Anthropic’s S-1 embodies all the contradictions of the AI boom, and the questions about what happens next
The bigger portfolio question is where the economic winners of AI will actually sit. If model developers remain dependent on enormous external compute purchases, investors may need to distinguish between exposure to AI application builders, semiconductor suppliers, hyperscale infrastructure, and power-intensive industrial enablers. A thematic ETF can bundle that story neatly, but it can also hide very different business models, margins, and balance-sheet risks behind a single “AI” label.
This also reinforces a familiar lesson about concentration risk. Investors who think they own diversified technology exposure through a passive index may, in practice, own a portfolio increasingly influenced by a narrow set of AI-linked narratives. Before adding more thematic exposure, it is worth checking whether existing broad-market holdings already provide substantial AI sensitivity. In many cases, the decision is less about finding access to the theme and more about avoiding accidental overexposure to one expensive, capital-hungry segment of the market.
Anthropic’s S-1 is a Rorschach test, perhaps.
The anxiously awaited go-public filing from the $965 billion company is filled with contradictions that, perhaps, wholly encapsulate the AI era. It starts with rhetoric: That AI, “one of the most consequential technologies in human history,” will commence an “age of miracles,” while simultaneously rife with “existential risks to society.” AI, Anthropic says in its S-1, can do everything from shortening DMV lines to potentially curing cancer—and “may pose catastrophic [risks].”
The contradictions, of course, filter through to the numbers, including the headline revenue figure for the first half of 2026: $16.29 billion in revenue (reflective of sweeping demand for Claude), posed right next to Anthropic’s net loss of $80.7 billion for the six months leading up to June 2026. Anthropic could be the largest IPO ever—with one of the largest net losses ever recorded by a public or soon-to-be-public company. That mammoth chunk of red ink is due to an accounting charge related to convertible notes, but the company’s losing money nonetheless.
Reading this S-1, it’s clear: This is a cost-of-compute story. And how could it be anything else? The massive costs and demands for compute are woven throughout (the word “compute” appears more than 580 times in the document) and embody their own dichotomies. For one, Anthropic and Elon Musk may have their highly-public squabbles, but that isn’t stopping Anthropic from buying compute from SpaceX to the tune of up to $120.5 billion through 2029 (this far outstrips what we saw disclosed in SpaceX’s S-1). The company’s present and future keep it handcuffed to compute, no matter the cost and every step of the way: Case in point, Anthropic spent about $9.7 billion on compute in the first half of this year for just model R&D. All in, Anthropic’s planning to spend as much as $567 billion on compute.
A number that mildly shocked me (but shouldn’t have): The company, in H1 2026, also spent over $3 billion on marketing, which features more prominently in the S-1 than you’d expect. Among the risks, the filing reads: “We may not successfully build, expand, and deploy our sales and marketing organization in a timely manner, or we may not successfully hire, retain, train, and motivate our sales personnel… We have limited experience building, managing, and scaling a direct sales force…. If our marketing programs are not effective… our sales and revenue may grow more slowly than expected or decline.”
In short: It’s a reminder that no matter how much the AI boom booms, AI remains a product that needs to be sold, in a competitive market that (I would argue) runs the risk of being commoditized over time.
And throughout the filing is implicitly acknowledged a profound truth of this moment in tech and startups: Just because a company is richly valued, it doesn’t mean it’s mature. The phrase “limited experience” shows up to describe the unknowns around management, commercialization, and management of infrastructure. Which perhaps marks the greatest contradiction (and inevitability) of all: That the IPO that will test the AI bubble is a trillion-dollar leap of faith.
Anthropic, like OpenAI, has become an index fund of sorts for Silicon Valley (at least 100 VCs back the company, but none seem to own more than 5% per the filing). And the public markets responded positively yesterday, seemingly, to the news of Anthropic’s go-public. Billions, conservatively, are at stake, but the era of Rorschach projection in AI may be about to end.
Because when Anthropic starts trading, that’s when the real, quantifiable test begins.
See you tomorrow,
Allie Garfinkle
X: @agarfinks
Email: [email protected]
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.