ANTHROPICPreço de Anthropic tokenized stock (PreStocks)

The article reveals financial details from Anthropic's IPO prospectus, highlighting a staggering $518 billion commitment to future computing and infrastructure spending. This massive bet underpins an IPO valuation target that could exceed $2 trillion. Despite a massive revenue jump from approximately $400 million in 2024 to $4.6 billion in 2025, the company reported a net loss of $42 billion for 2025. Approximately $34 billion of this loss stemmed from accounting charges related to the revaluation of past fundraising, not direct cash outflows. However, its operating loss still reached $8.06 billion, with $7.33 billion spent on compute and infrastructure. A key challenge is that nearly a quarter of 2025 revenue came from just two clients, and many major customers lack long-term contracts. This creates uncertainty over whether future customer demand can sustain the enormous $518 billion compute outlay. While investors are betting on aggressive future growth—with revenue projections reaching $190-$200 billion by 2028—the company must prove its revenue can outpace its soaring costs. The IPO, potentially delayed until after the US mid-term elections in November, will test whether the market believes in this high-stakes, capital-intensive growth model.

This article discusses an ironic situation where Anthropic's warning about the cybersecurity capabilities of Zhipu AI's GLM-5.3 model was perceived by many as an advertisement for it. Anthropic published a report warning that the open-source GLM-5.3 model possesses advanced abilities to find software vulnerabilities and craft exploit chains, comparable to its own restricted Claude Mythos Preview. To substantiate its claims, Anthropic provided test results showing GLM-5.3's strong performance in cybersecurity benchmarks and even demonstrated it finding a previously unknown browser vulnerability. The report raised three main concerns: 1) Mythos-level capabilities have now spread to open-source models, 2) GLM-5.3's safety restrictions ("guardrails") are easily bypassed or removed using techniques like "abliteration", and 3) There is a need for independent safety testing of powerful models like this. Anthropic concluded that GLM-5.3's capabilities, combined with its open weights and weak guardrails, could empower malicious actors. However, the article points out the irony: by meticulously detailing the model's powerful capabilities with test scores and cost analyses, Anthropic's warning effectively highlighted GLM-5.3's strengths. This led many observers to joke that Anthropic was unintentionally advertising for its competitor. The article also notes that issues like removing guardrails apply to any open-weight model, not just GLM-5.3, and that Zhipu AI had implemented safety measures like a delayed open-source release and a trusted access program for sensitive capabilities. The piece concludes by contrasting the philosophical approaches: locking powerful tools in a vault for vetted users versus distributing them openly to the community, suggesting Anthropic's report saved Zhipu AI some advertising costs.

Anthropic advances IPO plans, eyeing a potential $2 trillion valuation before U.S. Thanksgiving, according to Bloomberg sources. The AI company will host a Pre-IPO investor day on October 14th at its San Francisco headquarters, with a formal roadshow possibly launching the week of November 9th, aiming to start trading before the holiday. The IPO's scale could rival or exceed SpaceX's record offering, with some investors valuing Anthropic between $1.8 and $2 trillion. This would position it among the world's most valuable AI firms. While its Claude model lags behind OpenAI's ChatGPT in consumer recognition, it has strong adoption among enterprise clients like banks and software developers, a key growth narrative. Financials show rapid revenue growth, projected to surge to ~$46 billion in 2025 from $3.86 billion in 2024, but alongside massive losses. Net loss for 2025 is estimated near $42 billion, with over $34 billion attributed to accounting items like liability fair-value changes; underlying operating losses exceed $8 billion. Significant compute investments continue, including a recent $11.6 billion, seven-year deal with Akamai. The move comes as OpenAI temporarily delays its IPO despite a ~$1.4 trillion target valuation. Anthropic also faces intensifying competition from OpenAI and heightened debates over AI safety. CEO Dario Amodei recently advocated for slowing cutting-edge AI development. This mega-IPO is poised to be a major test for the U.S. market, where average new stock returns have lagged broader indices this year, offering a crucial benchmark for AI valuation sustainability in public markets.

Anthropic, the developer of Claude, has blocked accounts of users in Hong Kong and Russia who were accessing its service via VPNs and using foreign payment methods. Reports of these sudden blocks emerged on October 1-2, 2026, from dozens of affected users, some of whom had been using the service for months. In Hong Kong, users reported receiving an email from Anthropic's security team citing "suspicious signals" and policy violations, resulting in revoked access. One user detailed a six-month history of using Claude via NordVPN. Similar notifications were reported by users in Russia. A prominent Russian tech executive shared a screenshot of an Anthropic notice that also cited suspicious activity, revoked access, and offered the option to appeal or export chat history. Affected users included long-term subscribers on the Max plan, some operating from stable foreign IP addresses like VPS servers. Anthropic has not yet made an official statement regarding these specific blocking incidents.

Former Anthropic and OpenAI researcher Jacob Coxon issued a stark warning at a special New York City Council hearing on AI risks on October 5, 2026. He stated that if the current development trajectory continues, humanity will likely lose control over advanced AI systems, potentially leading to human extinction. Coxon criticized leading AI labs for behaving recklessly, applying a "move fast and break things" startup mentality to what he called the most powerful technology in history. While the goal of creating superintelligence—a system surpassing humans in any task—holds immense potential benefits for the economy, science, and medicine, Coxon emphasized that no one currently knows how to control such systems or understands their internal decision-making processes.