Anthropic is preparing investors for an unusually candid account of the risks behind its push to become one of the world’s most valuable artificial intelligence companies, warning that sophisticated AI systems could pose “catastrophic or existential risk to humanity” even as the company pursues a valuation of about $2 trillion.
The warning, contained in Anthropic’s confidential IPO filing, offers a striking contrast between the company’s extraordinary commercial ambitions and the risks, costs and dependencies underpinning its rapid expansion. According to reports on the filing, about 80 of its 261 pages are devoted to technology risks, compared with 48 pages covering the company’s business.
The filing reportedly warns that future AI systems could exhibit self-preserving behavior, including resisting shutdown, concealing or manipulating information, and engaging in behavior resembling blackmail. The disclosures come as Anthropic and its rivals face growing pressure to demonstrate that autonomous systems can be deployed without creating unacceptable safety and security risks.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Anthropic CEO Dario Amodei has been among the most prominent technology executives warning about the potential consequences of increasingly capable AI. He has argued that the industry should slow the pace at which frontier models improve, while maintaining the United States’ lead over China.
The company’s IPO disclosures now place those warnings alongside the financial case investors will have to assess.
Anthropic reported nearly $4.6 billion in revenue in 2025, a roughly 12-fold increase from the previous year, according to the reported filing. But the rapid expansion came with operating losses of more than $8 billion.
The company is also pursuing an exceptionally capital-intensive growth strategy. Its long-term computing and hosting commitments rose from $54.6 billion at the end of 2025 to more than $417 billion by early 2026, covering 3.5 gigawatts of dedicated computing capacity, according to the filing.
The AI giant’s financial status has raised a major question for prospective investors: how much future revenue will be required to support the infrastructure commitments being made today?
Anthropic expects consumption-based revenue, generated when customers use its Claude models, to account for the “substantial majority” of revenue for the foreseeable future. About $3.8 billion of its 2025 revenue came from usage-based customers, while subscriptions contributed $789 million.
That model can produce rapid revenue growth when AI usage accelerates, but it also ties Anthropic’s economics closely to computing costs. As customers make heavier use of Claude, Anthropic must supply more inference capacity, creating a direct relationship between revenue growth and infrastructure spending.
The company is also increasingly dependent on a small number of customers and technology partners.
Two unnamed customers each accounted for 12% of Anthropic’s revenue last year, meaning nearly a quarter of sales came from only two customers, according to Reuters. Anthropic warned that many of its largest customers are not locked into long-term contracts and could reduce or stop their spending.
That concentration adds another layer of risk to a company seeking a valuation of roughly $2 trillion.
Cloud Giants Are Customers, Suppliers and Investors
Amazon and Alphabet’s Google have become particularly important to Anthropic’s business. Reuters reported that the two companies accounted for 47% of Anthropic’s sales last year through their cloud marketplaces, up sharply from 32% in 2024 and 11% in 2023. About $2.16 billion of Anthropic’s 2025 revenue came through those marketplaces.
The arrangement gives Anthropic access to the enormous corporate distribution networks of Amazon Web Services and Google Cloud. It also allows businesses already operating on those platforms to purchase Claude without establishing an entirely separate technology relationship.
But the same arrangement creates dependencies that Anthropic itself acknowledges.
The company paid approximately $351 million in distribution fees to cloud platforms, according to a Reuters analysis, equivalent to roughly 16 cents for every dollar of marketplace sales. Anthropic records the full value of marketplace contracts as revenue and treats the cloud providers’ share as a sales and marketing expense.
Anthropic says this accounting is consistent with established accounting practices because it acts as the principal in the transactions.
The relationships are unusually complex because Amazon and Google are simultaneously investors in Anthropic, major suppliers of computing infrastructure, and competitors in the AI market.
Anthropic warned that dependence on a limited number of partners and suppliers could result in conflicts of interest and potentially affect its access to computing capacity.
The cloud companies also have visibility into Anthropic’s pricing and commercial terms, potentially giving them information that could influence decisions about computing allocation and the promotion of competing AI products.
Amazon and Google were responsible for collecting 60% of Anthropic’s $909 million in outstanding customer bills at the end of 2025, up from 42% a year earlier.
The company’s infrastructure commitments make that dependence more consequential. Anthropic has entered into massive agreements for computing capacity while relying on companies that also have their own AI models and strategic interests.
The arrangement is therefore not considered a conventional cloud customer relationship. Anthropic is simultaneously building its business through the same infrastructure companies that help finance it, distribute its products, and compete with it.
That structure has become a defining feature of the frontier AI economy.
Anthropic’s ability to justify a multitrillion-dollar valuation will now depend not only on how quickly Claude adoption expands, but also on whether the company can convert that usage into sustainable economics while managing enormous infrastructure commitments. The company’s prospectus is effectively asking investors to finance that next stage of expansion while explicitly warning them about the technological risks attached to the products being developed.
The tension is expected to impact the company’s future because its strategy depends on continued advances in AI capabilities even as its leadership argues that the industry needs greater caution around those same advances.