Anthropic’s valuation exceeds $950 billion as major tech firms provide compute and funding.

On May 6 in San Francisco, at the Anthropic Developer Conference, Chief Product Officer Ami Vora did not unveil a new model on stage. She said the company will take over the entire computing capacity of SpaceX’s Colossus 1 data center within a month—exceeding 300 megawatts and 220,000 NVIDIA GPUs.

The audience fell silent for a few seconds. The developers were waiting for a model update; Anthropic provided compute power.

Six days later, The New York Times reported that Anthropic was in talks for a new funding round with a valuation of up to $950 billion. Bloomberg later added that the funding amount was estimated between $30 billion and $50 billion, potentially closing before the end of May, though no term sheet had been signed yet. If completed, Anthropic would surpass OpenAI’s $852 billion valuation record set in March.

From 380 billion in February's Series G to 950 billion in May, it took only three months. This valuation trajectory has no precedent in the history of technology commerce.

But what truly deserves questioning is the transfer of power behind the numbers: cloud providers and computing giants are entrusting their most scarce resources to the same company. An AI laboratory founded just five years ago has, in a remarkably short time, become the de facto controller of the entire AI infrastructure layer.

The term "metropole" is precisely appropriate.

On April 21, Amazon announced an additional investment of up to $25 billion in Anthropic. In return, Anthropic committed to spending over $100 billion on AWS over the next decade, including on Trainium-series chips.

Amazon disclosed that its $800 million investment in Anthropic from the same period is now worth over $70 billion. Of course, at least for accounting purposes, this gain is treated as unrelated to the business relationship.

On April 24, three days from now, Google followed suit, announcing an immediate cash investment of $10 billion and committing to add up to an additional $30 billion once Anthropic achieves performance milestones, bringing the total ceiling to $40 billion. At the same time, Google Cloud pledged to provide approximately 5 gigawatts of computing power over the next five years, while Anthropic committed to spending $200 billion on Google Cloud over the same period.

In May, SpaceX joined. All of Colossus 1's computing power was integrated into Claude within a month.

Including the Azure capacity provided by Microsoft and NVIDIA, along with Broadcom’s collaboration on custom TPU chips, Anthropic has secured over 20 gigawatts of computing power commitments in less than six months.

But the billing isn’t one-way—Amazon and Google’s money comes with “rebate clauses” attached; Anthropic’s funding must be spent back on its investors’ cloud services and chips at a scale of hundreds of billions of dollars. This isn’t venture capital-style “give money and let them burn it”; it’s more like a compute supplier securing large customers to lock in demand and absorb capacity. Beneath the veneer of venture capital lies a pre-sale contract for infrastructure.

A report by Fortune magazine at the end of April revealed another layer of awkwardness: nearly half of Google and Amazon’s “impressive AI profits” in the first quarter came from the paper gains on their stakes in Anthropic, not from their own operations. For Amazon, this investment gain was 1.4 times its AWS profit. The quarterly earnings of the world’s two largest cloud providers are now being driven by the valuation of a single startup.

Thus, the funding structure of the parent company is fully revealed: SpaceX provides immediate GPU computing power to alleviate Claude’s urgent throttling issues during peak periods; Amazon and Google provide funding, proprietary chips, and future computing capacity, while distributing Claude models through their cloud platforms and taking channel commissions; Microsoft provides Azure capacity and, despite its deep ties to OpenAI, has made Claude one of the flagship models on its Foundry platform; and Broadcom offers a hardware pathway with customized TPU chips, reducing Anthropic’s dependence on NVIDIA.

Four-way support, one suzerain.

Each party believes it is leveraging Anthropic to achieve strategic goals: Amazon aims to fill the capacity of its Trainium chips with Claude, Google seeks to use Anthropic to counterbalance the alliance between Microsoft and OpenAI, Microsoft wants to prevent customers from leaving for AWS due to inability to access Claude, and SpaceX seeks to generate rental income from its idle Colossus 1.

But each party, while delivering resources, is also surrendering bargaining power. When all the giants stake their strategic assets on the same company, that company ceases to be an optional partner—it becomes a necessity no one can afford to lose.

The reason the scale of support continues to increase is that Claude's growth has exceeded everyone's expectations.

On April 7, 2026, Anthropic announced that its annualized revenue surpassed $30 billion, more than tripling from approximately $9 billion at the end of 2025. Subsequent Q1 global LLM market data from Counterpoint Research showed Anthropic leading the market with a 31.4% revenue share, surpassing OpenAI’s 29.0%.

More striking is revenue efficiency: Anthropic has only 134 million monthly active users—about one-seventh of OpenAI’s—yet generates approximately $16 in monthly revenue per user, more than seven times that of OpenAI and over 160 times that of Meta.

Generating more revenue with fewer users indicates that Anthropic is capturing the segment of productivity users with the highest willingness to pay and the greatest usage intensity.

By the time of the developer conference in May, CEO Dario Amodei provided updated figures: annualized revenue had surged to over $44 billion, gross margins on inference had risen from 38% a year earlier to over 70%, and the number of enterprise customers spending over $1 million annually had grown from just over a dozen two years ago to more than 1,000, including eight of the Fortune 10 companies as Claude customers.

The revenue growth rate exceeded Amodei's own expectations; he said at the meeting, "I hope the 80-fold growth doesn't continue—it's too insane and too difficult to manage."

The core product driving this growth curve is Claude Code. This AI-powered coding agent generated over $2.5 billion in annualized recurring revenue by early 2026, capturing approximately 54% of the market share in its category. Once enterprises deeply integrate Claude Code into their development workflows, the migration costs become extremely high, creating a strong stickiness barrier.

But it was precisely at this moment that OpenAI took action.

In mid-April, a four-page internal memo from OpenAI’s Chief Revenue Officer, Denise Dresser, was published in full, accusing Anthropic of using the “gross method” for revenue recognition. Specifically, when enterprise customers purchase Claude services through AWS, Google Cloud, or Azure, Anthropic records the entire amount paid by the customer as its own revenue, including the portion that must be shared with the cloud providers. In contrast, OpenAI uses the “net method,” recording only the net revenue after deducting Microsoft’s share.

According to Dresser’s calculations, if adjusted to a net basis, Anthropic’s claimed annualized revenue of $30 billion should be approximately $22 billion, below OpenAI’s $25 billion during the same period. “Their story is built on fear and constraints,” Dresser wrote in the memo.

Media analysis has pointed out that both accounting methods are permissible under U.S. GAAP, with the key issue being the company’s role in the transaction. Anthropic’s position is that Claude is its core product, and the cloud platform serves merely as a distribution channel, thus justifying the gross method. However, the problem lies in the fact that Anthropic’s relationship with cloud platforms extends beyond distribution—Amazon and Google are simultaneously its shareholders, compute providers, and distribution partners. When funds circulate within this triangular relationship, distinguishing between “channel fees” and “investment returns” becomes inherently challenging.

Bank of America estimates that Anthropic’s total channel commissions paid to AWS and Google will reach $6.4 billion in 2026, more than tripling from $1.9 billion in 2025. Under the gross method, this $6.4 billion is counted as part of revenue; under the net method, it does not appear in revenue at all.

OpenAI’s move at this juncture is not merely an accounting debate. Both parties are advancing toward their IPOs. Once the S-1 registration statement is filed with the SEC, regulators will require both sides to restate their revenues under a unified framework.

Khosla Ventures partner Ethan Choi previously cut to the chase in an interview with Forbes: "If they all go public over the next few quarters, I'm not sure the SEC will allow two companies to use different accounting treatments for essentially the same type of revenue."

This $8 billion dispute over accounting is both a battle of valuations and a stress test both companies must face before going public.

The home country's greatest weakness is not its growth rate, but the fact that its growth figures have not yet been audited. Every number in the S-1 prospectus related to revenue quality, depreciation periods, and related-party transactions could become a ruler in the hands of critics.

A mother country becomes a mother country not because of what it says about itself, but because the empires that sustain it are reorganizing themselves along its path.

Amazon is betting its next decade of Trainium chips on Claude, Google is making its most advanced TPU capacity available to others, Microsoft is deeply integrating Claude into Azure while renegotiating its partnership with OpenAI, and SpaceX has included collaboration with Anthropic to develop multi-gigawatt orbital AI computing in its preliminary memorandum of understanding.

When stewards gradually entrust the allocation of core strategic assets to the same entity, that entity no longer needs to compete for influence—because influence is handed to it directly.

The FTC warned as early as its January 2025 AI collaboration report that when cloud credits, computing power commitments, equity stakes, and revenue-sharing terms are interwoven, they “shape competition, switching incentives, and access to commercially sensitive information.” More than a year later, this warning nearly perfectly anticipated Anthropic’s support structure.

Anthropic is turning this structural advantage into ecological control. Claude is the only cutting-edge model available simultaneously on AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry—the three leading cloud platforms.

This cross-cloud capability means enterprise customers are not locked into a specific model simply because they chose a particular cloud provider. Conversely, cloud providers that reject Anthropic risk losing enterprise customers. The cross-cloud portability of models has reversed the dynamics of supply and demand. Anthropic’s relative share of enterprise AI spending has surged from approximately 10% at the beginning of 2025 to over 65% by February 2026.

Another structurally significant variable is Claude Code. This programming tool now accounts for approximately 4% of all public commits on GitHub worldwide. Code is not an ordinary consumer product—it is the foundational syntax of the modern digital economy. Whoever controls the gateway to code generation holds the power to define the future of software production. As developers grow accustomed to Claude Code’s workflow, it’s not just Anthropic’s revenue that grows, but the entire center of gravity in software development shifts toward it.

OpenAI is not without counterarguments. The Dresser memo claims that OpenAI expects to have 30 gigawatts of computing power by 2030, while Anthropic will have only 7 to 8 gigawatts by the end of 2027. But the significance of computing power lies not in total capacity, but in timing. Colossus 1 is currently the only option available for immediate delivery, while OpenAI’s future computing power commitments are years away from realization. Enterprise customers make migration decisions now, not in 2030.

Meanwhile, OpenAI is advancing its own multi-cloud strategy and independent computing infrastructure, and its exclusive relationship with Microsoft has already begun to loosen. Anthropic’s multi-cloud architecture was designed this way from the start—it doesn’t need to break free from any constraints, as it has never tied itself exclusively to any single giant.

Microsoft’s agreement with OpenAI permits OpenAI to serve customers through any cloud provider, while Microsoft’s license to OpenAI’s models and product IP extends until 2032. This clause essentially reflects that both companies are preparing for an eventual separation, while Anthropic is strategically benefiting from the resulting rift.

The suzerain state is a temporary arrangement.

Anthropic relies on its patrons for computing power, while patrons rely on Anthropic for models. Each depends on the other, yet both remain cautious.

Musk retained a clause in the SpaceX agreement: if Anthropic’s AI engages in behavior harmful to humanity, SpaceX has the right to reclaim computing resources. This clause appears as a safety statement but is, in fact, a substantial control arrangement and a microcosm of the power dynamics among all parties in the overarching system.

But at least within the current time window, the side with the most supporters is driving the industry forward according to its vision. Anthropic has transformed in five years from a research lab into the central orchestrator of AI infrastructure.

If the $950 billion valuation financing is completed by the end of May and a subsequent IPO actually takes place, this story will enter an entirely different chapter. The previous valuation was a pricing game in the private market; the IPO prospectus will be the first true external scrutiny.

At that moment, the $8 billion accounting dispute in the Dresser memo would no longer be an attack by one company against another, but a formal inquiry on the SEC’s desk. Amazon and Google’s earnings would shift from book value appreciation to related-party transactions requiring itemized explanation in audit reports.

To sustain long-term dominance, a dominant player must transform external inputs into self-sustaining capabilities. Anthropic’s reasoning gross margin increasing from 38% to over 70% forms the most solid narrative foundation for its business model. However, if the distinction between gross and net revenue methods casts doubt on revenue quality, or if high compute costs and slowing growth prevent rapid loss reduction, this narrative faces repricing.

Even when accounting for all these uncertainties, one irreversible fact remains: power in the AI industry is being redistributed along the compute supply chain, cloud distribution channels, and developer ecosystems. At this current juncture, the coordinate closest to the center of power is called Anthropic.

This article is from the WeChat public account "Silicon Starlight," authored by Yuan Tai.