Forty-seven percent of revenue passed through cloud channels run by AI rivals
Reuters' review of Anthropic's IPO filing shows that sales through Amazon and Google cloud marketplaces totaled about $2.16 billion in 2025, or 47% of annual revenue.
That does not mean Amazon and Google directly bought 47% of Anthropic's revenue. Customers bought access to Claude through those marketplaces while Anthropic set prices and delivered the service.
The unusual part is the role overlap. Amazon and Google are also major Anthropic investors, critical compute suppliers and developers of competing AI systems. One counterparty can be Investor, Compute Supplier, Distributor and Competitor at the same time.
Why use a competitor's distribution? Enterprise reach is difficult to reproduce
Anthropic says selling Claude through Amazon, Google and Microsoft cloud platforms lets it tap large existing sales networks and customers that already buy through those providers. That can accelerate market penetration compared with rebuilding every enterprise purchasing relationship from scratch.
Its public partner strategy reinforces the same logic. In March, Anthropic committed an initial $100 million to the Claude Partner Network for training, technical support and joint market development and said Claude was available through AWS, Google Cloud and Microsoft.
By June, Anthropic said more than 40,000 firms had applied to join the network and more than 10,000 consultants had earned Claude certification. The company is not only selling a model; it is expanding the organizations that can put the model into production.
The same companies appear again on the supply side
Anthropic expects to spend at least $518 billion over a decade building AI infrastructure with six partners, according to the IPO materials reviewed by Reuters. Roughly 80% is non-cancelable or requires payment regardless of actual usage.
Long-term infrastructure obligations include at least $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft. The platforms helping bring customers to Claude also supply much of the compute needed to run it.
Anthropic argues that compute availability is becoming a principal constraint on advanced AI development, so locking in capacity can be strategically valuable. The trade-off is that distribution speed and compute security come with large fixed obligations.
Partnership is both an accelerator and a dependency
Anthropic presents these relationships as a competitive advantage, but its filing also says reliance on a limited set of partners and suppliers creates complex incentives and could affect access to compute.
Cloud partners can gain visibility into pricing and commercial terms, while also deciding how they allocate infrastructure and how aggressively they sell their own products. Concentrating customer access and production resources in the same counterparties can change bargaining power.
Customer concentration sits on top of that partner concentration: two unnamed customers each accounted for 12% of Anthropic's 2025 revenue. The company therefore has to manage customer concentration and infrastructure/distribution concentration at the same time.
Multi-cloud can be a competitive strategy, not only a resilience architecture
Offering Claude across competing clouds can reduce the degree to which any single provider controls distribution or compute access. This is broader than technical failover.
Anthropic also says it is moving beyond a cloud-only model toward more dedicated infrastructure and directly leased hardware.
The strategic question becomes which capabilities to rent, which to internalize and how much optionality to preserve across partners. Infrastructure architecture and competitive architecture start to overlap.
The same question applies to any company building an AI business
A B2B product can reach customers much faster by using an established cloud marketplace, systems integrator or consulting network than by building direct distribution from zero.
But partner selection becomes more specific when the partner also sits in the supply chain. Which capability are you borrowing? Who owns the customer relationship? Who can see pricing or usage? Does the product survive if one partner changes its priorities?
Anthropic's structure shows that partnership can shape the operating model and the competitive map at the same time.
BANSEOG VIEW | Your most important growth infrastructure may also be your competitor
The lesson is not simply that Big Tech dependence is bad. Amazon and Google's infrastructure and distribution can accelerate Anthropic's growth, while that same dependence creates real strategic constraints.
The deeper shift is that Partner and Competitor are no longer mutually exclusive categories. A single company can sit across Capital → Compute → Distribution → Competition.
Competitive advantage may therefore depend less on eliminating partners and more on using their assets while preserving core capability, customer access, optionality and bargaining power.
BANSEOG VIEW
Banseog View — Investor → Compute → Distribution → Competitor
In 2025, 47% of Anthropic's revenue was routed through Amazon and Google cloud marketplaces; both companies are also investors, compute suppliers and AI competitors.
Anthropic plans at least $518B of AI infrastructure spending over a decade, with roughly 80% non-cancelable or payable regardless of usage.
Evaluating an AI company therefore requires looking beyond model quality to who controls compute, distribution and the company's strategic optionality.
SOURCES
Primary sources and references
- Reuters — Anthropic IPO filing shows dependence on Big Tech partners
Sep. 29, 2026. Confirms 47% of 2025 revenue was routed through Amazon and Google cloud marketplaces; those firms also act as investors, compute suppliers, distributors and AI competitors. It also reports two unnamed customers at 24% combined revenue and more than $417B in long-term compute commitments.
- Reuters — Anthropic's $518B AI infrastructure commitments
Sep. 29, 2026. Confirms at least $518B of planned infrastructure spending over a decade, with roughly 80% non-cancelable or payable regardless of usage, including $111.1B with Google, $110B with Amazon and $31.4B with Microsoft.
- Anthropic — Claude Partner Network
Mar. 12, 2026. Confirms the initial $100M partner investment, availability across AWS, Google Cloud and Microsoft, and support for partner training, technical enablement and market development.
- Anthropic — Services Track and Partner Hub
Jun. 3, 2026. Confirms more than 40,000 firms applied to join and more than 10,000 consultants earned Claude certification, with emphasis on production deployment experience.
The 47% figure refers to revenue sold to end customers through Amazon and Google cloud marketplaces; it is not the share of revenue directly purchased by Amazon and Google. The $518B figure is a minimum planned spend over roughly a decade across cloud, compute and infrastructure arrangements with six partners, not a one-time cash outlay or already-incurred cost. Reuters reports that about 80% is non-cancelable or payable regardless of usage. Partner → Dependency → Bargaining Power is Banseog's analytical synthesis of the disclosed relationships.