Anthropic’s S-1 confirms something that’s been an open secret in AI circles for a while: Amazon and Google aren’t just two of Anthropic’s biggest cloud bills, they’re also two of its biggest shareholders. The filing puts numbers on exactly how tangled that relationship is — and flags, in its own risk factors, that the tangle itself is a risk.
How Amazon and Google became investors
Google’s involvement goes back furthest. We reported in 2025 that Google had quietly built a 14% stake in Anthropic, capped at 15% under its agreement with the company, with no board seat or voting rights attached. The S-1 shows that relationship has only deepened since: Google bought roughly $10 billion of Series G-2 preferred stock in March 2026, followed by another $10 billion of Series H preferred stock in May 2026.
Amazon’s position goes back even further and has grown larger. Amazon first put money in via convertible notes starting in 2023, totaling around $8 billion by 2025. In the second quarter of 2026 it converted that relationship into two $5 billion preferred stock purchases, and as of June 30, 2026, the S-1 shows Amazon still holding $6.2 billion in notes that will convert at the IPO — subject to a cap that keeps Amazon’s total stake at or below 12% of the company.
Between the two of them, analyst estimates cited around the filing put Amazon’s post-IPO stake somewhere in the 15-21% range and Google’s around 10-15%, which at a $2 trillion valuation would make Amazon’s position worth $300-400 billion and Google’s $200-300 billion on paper — larger than most companies’ entire market caps, just as a minority stake in one AI lab.
Investor and customer, at the same time
What makes the arrangement unusual isn’t just the size of the stakes, it’s that Amazon and Google are also Anthropic’s two largest suppliers. AWS is designated Anthropic’s primary cloud provider, running workloads including Project Rainier, the Trainium2-based supercomputer cluster in Indiana that AWS has called the world’s most powerful AI training system. Google, meanwhile, supplies Anthropic with TPU capacity — Anthropic struck a deal for up to one million Google TPUs worth tens of billions of dollars, on top of its Google Cloud usage.
The S-1 puts hard numbers on those commitments: Anthropic has agreed to spend at least $110 billion with Amazon through April 2036, and at least $114.5 billion with Google through July 2033 — both described as owed “regardless of actual usage,” meaning Anthropic is on the hook for the spending even if it doesn’t end up needing that much compute. Both companies also act as sales channels, taking a cut of revenue that flows through their cloud marketplaces; combined, Amazon and Google’s marketplaces accounted for 47% of Anthropic’s 2025 revenue and 48% in the first half of 2026.
So Amazon and Google end up touching Anthropic’s business three separate ways: as equity holders whose stakes rise with Anthropic’s valuation, as infrastructure suppliers collecting on multi-year, non-cancelable compute contracts, and as distribution partners taking a fee on top. Interestingly, Google DeepMind’s Demis Hassabis was reportedly an early personal investor in Anthropic, even as Google runs a competing model in Gemini, and the irony of Google and Amazon as investors in a company whose AI could pose existential risk hasn’t gone unnoticed as the broader AI funding picture has gotten more interconnected.
The conflict-of-interest risk factor
Anthropic’s S-1 addresses this directly in its risk factors, warning that it relies on a limited number of cloud providers and suppliers, some of which are also strategic investors and customers, and that the overlap “may create complex situations that could lead to conflicts of interest and adversely affect our ability to secure computing power.”
The practical concern cuts a few ways. Amazon and Google have leverage over Anthropic as suppliers — they can shape pricing, prioritize capacity for their own AI products, or slow-walk delivery — while simultaneously having a financial stake in Anthropic’s success as shareholders, which could blunt their incentive to negotiate hard against their own portfolio company, or conversely give them outsized influence over strategic decisions precisely because Anthropic depends on them for both capital and compute. It’s also a double-edged sword if the relationship ever sours: Amazon and Google are also both building their own competing frontier models, in Nova and Gemini respectively, so the same companies keeping Anthropic’s servers running are also racing to make Claude obsolete.
What it means for the IPO
For investors, the Amazon and Google relationships are something of a mixed signal. On one hand, having two of the biggest cloud companies in the world locked into over $224 billion of minimum committed spending gives Anthropic a degree of funding certainty that smaller AI labs don’t have, and the FTC has previously examined whether these arrangements function as de facto acquisitions, which at minimum signals regulators see them as consequential. On the other hand, it means a meaningful share of Anthropic’s revenue and future compute runs through two counterparties who are also competitors, also shareholders, and whose incentives won’t always point the same direction as outside investors’. That circularity — the same dollars moving between investment, infrastructure spend, and revenue — is exactly the kind of thing the S-1’s risk factors exist to flag, and exactly the kind of thing public market investors will be pricing in once ANT starts trading.