On May 6, at the Anthropic Developer Conference in San Francisco, Chief Product Officer Ami Vora did not announce a new model on stage. She stated that the company would take over all computing power from 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. Developers were waiting for model updates, but Anthropic delivered computing power instead.

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 size would be between $30 billion and $50 billion, potentially completing by the end of May, though no investment term sheet had been signed yet. If successful, Anthropic would surpass OpenAI's $852 billion valuation record set in March this year.

From $380 billion in its Series G round in February to $950 billion in May, it took only three months. No precedent for such a valuation curve exists in tech business history.

But what truly warrants scrutiny is the power transfer happening behind the numbers: cloud providers and computing power giants are handing over their scarcest resources to the same company. A five-year-old AI lab has, in an extremely short time, become the de facto controller of the entire AI infrastructure layer.

The term "suzerain" fits perfectly.

01

Four Tributes, One Suzerain

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

Amazon simultaneously disclosed that its $800 million investment in Anthropic in 2025 was now valued at over $70 billion. Of course, at least on paper, this gain was treated as unrelated to business relationships.

Three days later, on April 24, Google followed suit, announcing an immediate $10 billion cash investment and committing up to an additional $30 billion after Anthropic met performance milestones, totaling up to $40 billion. Meanwhile, Google Cloud committed to providing approximately 5 gigawatts of computing power over the next five years, while Anthropic pledged to spend $200 billion on Google Cloud over the same period.

In May, SpaceX joined in. All computing power from Colossus 1 would be integrated with Claude within a month.

Combined with previous Azure capacity provided by Microsoft and NVIDIA, as well as the TPU custom chip collaboration involving Broadcom, Anthropic had secured computing power commitments totaling over 20 gigawatts in less than six months.

But the bills are not one-way. Amazon's and Google's money "comes with rebate clauses for every dollar," meaning Anthropic must spend hundreds of billions back on the investors' cloud services and chips. This is not VC-style "giving money to burn" but more akin to computing power suppliers finding major clients for their capacity, using financing to lock in demand and using demand to digest capacity. Beneath the venture capital veneer lies an infrastructure presale contract.

Fortune magazine highlighted another layer of awkwardness in a late-April report: nearly half of Google's and Amazon's "impressive AI profits" in Q1 came from the book value appreciation of their Anthropic stakes, not from their own operations. For Amazon, this investment gain was 1.4 times its AWS profit. The quarterly earnings reports of the world's two largest cloud providers were being dictated by a startup's valuation.

Thus, the suzerain's support structure fully emerged: SpaceX provides immediate GPU computing power, alleviating Claude's throttling crises during peak demand; Amazon and Google provide funding, self-developed chips, and long-term computing power while distributing Claude models through their cloud platforms and taking channel commissions; Microsoft provides Azure capacity and, despite its deep partnership with OpenAI, Claude has become one of the main models on its Foundry platform; Broadcom provides a hardware pathway for custom TPU chips, freeing Anthropic from complete dependence on NVIDIA.

Four tributaries, one suzerain.

Each party believes it is leveraging Anthropic to achieve strategic goals: Amazon aims to fill Trainium chip capacity with Claude, Google seeks to counter Microsoft's alliance with OpenAI through Anthropic, Microsoft prevents customer loss to AWS due to Claude's unavailability, and SpaceX seeks rental returns for its idle Colossus 1.

But each party, while delivering resources, is also ceding bargaining power. When all giants stake their strategic assets on the same company, that company ceases to be an optional partner—it becomes an indispensable necessity.

02

Blade Flashes on the Eve of IPO

The reason support continues to escalate lies in Claude's growth exceeding all expectations.

On April 7, 2026, Anthropic announced annualized revenue exceeding $30 billion, more than tripling from approximately $9 billion at the end of 2025. Counterpoint Research's subsequent Q1 global LLM market data showed Anthropic surpassing OpenAI with a 31.4% revenue share versus OpenAI's 29.0%, claiming the global top spot.

More striking was revenue efficiency: Anthropic had only 134 million monthly active users, about one-seventh of OpenAI's, but generated approximately $16 in average monthly revenue per user—over seven times OpenAI's and more than 160 times Meta's.

Generating more revenue with fewer users indicates Anthropic captures the highest-paying, most intensive productivity users.

By the May developer conference, CEO Dario Amodei provided updated figures: annualized revenue had surged past $44 billion, inference gross margins climbed from 38% a year ago to over 70%, and enterprise clients spending over $1 million annually grew from a dozen two years ago to over 1,000, including eight Fortune 10 companies as Claude customers.

The revenue growth rate exceeded even Amodei's expectations. "I hope 80x growth doesn't continue—that would be too crazy and hard to manage," he said at the conference.

The core driver of this growth is Claude Code. This programming agent tool exceeded $2.5 billion in annualized recurring revenue in early 2026, holding about 54% of the market share among similar tools. Once enterprises deeply integrate Claude Code into their development workflows, migration costs become prohibitively high, creating sticky barriers.

But precisely at this moment, OpenAI made its move.

In mid-April, OpenAI Chief Revenue Officer Denise Dresser's four-page internal memo was published in full, directly accusing Anthropic of using "gross revenue" accounting. Specifically, when enterprise clients purchase Claude services through AWS, Google Cloud, or Azure, Anthropic books the entire amount paid by clients as its own revenue, including channel fees distributed to cloud providers. In contrast, OpenAI uses "net revenue" accounting, recording only net income after Microsoft's share is deducted.

According to Dresser's calculations, if adjusted to net terms, Anthropic's claimed $30 billion annualized revenue would be approximately $22 billion, below OpenAI's $25 billion for the same period. "Their story is built on fear and restriction," Dresser wrote in the memo.

Media analysis pointed out that both accounting methods are legal under U.S. GAAP, with the key lying in the company's role in transactions. Anthropic's stance is that Claude is its core product and cloud platforms are merely distribution channels, justifying gross revenue accounting. The issue arises because Anthropic's relationship with cloud platforms extends beyond distribution—Amazon and Google are simultaneously its shareholders, computing power suppliers, and distribution channel partners. When funds circulate within this triangular relationship, distinguishing "channel fees" from "investment returns" becomes a difficulty (challenge).

Bank of America provided a calculation: Anthropic's total channel commissions paid to AWS and Google are expected to reach $6.4 billion in 2026, more than tripling from $1.9 billion in 2025. Under gross revenue accounting, this $6.4 billion counts as part of revenue; under net revenue accounting, it wouldn't appear in revenue at all.

OpenAI's move at this juncture is not merely an accounting debate. Both companies are advancing toward IPOs. Once S-1 filings are submitted to the SEC, regulators will force both to restate revenue under a unified framework.

Ethan Choi, a partner at Khosla Ventures, previously told Forbes: "If they both IPO in the coming 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 accounting dispute is both a valuation battle and a stress test the companies must pass before rushing toward public markets.

The suzerain's greatest vulnerability lies not in growth speed but in growth figures not yet audited. Every line in the S-1 prospectus related to revenue quality, depreciation schedules, and related-party transactions could become a measuring stick for skeptics.

03

Voice Comes From Being Handed Power

The suzerain became the suzerain not because of what it said but because the empires supporting it are reorganizing themselves along its path.

Amazon has bet Trainium chips' next decade on Claude, Google has surrendered its most advanced TPU capacity, Microsoft is deeply integrating Claude into Azure while renegotiating its partnership with OpenAI, and SpaceX has included developing multi-gigawatt orbital AI computing power with Anthropic in its preliminary intentions.

When supporters sequentially hand over configuration rights to core strategic assets to the same entity, that entity no longer needs to seize voice—it is handed power.

The FTC warned about this structure in its January 2025 AI cooperation report: when cloud credit lines, computing power commitments, equity stakes, and revenue-sharing clauses interlock, they "shape competition, shift incentives, and grant access to commercially sensitive information." Over a year later, this warning almost precisely predicted Anthropic's support system.

Anthropic is now converting this structural advantage into ecosystem control. Claude is the only cutting-edge model simultaneously running on AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry—the three major cloud platforms.

This cross-cloud capability means enterprise clients won't be locked into a specific model by choosing a cloud provider. Conversely, cloud providers refusing Anthropic risk abandonment by enterprise clients. The model's cross-cloud portability reverses supply-demand dynamics. Among enterprise AI expenditures, Anthropic's relative share has jumped from about 10% in early 2025 to over 65% by February 2026.

Another structurally significant variable is Claude Code. This programming tool accounts for approximately 4% of all public GitHub submissions globally. Code is no ordinary consumer product—it's the underlying grammar of the modern digital economy. Whoever controls code generation entry points defines software production methods. As developers grow accustomed to Claude Code's workflow, not only does Anthropic's revenue grow, but the entire software development paradigm shifts toward it.

OpenAI is not without countermeasures. Dresser's memo claimed OpenAI expects to have 30 gigawatts of computing power by 2030, while Anthropic will have only 7–8 gigawatts by the end of 2027. But computing power matters not in total volume but in timing. Colossus 1 is the only immediately deliverable option, while OpenAI's long-term computing power pledges won't materialize for years. Enterprise migration decisions happen now, not in 2030.

Meanwhile, OpenAI is advancing its multi-cloud strategy and computing power independence, loosening its exclusivity with Microsoft. Anthropic's multi-cloud architecture was designed from the start—it never bound itself to any single giant.

Microsoft's partnership agreement with OpenAI allows OpenAI to serve clients through any cloud provider while extending Microsoft's IP license for OpenAI models and products until 2032. This clause essentially means both companies are preparing for an inevitable split, and Anthropic is reaping strategic space from this fissure.

04

The Suzerain Is a Time-Limited Game

The suzerain is a temporary pattern (structure).

Anthropic relies on supporters for computing power; supporters rely on Anthropic for models. Both need and guard against each other.

Musk retained a clause in the SpaceX agreement: if Anthropic's AI harms humanity, SpaceX reserves the right to reclaim computing power. This clause appears to be a safety declaration but is in essence (substantially) a meaningful control arrangement—a microcosm of game theory (gameplay) across the entire suzerain system.

But in the current time window, the party with the most supporters is driving the industry forward according to its will. In five years, Anthropic has transformed from a research lab into AI infrastructure's central dispatcher.

If the $950 billion valuation funding round completes by late May and a subsequent IPO materializes, this story will enter an entirely new chapter. Previous valuations were private market pricing games; the IPO prospectus represents the first true external scrutiny.

At that moment, the $8 billion accounting dispute in Dresser's memo will cease to be one company's attack on another and become a formal SEC inquiry on the table. Amazon's and Google's gains will shift from book value appreciation to related-party transactions requiring itemized explanation in audit reports.

To sustain the suzerain long-term, it must convert nourishment into self-generating capabilities. Anthropic's inference gross margins climbing from 38% to over 70% form the most solid narrative foundation of its business model. But if revenue quality doubts persist due to gross vs. net accounting differences, or if computing costs remain high and growth slows without rapid loss compression, this narrative faces revaluation.

Even if all these uncertainties are laid on the table, one fact is irreversible: the power of the AI industry is being redistributed along the computing supply chain, cloud distribution channels, and developer ecosystems. And at the current juncture, the coordinate closest to the center of power is called Anthropic.