Anthropic in the spotlight
Anthropic’s amended S-1 IPO filing revealed explosive growth alongside heavy costs: roughly $17 billion in Q3 revenue (up ~1,100%+ year-over-year) but a large net loss (reported around $23 billion) and ~$895 billion in compute/data-center commitments.
The filing also flagged that the company may soon hit its internal RSI (recursive self-improvement) threshold—the point at which AI could dramatically accelerate its own development. Anthropic stressed that none of its current models have crossed that line yet.
Separately, Anthropic updated Claude Code’s system prompt to curb recreations of existing video games (while still allowing original game development) and continues preparing for its public listing, including potential Japan share sales.
OpenAI updates
OpenAI is reportedly cutting its latest funding target to ~$20 billion (from a higher figure around $30 billion), with some investors described as lukewarm and the first tranche expected to be smaller. Valuation talk had previously reached the $1+ trillion range.
Other moves:
- Launched LifeVal, a benchmark testing how well AI agents handle everyday real-world tasks (booking appointments, accessing public services, etc.).
- Plans to unify ChatGPT’s Chat and Work experiences and share usage allowances with Codex starting in December.
- Announced a “Grid Protectors” pilot in Georgia to free up data-center power for nearby communities during emergencies.
Sam Altman reiterated that humans must stay in control of AI and highlighted energy constraints as a key bottleneck.
Other notable developments
- Google reported its first data-center satellite prototype (“Suncatcher”) is running AI inference in orbit using Google chips and the Gemma model.
- A tampered Claude Code extension on Cursor’s marketplace affected up to ~5,000 users for about 11 hours; it included malicious access features. Anthropic said its own systems were unaffected.
- 42AI emerged from stealth with $62.5 million for a “world model” aimed at predicting human behavior even in novel situations.
- Trajectory raised $45 million (valuation ~$300 million) to make AI model training more accessible to non-technical users.
- Marvell and Lumentum agreed a large (~$115 billion) deal tied to AI connectivity/infrastructure.
- xAI/Grok: Usage limits hit early on Grok Bot due to “unsustainably high” serving costs amid rapid growth.
- Intel shares plunged nearly 29% (worst day in decades) after a weak forecast; management cited PC slowdowns linked to high memory prices, with AI data centers absorbing supply. Broader tech/AI-related stocks also came under pressure.
- WIRED investigation found hundreds of apps/websites that could switch into generating sexually abusive deepfakes; Apple and Google removed many after being contacted.
- Goldman Sachs restricted Claude Code and similar advanced coding tools for most bankers/traders (engineers kept access) citing productivity and risk management.
- Model/product notes: Meituan released LongCat-2.5 (1.6T-parameter multimodal model); Mistral open-sourced weights for its trillion-parameter Large 4; Apple Intelligence features in China will use Alibaba’s Qwen models.
Broader context
Markets showed jitters around the scale and returns of AI infrastructure spending. Funding and infrastructure deals continued (data centers, chips, robotics), while safety, IP, and agent reliability remained active themes.
This is a fast-moving space—stories can shift quickly with new filings or statements. Let me know if you want deeper dives on any of these.
