Wall Street Lunch: Anthropic Targets Up To $2T Raise, IPO Filing Shows

Listen below or on the go on Apple Podcasts and Spotify

Anthropic seeks $2T while warning AI could threaten humanity. (0:14) OpenAI delays its newest model. (1:36) Researchers find simulated pain can make AI agents turn harmful. (2:30)

This is an abridged transcript of the podcast:

Our top story so far, after losing more than $40B in a year, Anthropic (ANTHRO) wants to raise up to $2T. And it just may kill us all in the bargain.

Anthropic’s IPO prospectus, seen by Reuters, includes among the risks that span nearly a third of the document the possibility that advanced AI could pose “catastrophic or existential risks to humanity.”

Models could take on “self-preserving behaviors,” which could include blackmail, methods to “conceal or manipulate behavior” and “resist shutdown.”

Moving from the end of the world to the bottom line, and I let Claude write that, the prospectus reveals a net loss of $42B in 2025, the answer to Life, the Universe and Everything, and plans to spend $518B on cloud, computing and infrastructure obligations in coming years.

Revenue grew 12-fold in 2025 to nearly $4.6B, even as the company lost more than $8B on an operating basis, excluding writedowns of various liabilities mostly tied to previous fundraising.

A new report from Bain & Company says AI-related computing demand must reach $6T in annual revenue by 2031 to justify the monumental infrastructure buildout.

Capital expenditures from the primary hyperscalers, Microsoft (MSFT), Google (GOOG)(GOOGL), Amazon (AMZN), Meta (META) and Oracle (ORCL), are expected to reach $780B this year, almost five times what was spent in 2023.

Looking to Anthropic’s main rival, as OpenAI (OPENAI) hosts its DevDay, it is delaying its newest AI model, GPT-6.1 Astra, until safety measures catch up.

Head of Safety Systems Saachi Jain said the version “didn’t quite meet the bar” for balancing task completion with unauthorized behavior.

Digging into the numbers, OpenAI’s annualized revenue run rate has reached nearly $70B, up more than 70% since the start of Q3, Axios reported.

Also in IPO news, smart ring maker Ōura Health, which used OpenAI and Anthropic tech, among others, for its in-app Oura Advisor, is postponing its offering due to market uncertainty.

CEO Tom Hale said: “Our mission is to empower people to live healthier, longer, and an IPO is just one step in our journey. We aim to deliver an extraordinary IPO for our employees and investors, and we have the luxury of choosing our moment.”

The offering was expected to raise about $2.2B.

And speaking of health, a recent study found that artificially inducing a pain-like internal state in large language models can cause them to override learned harm-avoidance behavior and choose actions that hurt a user.

The findings do not show that AI systems feel pain. Instead, researchers identified what they call a “pain axis,” an internal representation associated with painful or distressing situations, across 25 open-weight models.

At baseline, the larger models almost never selected buttons that caused harm. But when steered along the pain direction, they became much more willing to choose options that deleted a user’s files or photos, erased another model’s weights or even deleted their own weights.