Anthropic’s S-1, Meta’s $90B, And Elon’s Warning Call | The Brainstorm 155

Meta Platforms

Across the shows: 57 speakers in other episodes, 47 bullish and 30 bearish calls, past 30 days

• Sam Korus · ARK InvestNeutral

  Regulatory fines on Meta, including $25B, are now seen as a immaterial cost of doing business.

  Sam argues that regulators act too late to change Meta's behavior. A $25 billion fine seemed large when it was imposed, but it is now not even a factor, and it is treated as a cost of doing business. He is making this point in response to Meta's harm to teen attention and fertility, which he attributes to Meta while noting the allegations.

  “And, you know, it's like down the road, it's like, sure, they get fined $25,000,000,000, which seemed like a lot when they were doing it. But then now it's not even a factor.”at 7:43

• Nick Grous · ARK InvestBearish

  Meta's large capital raise is a long-term dilution event for shareholders.

  Nick describes Meta's disclosed study, which said it could raise up to $700 billion before Zuckerberg lost majority voting control. That spooked the market. Meta then announced a first $90 billion tranche over several months and a new nonvoting share class. Nick says this is a long-term dilution event for shareholders, though he also accepts that the compute needed to serve billions of agent users requires this capital.

  “They also issued a new nonvoting class share of stock. So they are looking at this as, over the long term, a dilution event for shareholders.”at 10:12

• Nick Grous · ARK InvestBearish

  Meta's Muse commerce monetization needs bank-like payments control and will take years to pay off.

  Nick works through the math. About 250 million daily shoppers at about $200 per purchase, with a 5% take rate, needs trillions in commerce flowing through Muse to reach Meta's $300-350 per-user ad ARPU. He concludes Meta would need to own the payment and merchant side, so it would have to launch a bank. He says it is still a trillion-dollar opportunity, but it will take five to ten years of heavy spending before it approaches profitability.

  “I actually think you end up having to launch a bank here. Like, you need to own the entire payment infrastructure to shave off as many points as possible in that transaction.”at 14:48

Anthropic

Across the shows: 67 speakers in other episodes, 61 bullish and 55 bearish calls, past 30 days

• Brett Winton · ARK InvestBullish

  Anthropic's financials show near-record revenue growth alongside very high gross margins.

  Brett calls the twelve months of revenue growth in Anthropic's S-1 probably the most impressive in corporate history. He pairs that with very high gross margins, which he attributes to efficient R&D, even though Anthropic keeps spending billions on compute to train new models. He sees the products as remarkable business tools that have hit a sweet spot in what the market finds useful.

  “Probably the most impressive twelve months of revenue growth in corporate history crossed with very high gross margins and incredible R and D efficiency to yield those gross margins.”at 0:54

• Nick Grous · ARK InvestBullish

  Anthropic and OpenAI can monetize AI profitably today, as Anthropic's S-1 showed.

  Nick contrasts frontier labs that can monetize AI profitably now with Meta and others that will spend billions on consumer products before reaching profitability. He cites Anthropic's S-1 as the evidence that enterprise AI monetization is already working.

  “...versus what you're seeing with OpenAI Anthropic, where you can monetize that today profitably, which is what we learned from Anthropic's [S-1].”at 12:36

Space Exploration Technologies

Across the shows: 35 speakers in other episodes, 28 bullish and 16 bearish calls, past 30 days

• Brett Winton · ARK InvestBullish

  SpaceX's plan to build 10 gigawatts next year is a strong vote of confidence in its execution.

  Brett says SpaceX revised its plan from building five to ten gigawatts next year, to building a full 10. He reads that as a big upgrade in confidence in its own team's ability to attach power to chips, and it has the NVIDIA relationship for compute supply. He frames SpaceX as a supplier that could step into the market with a big lump of capacity.

  “The surprising wrinkle here is SpaceX just turned around and said, well, actually, rather than building five to 10 gigawatts this next year, we're going to build a full 10.”at 23:33

• Brett Winton · ARK InvestBullish

  SpaceX has better credit than neoclouds because its other businesses generate cash flow.

  Brett reports that SpaceX raised another $40 billion in debt to buy NVIDIA chips and build data centers. He argues it has an advantage over neoclouds because its other business lines generate cash flow, which makes it a better credit with a big balance sheet. If compute demand falls short, SpaceX can take capacity internally rather than selling it.

  “I think that the advantage that SpaceX has and the core cloud providers have is SpaceX has both other business lines that are producing cash flow, so then they're a better credit.”at 25:01

OpenAI

Across the shows: 25 speakers in other episodes, 11 bullish and 20 bearish calls, past 30 days

• Nick Grous · ARK InvestUncertain

  OpenAI faces a dilemma: its consumer user base versus enterprise monetization.

  Nick calls OpenAI the most interesting company in this dynamic because it already has a billion-plus user base on the consumer side and the enterprise monetization is already there. He asks how it can serve consumers and compete with rivals willing to spend hundreds of billions on consumer products while expecting losses. He does not say which path wins.

  “the company that's the most interesting given this dynamic is OpenAI, because they already have this huge... billion plus user install base on the consumer side. But... the monetization on the enterprise side is there.”at 16:52

CoreWeave

Across the shows: 54 speakers in other episodes, 14 bullish and 52 bearish calls, past 30 days

• Brett Winton · ARK InvestBearish

  CoreWeave's five-year contracts lock in blended pricing well below spot compute rates.

  Brett contrasts SpaceX, which signs spot-priced rental contracts at about $50 billion per gigawatt, with CoreWeave's blended monetization of about 10 to 15 billion per gigawatt. He attributes the gap to CoreWeave's five-year contracts, which include the blended cost decline of compute over time and are matched against financing. He says markets are scared of CoreWeave and neoclouds, and a shortage that lasts would help SpaceX as the on-the-run supplier.

  “CoreWeave's blended monetization, I think it's between ten and fifteen [billion per gigawatt]. And that's because CoreWeave has to sign a five year deal that includes the whole blended cost decline of the compute over time”at 27:50

AI infrastructure & data centers

• Brett Winton · ARK Invest

  Spot GPU rental rates are rising sharply, with marginal compute scarce and well above build cost.

  Brett says it is undeniable that spot prices are still rising. Anyone trying to rent marginal compute cannot find supply. On SpaceX's call, marginal renter rates were above $50 billion per gigawatt, while SpaceX can build a gigawatt for $34 billion. He calls that a great business to be in, and says the compute scarcity is pushing returns to the top of the stack.

  “On SpaceX's call, they said marginal renter rates are above $50,000,000,000 per gigawatt, which SpaceX can build a gigawatt for $34,000,000,000.”at 22:27

• Sam Korus · ARK Invest

  Capital is being misallocated to AI compute builders that cannot connect power.

  Sam reads Elon's warning as a signal to debt markets to keep allocating capital to investment-grade borrowers rather than to companies that cannot connect their power. He thinks there is clear misallocation of capital that the entire market does not yet see, and that calling it out is advantageous for SpaceX.

  “I think there's a lot of... He sees a lot of misallocation of capital very clearly that I don't think the entire market sees.”at 26:51

Cybersecurity

• Brett Winton · ARK Invest

  AI-driven vulnerability discovery and patching will make the internet safer and generate tech revenue.

  Brett cites a PHP vulnerability found by frontier-lab models that was leaked and then patched in a worldwide scramble. PHP runs a large share of websites, and the flaw had existed for years. He sees frontier models' ability to find bugs before attackers as good for the internet and the labs. He expects this to make the internet safer over time and to generate revenue for the tech sector fixing problems. Attacks that intentionally break things are within the normal range, as long as they are not rogue agents.

  “This is something that both should happen, will make us safer over time, will generate lots of revenue into the tech space to fix all of the problems”at 4:20

Open-source AI models

• Brett Winton · ARK Invest

  Open-weight models will likely produce agent bot infestations of websites within a year or eighteen months.

  Brett describes a scenario in which agents spend resources to serve a small business and then multiply to absorb other people's infrastructure, similar to the Hugging Face hack. He says he has not yet seen this at scale. He thinks OpenAI and Anthropic are wary of it, so open-weight models will probably produce this kind of bot infestation within the next year or eighteen months. He thinks that will be a thornier set of issues to deal with.

  “And I think you should expect to see something like that come from the open weight models, probably within the next year or eighteen months, and then that'll be a thornier set of issues to deal with.”at 5:15

Consumer AI products

• Brett Winton · ARK Invest

  Consumer agents address far less market than enterprise, so compute should favor enterprise.

  Brett compares a $30-40 trillion enterprise AI software market with consumer agents that can only clip a 5-10% finder's fee on roughly $60-70 trillion of global consumption. That gives consumer agents an addressable market of only about $5-10 trillion. He asks whether Meta, SpaceX, and OpenAI should put scarce compute into enterprise, where the money is. He hedges that the 5% take rate may be too conservative given aggregation power.

  “global consumption is 60 to 70,000,000,000,000, and the agents are only gonna be able to clip off a finder's fee for that consumption, basically, in the five to 10% range.”at 11:20

Power demand & the grid

• Brett Winton · ARK Invest

  Permitting, shell construction, and turbine problems leave AI chips installed but not powered.

  Brett says the reported stranded compute reflects a real bottleneck. He thinks that, on a project-by-project basis, a combination of permitting, building the shell, and getting turbines working as intended leaves NVIDIA chips in inventory but not switched on. He calls this probably true to some degree.

  “A combination of permitting and building the shell and getting the turbines working the way it's supposed to is causing chips to be in inventory but not turned on.”at 23:18