Anthropic Profitability, Compute Deals, IPO & Funding; Claude 5 Fable Released
Key Questions
What are Anthropic's latest financial and valuation figures?
Anthropic neared its first profitable quarter with $10.9B Q2 revenue and a $47B run rate. It closed a ~$65B round at $965B valuation and filed confidential IPO papers targeting an August debut.
What new model did Anthropic release and what are its capabilities?
Anthropic released Claude 5 Fable with strong CursorBench and cybersecurity benchmarks matching GPT-5.5 levels. Early reports note improved speed and writing quality alongside high token consumption in agent workflows.
How is SpaceX involved with Anthropic's compute needs?
SpaceX has become a neo cloud provider supplying compute infrastructure to Anthropic. This supports Anthropic's $50B data center investment plans amid rising compute costs.
Anthropic nears first profitable quarter ($10.9B Q2 revenue, $47B run rate); closed ~$65B round at $965B valuation, surpassing OpenAI. QIA increased investment in the round. Filed confidential IPO papers with the SEC, likely August debut. President cites high compute costs as driver for IPO. Plans to invest $50B in US data centers. Blackstone and Apollo working on $36B debt deal for compute infrastructure. SpaceX is now providing compute as a neo cloud provider. Andrej Karpathy leading pre-training. Released Opus 4.8 with dynamic workflow for up to 1,000 subagents; Claude Code also got dynamic workflows. Mythos model with advanced cyber capabilities now scaling to 150 orgs across 15+ countries (Okta, Samsung, NATO). Compute deal with xAI. Practical concern: Opus 4.8 may consume excessive tokens in agent workflows. IPO sparks compute gold rush analysis. Historical IPO patterns from Palantir, Snowflake suggest post-IPO volatility. New: Anthropic released Claude 5 Fable (Mythos-class) with strong CursorBench scores matching GPT-5.5, insane cybersecurity benchmarks, but token-hungry. Early user reports highlight speed and writing quality improvements, but long-context issues and aggressive pricing. Cybersecurity researchers are unhappy about Fable's guardrails hindering research. OpenAI counters with Ona acquisition for agent cloud infrastructure. Fable 5 achieved 255 tok/s for Gemma 4 on WebGPU, signaling local inference progress.