat, cost and headcount, and since about 2023 they have been waiting for an AI way to rewrite that software, which is why SaaS growth stalled. Rebuilding software from scratch is now dramatically cheaper: a team of 30 at 8090 can service hundreds of millions of dollars of work by rebuilding the whole development lifecycle with these tools, and an owner can rip out hundreds of millions of dollars of software licenses and replace them with tens of millions of dollars of custom software. The cartel of influence inside IT organizations that justified budgets as large as 18 billion dollars a year gets undone, so the jig is up for software.
Consumption pricing eventually destroys Snowflake's business.
As the per-seat model breaks down, SaaS vendors are shifting to consumption pricing, and Chamath argues that model can win short-term adoption but destroys the business long term. Customers will not tolerate a variable bill that grows with every terabyte they are forced to store, and they cannot tell in advance which data is valuable, so alternatives grow up around the vendor. Snowflake is his example: users are moving to Postgres and Supabase because the consumption economics stop making sense.
Dispersion spreads from Mag 7 to 493.
Laffont expects the Magnificent 7 debate to become a broader lens on the S&P 493: the same arguments about who is well positioned, who can win, and which management teams are being as aggressive and bold as Zuckerberg will play out inside every boardroom and investment committee. For a stock picker that makes the next five years one of the most interesting windows he can imagine for separating winners from losers inside the index.
Own defensible offline assets like specialty chemicals.
Rather than an IT services or professional services rollup, where he fears there may be no terminal buyer once agents can do the work, Chamath would screen the S&P 493 on what offline and online assets each company owns and what share of those assets stays defensible and unique in a post-AI world versus simply disappearing. That filter pushes him toward physical, hard-to-disintermediate businesses; the example he gives is owning a specialty chemicals company, because the world still needs lubricants no matter what the agents can do.
Amazon is physical AI's demand kingmaker.
Amazon is a kingmaker for physical AI because it is a sink for demand: if Figure lands the BMW or UPS robot, if Optimus proves itself in a Tesla factory, or if delivery drones work, Amazon will buy enormous quantities of them, and that gives its retail operation a large opex lift. The harder problem is AWS, where success as a marketplace of many different things is also the bottleneck; Jassy has to decide whether to differentiate his own silicon from Nvidia's and whether to make a real model bet, most plausibly buying Anthropic and coupling it tightly so next-generation codegen has to run inside AWS, which means spending hundreds of billions of dollars.
Microsoft grows; AI-written code is crap.
Chamath expects Microsoft to have more employees in five years, not fewer, and the business to grow at the margin. The share of code written by AI is a dangerous vanity metric because most AI-generated code is poor and error rates compound over long, complex enterprise tasks, so coding agents are not what is actually driving layoffs, which look more like air cover for cuts management already wanted. Meanwhile some of the S&P 493 shrink and disappear, and it becomes cheaper for Microsoft to bundle together the point features those companies sell today, which means more products, different skill sets and a larger employee base.
Microsoft's cloud and apps get competed away.
Friedberg expects Microsoft to shrink because he sees a real probability of revenue decline over the next five years. The enterprise install base gets competed away as cloud commoditizes, and on the application software layer the old-school customers that buy Microsoft are the ones most likely to die, while the companies that build native software and agentic workflows will not use them. He also disagrees with Chamath on AI-written code, expecting it to be good in three or four years given the pace of improvement, which makes the core business more vulnerable, not less.
Microsoft gets bigger and more relevant.
Laffont sides with Chamath as the tiebreaker: Microsoft's business will be bigger if anything, because the application dollars flow toward it and it will simply need more people to support a larger and more relevant franchise. He expects more productive employees, but more of them.
GENIUS Act levels playing field for Circle.
Calacanis argues the GENIUS Act is an example of good regulation pulling an industry back onshore: the grey or offshore actors, with Tether the obvious case given its New York Attorney General settlement and bans in several jurisdictions, now have to compete with fully buttoned-up US issuers such as Jeremy Allaire's Circle on a level playing field. He also points to roughly 50 million wallet holders in the US as evidence the consumer and business demand is already there.
Up Next
This All-In Podcast video, published June 21, 2025, features Thomas Laffont, Chamath Palihapitiya, Jason Calacanis, David Friedberg discussing META, MAGS, TSLA, AAPL, NVDA, GOOG, CRCL, CoreWeave, CHYM, SPY, IGV, SNOW, Specialty chemicals, AMZN, MSFT. 27 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Thomas Laffont, Chamath Palihapitiya, Jason Calacanis, David Friedberg · Tickers: META, MAGS, TSLA, AAPL, NVDA, GOOG, CRCL, CoreWeave, CHYM, SPY, IGV, SNOW, Specialty chemicals, AMZN, MSFT