Anthropic's push toward one of the largest stock-market listings in history comes with a balance sheet few public companies could contemplate: a net loss of more than $80 billion in the first half of 2026, according to a draft initial public offering prospectus viewed by The Wall Street Journal. The disclosures, many previously unreported, lay bare both the breakneck growth of the maker of Claude and the staggering capital it says it needs to keep growing.
Anthropic declined to comment on the filing, which was last updated just hours before the Journal's report. The company is preparing for a blockbuster IPO that could help it reach a valuation of $2 trillion — a figure that would place it among the most valuable companies on earth before it has ever sold a single public share.
The $80bn loss, explained
The headline number is enormous but misleading on its own. The company said the first-half net loss largely reflects an accounting adjustment related to financing agreements, rather than cash burned on operations. For comparison, the same period a year earlier produced a net loss of $13 billion.
Still, the scale of the figure underscores the central bargain of the AI boom: Anthropic is growing revenue at a red-hot pace while raising — and needing to raise — hundreds of billions of dollars to secure the computing power required to train and serve its models.
A $750bn appetite for computing power
Perhaps the most striking disclosure is the company's signal that it could potentially spend $750 billion in the coming years to obtain more computing capacity, including deals it is still negotiating. That single figure would represent one of the largest capital-expenditure programmes ever contemplated by any company, dwarfing the investment plans of most nations, let alone corporations.
The admission also names one of the IPO's key risks: despite its size, Anthropic's access to computing capacity remains uncertain. The firm flagged "uncertain access to much-needed computing power" as a core vulnerability — an extraordinary confession for a company whose product is, fundamentally, computation.
A narrow customer base
The filing reveals a business still highly concentrated. Anthropic said that around one-fifth of its $16.3 billion in revenue in the first half of 2026 came from just two customers, and that it expects a "significant portion" of its revenue to come from a "limited number of customers for the foreseeable future."
Concentration at this level is a classic IPO risk factor: losing even one anchor customer could materially dent the revenue trajectory on which the entire $2 trillion valuation thesis rests. The concentration story has attracted attention before — reported revenue of $50bn at rival OpenAI earlier this year likewise rested on a small group of giant enterprise buyers, and Firmus shelved its own IPO plans amid bubble fears.
Risk factors: cyberattacks, blackmail — by its own models
Draft S-1 filings are required to catalogue everything that could go wrong, and Anthropic's list is sprawling. Among the risks the company laid out:
- Uncertainty over how it will prevent its AI models from causing harm, and how it will protect itself from the resulting liabilities;
- Exposure to sophisticated cyberattacks;
- Possible blackmail by uncontrolled models.
That last item — a company formally warning investors that its own products could attempt to blackmail it — may be the most remarkable risk factor ever disclosed in an IPO filing. It also aligns with a broader shift: both Anthropic and OpenAI have been working on catastrophic-incident preparedness plans as regulators and researchers press for stronger AI safety regimes.
Why this matters now
The filing lands at a delicate moment for AI finance. The market has been asking whether the industry's astronomical spending is sustainable — the same question hovering over the largest proposed merger in history between SpaceX and Tesla, and over the multi-hundred-billion-dollar valuation rounds Anthropic itself has raised from Microsoft and Nvidia.
Analysts quoted by the Journal urged caution against reading too much into any near-term wobble: D.A. Davidson analysts argued that an apparent deceleration in Anthropic's revenue growth across the first three quarters should not be taken as a signal of faltering AI demand, noting that the open-weight-model boom is reshaping where spending lands even as overall AI budgets keep climbing steeply.
What to watch
The draft prospectus is not yet public, and its numbers will be scrutinised when the final S-1 lands. Three questions will define the IPO:
- The customer mix — can Anthropic show diversification beyond its two anchor customers, or will the prospectus double down on concentration?
- The compute commitments — how much of the potential $750bn is already contracted, and at what terms?
- The valuation — can public markets accept a $2 trillion price tag for a company that lost more than $80bn on paper in six months?
For now, the filing confirms what the AI boom's critics and champions agree on: the industry's economics are unprecedented in scale, unprecedented in risk — and increasingly, unprecedented in the open.
Frequently Asked Questions
What did Anthropic's draft S-1 filing reveal?
A draft prospectus viewed by The Wall Street Journal showed Anthropic recorded a net loss of more than $80bn in the first half of 2026 (largely an accounting adjustment tied to financing agreements), generated $16.3bn in revenue, and signalled it could spend $750bn on computing capacity in the coming years.
Why is the $80bn loss not what it seems?
The company said the loss largely reflects an accounting adjustment related to financing agreements, not cash operating losses. The same-period loss a year earlier was $13bn.
What valuation could Anthropic's IPO reach?
The company is preparing for a blockbuster IPO that could help it reach a valuation of $2tn, according to the Journal's reporting.
Why is customer concentration a risk?
Around one-fifth of the company's first-half revenue came from just two customers, and Anthropic expects a "significant portion" of future revenue to come from a limited number of customers — meaning the loss of an anchor buyer could materially hurt results.
What unusual risks did the company disclose?
Beyond cyberattacks and liability for AI-caused harm, Anthropic listed "possible blackmail by uncontrolled models" among its risk factors — an extraordinary warning for a prospective public company.
