Anthropic’s planned public offering is revealing a central contradiction in the economics of frontier artificial intelligence: the faster an AI company expands, the more heavily it may have to rely on the technology giants it ultimately competes with. The company’s confidential IPO prospectus shows that Amazon and Google are simultaneously investors, computing suppliers, distribution partners and direct competitors. In 2025, 47% of Anthropic’s sales were routed through the cloud marketplaces of those two companies, which also collected customer payments on its behalf. At the same time, Anthropic has committed to enormous infrastructure spending as it attempts to satisfy rapidly increasing demand for its models.
The company expects to spend at least $518 billion over a decade on AI infrastructure with six partners, with about 80% of those commitments either non-cancellable or payable regardless of actual usage. The figures illustrate why the frontier AI business cannot be assessed simply by looking at revenue growth or model performance. Its expansion depends on access to computing capacity, and that access is increasingly controlled by a small number of companies with interests that overlap with Anthropic’s own.
The dependence matters because computing is no longer an ordinary operating expense for the companies developing the most advanced AI systems. It is becoming a strategic constraint on growth. Anthropic says in its filing that future demand for advanced AI systems is likely to exceed available computing supply, meaning access to processors and data-centre capacity could determine how quickly the company can expand. That changes the balance between an AI developer and its cloud suppliers. Anthropic needs Amazon and Google not merely to host applications but to provide the infrastructure required to train and operate increasingly demanding models. Amazon and Google, meanwhile, have their own AI ambitions and are developing competing products and services.
The relationship therefore combines cooperation and competition in a way that is unusual even by technology-industry standards. Anthropic can use those relationships to accelerate distribution and secure computing, but the same arrangements mean that a significant part of its growth is tied to companies whose commercial interests extend well beyond supporting Anthropic. The IPO prospectus consequently exposes a structural feature of the AI economy that rapid revenue growth can obscure: technological independence is difficult to achieve when the infrastructure required for technological leadership is concentrated in the hands of a few powerful suppliers.
Computing Has Become a Strategic Dependency
The scale of Anthropic’s infrastructure commitments helps explain why the company has accepted such extensive obligations. Its filing identifies at least $111.1 billion of planned spending with Google, $110 billion with Amazon and $31.4 billion with Microsoft over periods extending seven to ten years. These agreements provide Anthropic with something increasingly valuable in the AI race: predictable access to computing capacity. Without such commitments, the company could face a situation in which demand for its services grows faster than its ability to secure the processors and data-centre resources necessary to serve customers.
The problem is particularly acute because frontier AI systems require vastly more computing power than conventional software applications. Training a new generation of models can involve enormous amounts of specialised hardware, while operating those models for millions of users creates a continuing demand for inference capacity. Long-term infrastructure agreements therefore protect Anthropic against one form of supply uncertainty. But they create another form of exposure because the company remains financially committed even if its own demand changes. The very contracts that provide certainty about infrastructure can therefore reduce flexibility if technology, prices or customer demand develop differently from current expectations.
That tension becomes clearer when Anthropic’s relationships with its cloud partners are viewed alongside its distribution model. Nearly half of its 2025 sales passed through Amazon and Google, meaning these companies have influence over both the infrastructure on which Anthropic operates and the channels through which a substantial share of its business reaches customers. This does not mean that Anthropic lacks commercial independence, but it does mean that its growth is intertwined with the platforms of companies that have enormous bargaining power.
Amazon and Google can benefit from Anthropic’s growth because it increases demand for cloud services and related infrastructure, while Anthropic benefits from their scale and technical resources. The arrangement can therefore be mutually profitable. The potential vulnerability arises when one side becomes difficult to replace. If access to computing is scarce and alternative suppliers cannot provide equivalent capacity, an AI developer may have limited freedom to change infrastructure providers even when commercial conditions change. Anthropic’s situation illustrates why the AI industry is developing a new form of dependence: the most important strategic asset may not be the model alone, but the infrastructure required to keep that model running at scale.
Rapid Growth Raises a Different Financial Question
Anthropic’s extraordinary revenue growth has strengthened the case for investing heavily in infrastructure, but it does not remove the financial risk created by that spending. A frontier AI company can generate rapidly rising sales while simultaneously committing enormous amounts of capital to future computing capacity. The underlying assumption is that demand for advanced AI will continue expanding quickly enough to justify those commitments. If that assumption holds, securing infrastructure early could give Anthropic an important advantage because it would have capacity available when competitors and customers are competing for the same resources. If demand grows more slowly, however, fixed infrastructure obligations could become a heavier burden.
This is particularly relevant because the AI industry remains in a period of rapid technological change. Computing hardware becomes more efficient, model architectures evolve and companies continually search for ways to reduce the amount of computing required for a given task. An infrastructure agreement that looks necessary today could therefore carry a different economic value several years from now. Anthropic’s challenge is not simply to predict how much AI will be used, but to predict which kinds of computing will remain economically valuable as the technology changes.
The IPO prospectus therefore reveals a business model in which technological ambition and infrastructure dependence are closely connected. Anthropic needs massive computing commitments to compete at the frontier, yet those commitments increase its exposure to a small group of suppliers. Amazon and Google are particularly significant because they occupy several positions in the relationship at once: they provide infrastructure, facilitate distribution, invest in Anthropic and compete against it.
That overlap does not make the partnership inherently unstable, but it does create strategic complexity that investors will have to assess alongside revenue growth and model capabilities. Anthropic’s public offering is consequently about more than the valuation of an AI company. It provides a test of whether frontier AI developers can convert extraordinary demand into sustainable businesses while reducing the infrastructure dependence created by the very scale at which they operate. The central economic question is becoming clearer: as AI companies become more powerful, can they also become sufficiently independent from the technology giants whose infrastructure makes that power possible?
(Adapted from Business-standard.com)
