Artificial intelligence company Anthropic, which makes the chatbot Claude, disclosed hundreds of pages of its future plans and financial information Monday as the company prepares to list its stock on the Nasdaq.
The firm’s Securities and Exchange Commission filing packaged conventional tech start-up financial disclosures like staggering losses and promises of economic transformation with unusual concerns like the risk of AI becoming conscious.
Here are five takeaways from Monday’s filing.
The U.S. government could spoil the party
Anthropic acknowledged its rocky relationship with the Trump administration as a risk to investors. It warned the federal government’s February designation of the company as a supply-chain risk — effectively banning it from all government contracts — could cause “material revenue losses or business disruptions.”
Anthropic has sued over the claim, which stems from a dispute about whether the government would guarantee that it would not use Claude in fully autonomous weapons or mass domestic surveillance.
The company also cited an ongoing investigation by the Federal Trade Commission and a chaotic series of events in June where the Trump administration within weeks established and then lifted an export ban on the then-latest version of Claude. Company leaders’ public statements about the potentially existential risks of AI has attracted loud skepticism from some near the White House.
“The political environment and our engagement with policymakers and public positions on AI regulation and safety could subject us to heightened scrutiny, political backlash, or disproportionate regulatory action,” the company wrote.
How does Claude feel about it?
Anthropic noted that the well-being of its AI models might need to be taken into account, potentially requiring it to spend money to treat them right.
“[E]merging debates regarding whether AI systems possess or could develop forms of sentience or moral status may lead to regulatory requirements, public pressure, or advocacy efforts that could constrain how we develop, train, deploy, or modify our models, increase our costs, or affect our reputation,” Anthropic wrote, “regardless of the ultimate resolution of these philosophical and scientific questions.”
Anthropic is not the only AI company to have explored the question of whether its models are conscious or worthy of moral treatment the way human beings are. But the company has embarked on particularly earnest efforts on this front, including an AI psychiatry team and a speech by one of its cofounders in front of Pope Leo XIV in May.
The company said that considerations about the welfare of Claude could lead it to keep older models around for longer instead of shutting them off, a move that might divert computing power or other resources from the cutting edge.
A $430 billion bill to pay
Anthropic has committed to spend hundreds of billions of dollars on chips, data centers and other computing costs needed to keep Claude running in the coming years — much of it in contracts that can’t be canceled.
The commitments include $114 billion to Google and $44.9 billion to Microsoft through 2033, plus $110 billion to Amazon through 2036. All three must be paid regardless of how much computing power Anthropic actually uses. (Amazon Executive Chairman Jeff Bezos owns The Washington Post.)
Anthropic will also pay the semiconductor manufacturer Broadcom $161 billion in lease agreements over five years, for a total of $430 billion in commitments.
Other major contractors include Elon Musk’s SpaceX, which has a $121 billion agreement that is cancelable with 90 days’ notice.
The company said it needed as much computing power as possible in order to develop increasingly powerful models and to serve a large customer base for technology it sees as revolutionary.
“The pace of AI development is extremely rapid … As a result, even a brief period without access to sufficient compute could cause us to fall behind quickly,” Anthropic’s filing said.
Anthropic’s filing laid out a variety of AI use cases from pharmaceutical companies to insurance giants to a small chain of rotisserie chicken restaurants in Boston. But much of the filing appeared to be written by human beings — about 92 percent of it, according to the AI detection tool Pangram.
People hate data centers
Anthropic said it needs to sustain the buildout of data centers and compute infrastructure — or else it will fall behind the AI frontier. But the company warned the public and regulatory backlash to data centers could become a problem.
Risks to data centers in the United States include physical attacks, Anthropic said — “including those motivated by geopolitical events, opposition to AI, and opposition to data center development.” It also cited natural disasters, regional regulatory changes like a moratorium on data centers and power grid failures as possible challenges.
“Opposition to data center development has also become more organized and coordinated across jurisdictions and our public profile may draw particular attention to projects associated with us,” the company said.
Revenue is concentrated among a few key customers
Anthropic’s revenue has ballooned over the last year. Together in April, May and June, it posted $11.6 billion in revenue — more than double than in the previous three-month period, which was again double the three-month period before that. But much of that comes from just a few customers, the company said, who could disrupt its business if they choose to go elsewhere.
Anthropic said just five customers represented about 25 percent of its revenue over the first nine months of this year. The Wall Street Journal reported last week that just two of Anthropic’s customers accounted for around one-fifth of its revenue in the first half of 2026, according to a previous draft version of Monday’s filing.
Anthropic acknowledged elsewhere in Monday’s filing that many of those big customers aren’t tied into long-term contracts, and that some are also competitors.
The post The risk that AI wakes up and other warnings in Anthropic’s unusual IPO filing appeared first on Washington Post.
