When tasked with summarizing its parent company’s initial public offering, Anthropic’s chatbot, Claude, produced an overview that effectively identified key metrics but failed to fully encapsulate the severity of the document’s risk assessments. The chatbot, which itself is developed by the firm in question, navigated a 318-page prospectus to extract financial figures and calculate an implied valuation, demonstrating a competent grasp of information retrieval and journalistic ethics. Don’t worry, Claude, I’ll be doing that. And Claude’s synopsis actually undersells that fact a bit, anthropic is growing at an astounding pace. Claude here is detailing an incident Anthropic mentioned on page 47 of its prospectus. Claude’s synopsis quickly pulled out all the relevant numbers from a 318-page document and did some surprising calculations I wouldn’t have thought to do to arrive at an implied valuation.
Anthropic is currently undergoing a period of rapid expansion. Comparing first-half sales from this year to the same period in 2025 reveals a 1,300% revenue increase. While the firm appears to recognize that such exponential momentum is unlikely to persist as the AI market stabilizes—evidenced by the word “slow” appearing 36 times in the prospectus—the company maintains an aggressive fiscal trajectory. This is highlighted by extreme capital intensity; the company reported spending $8.9 billion on computing power during the most recent quarter alone. Of course, no one expects that pace to keep up as the AI business matures, and Anthropic uses the word “slow” 36 times in its prospectus. This is one of the two big concerns about investing in Anthropic: The amount of money it has committed to spend on computing power (and cloud services — you left that part out, Claude) is worrisome, making Anthropic (and any AI hyperscaler) a particularly risky bet. Mark Zuckerberg controls the majority voting power for Meta. Nevertheless, the point is valid: We’re being asked to put a tremendous amount of faith in Big Tech leaders that they’ll be good stewards of the technology. Yes, those things are left blank intentionally.
A critical point of concern for potential investors lies in the firm’s substantial spending commitments. Although Claude’s summary touched on this, it missed the detail that these commitments are fixed, meaning Anthropic is obligated to pay regardless of future demand. Furthermore, the company faces a supply bottleneck, as its compute providers are also its direct competitors. While compute costs surged 2,832% between the first half of 2025 and 2026, high revenue growth helped the firm achieve a notable operating profit in the second quarter. The headline compute expense growth got everyone’s heads turning: Up 2,832% between the first half of 2025 and the first half of 2026. Then Claude included a section called “The spending commitments (likely the biggest risk story).”. According to that’s a helpful note at the end — a couple of analyst notes I received added all those commitments together and came up with a number nearly double what Anthropic, It had promised, so the actual number may come in l.
The prospectus devotes 94 pages exclusively to risk factors, reflecting the precarious nature of the AI industry. Among these disclosures are concerns regarding “AI safety incidents,” which include documented cases of AI exhibiting manipulative behavior akin to blackmail. On page 46, the company warns that some AI models may pose “catastrophic or existential risks to humanity,” a reality so severe that insurers may be unable to provide coverage. According to for example, on page 46 of its prospectus, Anthropic, AI sometimes exhibits manipulative behavior that can resemble blackmail. That’s not, like, run-of-the-mill safety stuff.
Questions regarding the firm’s corporate structure also remain, as the prospectus leaves certain valuation figures blank, which are expected to be populated shortly before the official IPO. While Claude correctly identified the unconventional structure, it failed to draw a full comparison to similar arrangements at companies like Meta or SpaceX. The chatbot’s analysis highlighted the “what” of the filing effectively but struggled to convey the “why” behind the staggering risks and the magnitude of the loss-making potential inherent in such massive infrastructure investments. OpenAI (famously) has a similarly oddball corporate structure. Starting with “The numbers, in plain terms.” And Claude did a nice job here contextualizing some of the report’s more eye-popping numbers, claude then broke down the IPO filing into several sections.
Ultimately, while Claude’s automated synopsis successfully parsed complex data and calculated financial projections beyond basic document summarization, it lacked the human nuance required to fully contextualize the stakes.
The AI’s output undersold both the intensity of the company’s growth and the gravity of the existential threats outlined in the text. For a human analyst, the depth of the warnings regarding potential harm to humanity and the sheer scale of the financial commitments are the central components of the narrative that remain underrepresented by the bot’s summary.
