A Cap Table That Doubles as a Sales Pipeline
When Anthropic eventually prices its listing, the order book will look less like a conventional technology float and more like a seating chart at a very small dinner party. Several of the largest shareholders in the company behind the Claude models are also among its most consequential commercial counterparties — the cloud providers that rent it computing power, the platforms that resell its software, and the enterprise vendors that build products on top of it.
That overlap is not an accident. It is the architecture of the modern AI capital stack.
Anthropic, founded in 2021 by a group that included former OpenAI researchers Dario and Daniela Amodei, has spent its short corporate life raising enormous sums from a narrow band of institutions. Amazon and Google are the two giants of that band. Between them, they have committed billions: Amazon’s total investment has been reported at up to $8bn across successive tranches, while Google’s commitment, including its initial stake and a subsequent multi-billion-dollar arrangement, has been put at around $3bn. Venture firms including Lightspeed, ICONIQ, Fidelity, Menlo Ventures, Coatue, General Catalyst, Spark Capital and Salesforce Ventures have taken meaningful positions along the way, with later rounds reportedly valuing the business at figures that have climbed from roughly $61.5bn in early 2025 to far higher numbers since.
A public listing turns all of that private goodwill into a quoted price. It also turns a set of overlapping commercial relationships into something investors will need to underwrite.
Landlord, Backer, Reseller
The Amazon relationship is the clearest illustration of how tangled this has become.
Amazon is an equity holder in Anthropic. It is also the company’s principal cloud supplier, with Anthropic training and serving models on Amazon Web Services infrastructure, including AWS’s own Trainium silicon. And through Amazon Bedrock, AWS resells Claude to enterprise customers who never have to touch Anthropic directly. Three revenue lines, one counterparty.
Google occupies a similar position from a different angle. Beyond its stake, Google Cloud offers Claude models to customers through its Vertex AI platform, and Anthropic has used Google’s tensor processing units for workloads. Google is simultaneously funding a company whose success reduces dependence on its own in-house models, and selling that company’s output to its own cloud customers.
Neither arrangement is improper. Both were reviewed and cleared by competition authorities, including the UK’s Competition and Markets Authority, which examined the Google and Amazon tie-ups and concluded they did not raise the jurisdictional concerns that dogged Microsoft’s relationship with OpenAI.
But the economics deserve a closer look than they usually get.
Consider what happens in the ordinary course of business. Anthropic spends money with Amazon and Google on compute. That spending shows up as revenue for two of its largest shareholders. When Anthropic signs a large enterprise deal routed through Bedrock or Vertex, the cloud partner captures hosting revenue while Anthropic books the model revenue — and the shareholder’s stake appreciates. When Anthropic raises again at a higher valuation, those same institutions mark up their holdings.
The incentives are aligned, in other words, in almost every direction at once. That is either a virtuous flywheel or a circular financing arrangement, depending on your tolerance for ambiguity.
What Public Shareholders Are Actually Buying
Anthropic’s pitch to public markets rests on a genuinely impressive commercial record. The company crossed a $1bn annualised revenue run rate faster than most enterprise software businesses manage in a decade, driven heavily by corporate adoption and by coding use cases that have proven stickier and more monetisable than early consumer chatbots. Its Claude models sit inside products from Salesforce, GitLab, Notion, Zoom, Snowflake and Palantir, among others.
Yet the concentration risk is real, and it cuts both ways.
A meaningful share of Anthropic’s revenue reaches the market through a handful of channels, two of which are controlled by shareholders. A change in how Amazon or Google prioritises their own first-party models, or in the commercial terms they extend to a partner that is also a competitor’s supplier, could move Anthropic’s numbers materially. Public investors typically demand disclosure on customer concentration. Here, the concentration runs through the ownership structure itself.
There is also the question of what a listing does to the pace of capital formation. Anthropic’s funding needs are extraordinary — training frontier models consumes cash at a rate that has no real precedent outside state-directed programmes. Staying private lets a company raise from a cosy group of believers who understand the burn. Going public opens the door to index funds, pension schemes and retail buyers who are being asked to underwrite multi-year losses in exchange for a claim on a technology whose economics are still being discovered.
Governance: The Trust Meets the Tape
Anthropic has structured itself as a public benefit corporation with a Long-Term Benefit Trust, an arrangement designed to give a body of trustees meaningful influence over the company’s direction and to keep safety considerations in the boardroom. Co-founder Dario Amodei has long argued that frontier AI development carries risks serious enough to justify slowing down, and the company has built its public identity around responsible scaling.
Public equity markets are not natural habitats for that kind of institutional design.
Dual-class structures, golden shares and benefit-corporation charters all exist, and investors have learned to price them. But the specific tension here is sharper than usual: a shareholder base that includes some of the world’s most aggressive distributors of AI products, alongside trustees whose mandate may counsel restraint, alongside public funds that want quarterly growth. Something eventually gives.
None of this makes the offering uninvestable. Scarcity value is real, and Anthropic is one of a very small number of assets giving investors direct exposure to frontier model development outside a handful of mega-cap technology stocks. The demand for that exposure is not in doubt.
The diligence question is narrower and more useful: how much of Anthropic’s growth is powered by partners who also own it, and what happens to the growth rate if those partners’ priorities shift?
Why it Matters
Anthropic’s float is a test case for whether public markets can price an AI company whose most important relationships are also its most important conflicts. If Amazon and Google can be simultaneously landlord, distributor, customer and shareholder without eroding the value of the business being sold to outside investors, then the strategic-partnership model becomes the template for the entire sector — and every subsequent AI listing will be judged against it. If the concentration proves harder to disentangle than the prospectus suggests, investors will discover that they bought into a company whose revenue and ownership are drawn from the same well. Either way, the ordinary shareholder is being asked to underwrite a business in which the smartest money in the room has already agreed on the price.
