Author：TechFlow
Author: Deep Tide TechFlow
Deep Tide Guide:On April 6, Anthropic announced that its annualized revenue had surpassed $30 billion, more than doubling from $9 billion at the end of 2025. The number of enterprise customers with annual spending exceeding one million US dollars doubled from 500 to 1,000 within two months. On the same day, Broadcom's SEC filing confirmed that Anthropic will acquire approximately 3.5 gigawatts of next-generation TPU computing power starting in 2027, its largest single computing power commitment to date.
Anthropic released two major figures on the same day.
According to Anthropic's official blog on April 6, the company's annualized revenue has exceeded $30 billion, more than doubling from approximately $9 billion at the end of 2025. On the same day, a filing with the SEC by Broadcom revealed that Anthropic will acquire approximately 3.5 gigawatts (GW) of next-generation TPU computing power from Broadcom starting in 2027, as part of an expanded collaboration among the three parties (Broadcom, Google, and Anthropic).
Broadcom shares rose about 3% in after-hours trading.
(Image source: X user @damianplayer)
From $1 billion to $30 billion in 14 months, Claude Code is the core engine.
Anthropic's revenue curve is unprecedented in the AI industry. According to publicly disclosed data and timelines reported by Bloomberg and other media outlets: approximately $1 billion in December 2024, approximately $4 billion in mid-2025, approximately $9 billion by the end of 2025, approximately $14 billion in February 2026, approaching $19 billion in early March, and officially confirmed to have surpassed $30 billion on April 6.
In a statement, Anthropic CFO Krishna Rao said the company is making "the most significant computing power commitment to date to match unprecedented growth."
Client-side data is equally impressive. During its Series G funding round in February, Anthropic had 500 enterprise clients with annualized revenue exceeding $1 million; less than two months later, that number more than doubled to over 1,000. According to Anthropic's previous disclosures, the core driver of this growth is Claude Code, a product launched in May 2025, which had already generated over $2.5 billion in annualized revenue by February 2026.
For reference, Sacra estimates OpenAI's annualized revenue at approximately $25 billion (as of February 2026). According to Epoch AI analysis, Anthropic's annualized growth rate since surpassing $1 billion in revenue is approximately 10 times, while OpenAI's is approximately 3.4 times over the same period. Based on this trend, the revenue gap between the two companies may appear in mid-2026.
Please note: The above figures are all annualized revenue (run-rate revenue), which is an estimated value of recent monthly revenue multiplied by 12, and not actual cumulative revenue.
3.5 GW TPU Protocol: The latest addition to Anthropic's computing power footprint.
According to Broadcom's SEC filings, the core of the agreement is that Broadcom will design and supply Google with next-generation TPU chips, with the supply relationship continuing until 2031; and Anthropic will obtain approximately 3.5 gigawatts of next-generation TPU computing power from Broadcom starting in 2027 as part of its "multi-gigawatt" computing power expansion plan.
Broadcom added a key qualifier to the document: "Anthropic's consumption of extended computing power is contingent on its continued commercial success." The three parties are also discussing deployment support with "operations and financial partners."
This is not the first time Anthropic has signed a large-scale computing power agreement. In October 2025, Anthropic signed a cooperation agreement with Google Cloud, gaining access to up to 1 million TPUs, which is expected to bring more than 1 gigawatt of computing power in 2026. Broadcom CEO Hock Tan confirmed in the December 2025 earnings call that Anthropic had placed two TPU orders worth $10 billion and $11 billion respectively. Tan further stated in the March earnings call this year that he expects to receive approximately $21 billion in AI revenue from Anthropic in 2026 and more than $42 billion in 2027 (estimated by Mizuho analysts).
On the AWS side, Project Rainier went live in October 2025, deploying nearly 500,000 Trainium2 chips across multiple data centers in the United States. Amazon has invested a total of $8 billion in Anthropic, with Anthropic engineers directly involved in the development of the Trainium underlying kernel and providing design input for the next-generation Trainium3 chip.
Anthropic's computing power now comes from three chip sources (AWS Trainium, Google TPU, and NVIDIA GPU) and three cloud platforms (AWS, Google Cloud, and Microsoft Azure).In his blog post, Anthropic emphasized that Claude is "the only cutting-edge AI model available on all three major cloud platforms in the world."
A divergence from OpenAI Stargate's approach
Anthropic's computing power model stands in stark contrast to OpenAI's.
OpenAI has chosen a capital-intensive approach: in January 2025, it jointly established Stargate LLC with SoftBank and Oracle, aiming to invest $500 billion over four years to build 10 gigawatts of AI infrastructure. OpenAI retains operational responsibility and design control, Oracle is responsible for construction, and SoftBank assumes financial responsibility. To date, Stargate plans to have nearly 7 gigawatts of computing power, with cumulative investment commitments exceeding $400 billion.
However, Stargate's progress has been marred by control disputes among its partners. According to a Tom's Hardware report in February, OpenAI, Oracle, and SoftBank disagreed on data center ownership, causing delays to some projects. Furthermore, OpenAI's total cloud service procurement commitments have exceeded $500 billion ($250 billion from Microsoft, approximately $300 billion from Oracle, and approximately $50 billion from AWS), with projected cash burn of approximately $17 billion by 2026 and positive cash flow not expected until 2030 at the earliest.
Anthropic adopts a "light asset" strategy, meaning it doesn't build data centers or buy chips. Capital expenditures are borne by cloud service providers, while Anthropic, as a customer, locks in production capacity and prices through long-term agreements, retaining the flexibility to switch between different chip architectures. The trade-off is not owning infrastructure, potentially leading to higher long-term unit costs. Reportedly, Anthropic has a gross margin of approximately 40% and is projected to incur a loss of approximately $14 billion in 2026.
The merits of the two models remain to be determined. OpenAI is betting on economies of scale and infrastructure autonomy, while Anthropic is betting on supply chain flexibility and capital efficiency. However, one fact is clear: in the AI computing power arms race, long-term computing agreements are becoming a competitive factor as important as funding and technology.
