IPO imminent: Anthropic splashes out $517 billion on computing power, yet still trails OpenAI

Bitsfull2026/09/07 14:407831

概要:

Despite Anthropic's annualized revenue surpassing $65 billion and exceeding OpenAI, its computing power reserves still lag behind OpenAI's planned target of 30 gigawatts.


Over the past 11 months, Anthropic has aggressively signed computing power agreements totaling a staggering $517 billion. The direct trigger for this procurement frenzy is the demand shock driven by the explosive growth of its two products, Claude Code and Cowork, this year.


On September 6, according to a comprehensive analysis by The Information of public statements and previous reports, since October last year, Anthropic has signed agreements to lock in at least 14.8 gigawatts of computing capacity, available for use over the coming years. Combined with the 1 to 2 gigawatts previously secured, the total scale has further expanded.


This figure far exceeds the plan Anthropic disclosed to investors in December 2025—at that time, the company stated it would spend $180 billion on server rentals by 2029. However, based on currently announced agreements, total spending over the next decade or so could reach as high as $517 billion, covering cloud capacity leases, chip purchases, and data center leases.


Notably, Anthropic may have more undisclosed agreements, and the capacity under signed deals may not be fully utilized.


Additionally, despite Anthropic's annualized revenue surpassing $65 billion, exceeding OpenAI's $40 billion, its computing power reserves still lag behind OpenAI's planned target of 30 gigawatts.


Core Suppliers: Amazon and Google Shoulder the Bulk


Anthropic's early partners and investors, Amazon and Google, remain its largest computing power suppliers, collectively providing 11 gigawatts of capacity, with costs exceeding $300 billion over the next decade or so.


Over the past year, Anthropic has also extended its reach to partners previously more closely tied to OpenAI:


Microsoft: In November last year, Anthropic announced a deal with this long-time OpenAI backer, committing to spend at least $30 billion to lease approximately 1 gigawatt of Microsoft Azure servers (equipped with Nvidia chips).


SpaceX: In May this year, Anthropic announced a deal with Musk's SpaceX to pay $1.25 billion per month to lease AI servers, with the agreement running through May 2029 and a total potential cost of up to $45 billion.


Lambda and Nscale: Just last month, Anthropic signed cloud contracts worth $80 billion with startup Lambda and UK-based Nscale, locking in at least 460 megawatts of capacity for a six-year term.


Building Its Own Data Centers: Betting on Long-Term Control


Beyond leasing, Anthropic is also accelerating its push into proprietary data centers.


In November last year, Anthropic announced plans to jointly invest $50 billion with cloud startup Fluidstack to build data centers in Texas and New York, with some facilities already coming online this year.


Additionally, Anthropic signed a lease with former Bitcoin miner TeraWulf, which is developing an AI data center in Kentucky that will ultimately deliver 401 megawatts of computing power.


At the chip level, Anthropic signed an agreement with Google in April this year to procure multi-gigawatt-scale tensor processing units (designed by Broadcom); in July, it added another 2 gigawatts of AMD chip orders.


The Gap with OpenAI: Leading in Revenue, Lagging in Compute


Although Anthropic has overtaken OpenAI in revenue—according to Bloomberg, its annualized revenue has exceeded $65 billion, surpassing OpenAI's revenue of over $40 billion as of July this year—the gap in compute scale remains significant.


OpenAI told investors in April this year that it plans to have 30 gigawatts of computing power by 2030. According to sources familiar with the matter, OpenAI estimated last month that its compute spending before 2030 would be around $750 billion, up from its previous forecast of $665 billion.


OpenAI has signed a flurry of deals with Oracle, Microsoft, CoreWeave, and others. OpenAI executives say these commitments have already built the company a competitive edge over rivals.


By comparison, Anthropic's publicly disclosed total compute plans still appear to fall well short of OpenAI's target level.


Key Variables Ahead of the IPO


Whether computing power can keep pace with growth directly impacts Anthropic's profit margins and customer retention.


Over the past year, Anthropic and its Chief Financial Officer Krishna Rao have prioritized compute agreements. With the IPO prospectus expected to be made public within the coming weeks, what investors are most eager to see is the company's latest disclosure on compute capacity and spending.


However, these agreements are not set in stone. Data center construction faces multiple hurdles, including site selection, grid connectivity, and government approvals, while also encountering resistance from local governments and component shortages.


Uncertainty also exists at the contract level. Take the SpaceX-Anthropic agreement, for example—either party can cancel the contract with 90 days' prior notice.


Original Link


Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia