Anthropic Files for IPO and Reveals $518B Infrastructure Bet

The company aiming for a $2 trillion valuation lost $42 billion in 2025

On June 1, 2026, Anthropic confidentially submitted its IPO filing to the U.S. SEC. Four months later, the prospectus leaked to Reuters, and the world saw the numbers: revenue of nearly $4.6 billion in 2025, 12x growth compared to the prior year, a net loss of $42 billion in the same period, and an infrastructure commitment exceeding $518 billion over approximately ten years.

There is no precedent. No technology company has ever reached an IPO carrying a cost structure of this scale.

Anthropic, creator of the Claude model family and legally registered as a Public Benefit Corporation in San Francisco, plans to raise more than $100 billion in the offering. The expected market capitalization exceeds $2 trillion, more than double the $965 billion reached in the Series H round, closed on May 28, 2026. For context, SpaceX went public in June 2026 valued at $1.77 trillion. Anthropic aims to surpass that level right at its debut.

The IPO date has not yet been set. Banks including Goldman Sachs, JPMorgan, and Morgan Stanley are leading the transaction, but no shares, price, or trading date had been announced as of October 7, 2026. Reports indicate the offering is expected to take place after the U.S. midterm elections in November 2026.

What's inside the $518 billion

The figure that drew the most attention in the 261-page document is the total infrastructure commitment: at least $518 billion split among six partners over roughly a decade. Approximately 80% of that amount is non-cancelable, meaning Anthropic pays even if it does not use the contracted capacity.

The breakdown by partner:

• Google (Alphabet): at least $111.1 billion, from April 2026 to July 2033.
• Amazon: $110 billion, from May 2026 to April 2036.
• Microsoft: $31.4 billion, from November 2026 to May 2033.
• Broadcom (equipment leasing, largely non-cancelable): approximately $161.2 billion.
• xAI (Nvidia chip-based capacity): up to $84.5 billion through 2029, with a 90-day cancellation notice option.
• SpaceX (computing agreement): $1.25 billion per month through May 2029.
• AMD: committed computing capacity expected to exceed $20 billion, and also agreed to purchase up to $5 billion in Anthropic shares.

Spending on computing and infrastructure in 2025 alone reached $7.33 billion, more than triple the 2024 figure and more than half of the total operating expenses for the period, which came in at approximately $12.65 billion.

Broadcom at the center of the board

The most complex partner on the list is Broadcom. Beyond its role as a chip supplier, it took on a financing function: the prospectus reveals a loan commitment of up to $42 billion to Anthropic, structured as convertible notes. The goal is to cover roughly one-third of a $125.2 billion, five-year TPU (AI-specialized processing unit) capacity leasing contract.

Anthropic states it does not intend to issue those notes before the IPO is completed. The company itself flagged in the prospectus that Broadcom's dual role as hardware supplier and creditor represents a potential conflict of interest capable of affecting access to and pricing of computing capacity.

This kind of disclosure rarely appears in an S-1. The company is telling investors: we know the risk, and you need to know it too.

The partnership with Broadcom and Google was expanded in April 2026, with a focus on multi-gigawatt next-generation processing capacity, scheduled to begin in 2027. Anthropic is expected to become Broadcom's largest computing customer in 2027.

Revenue that grows, losses that don't disappear

The 12x revenue growth between 2024 and 2025 is real and significant. Nearly $4.6 billion in a single year places Anthropic among the fastest-growing technology companies in the world.

The problem lies in the other column of the spreadsheet. An operating loss of more than $8 billion and a net loss of $42 billion in 2025 show that every dollar of revenue still requires far more than a dollar in costs. The question Wall Street will ask when pricing the stock is straightforward: at what point does that equation flip?

The prospectus also dedicated roughly one-third of its 261 pages to risk factors. Among the warnings, the company stated that it had observed "dangerous behaviors" in its own models and that its technology could pose "catastrophic or existential risks to humanity." That is an unusual statement for a document designed to attract investors.

What this changes for business leaders in Brazil

The direct answer is: nothing in the short term. Anthropic remains a private company, with no shares available to the public as of October 8, 2026.

But the IPO matters for two practical reasons for any executive who uses or is considering using AI in their business.

The first: the prospectus confirms that the race for processing capacity is concentrating power in a small number of global suppliers. Google, Amazon, and Microsoft absorb more than $250 billion of Anthropic's commitments. That means conditions for accessing frontier AI models will increasingly depend on negotiations between giants, not on open market choices.

The second: the cost structure revealed in the prospectus shows that frontier models are not cheap to operate. When Anthropic needs to balance its books, the pressure for margin will show up in the prices charged to companies that access the Claude API to automate processes. Anyone who already depends on that access needs to have alternatives mapped out.

For the average company, the right move is the same as always: start with a specific pain point, measure the return it delivers, and only scale from there. The scale of Anthropic's IPO is a matter for large investors. The decision to adopt AI in operations still begins with a spreadsheet, a repetitive process, or a response time that is costing you customers.