Anthropic hits $380B valuation on $14B run-rate revenue, Claude Code at $2.5B

Meta commits $50B to Louisiana data center for 5GW+ capacity; economists warn of AI job displacement risks

Meta announced an additional $40 billion investment in its Louisiana data center campus, pushing total expected investment beyond $250 billion for at least 5 gigawatts of computing power. The expansion represents a 5x increase in announced investment at a single campus and directly contradicts recent market concerns about AI capex slowdown; Morgan Stanley simultaneously raised 2027-2028 hyperscaler capex forecasts to $1.2-1.4 trillion, signaling continued infrastructure acceleration. In parallel, nearly 200 economists including 15 Nobel laureates and Anthropic's Jack Clark signed a "We Must Act Now" letter warning that AI could drive economic transformation larger than the Industrial Revolution over a vastly shorter timeframe, with risks including large-scale job displacement. The statement calls for policymakers and tech leaders to act immediately on understanding transformative AI economics—a significant acknowledgment from the mainstream economics establishment that AI poses material labor market risks requiring urgent policy response.

Codex hits 7M weekly active users with >10x growth in 6 months, outpacing Claude Code's last reported 2M users

OpenAI disclosed that Codex reached 7 million weekly active users on July 13, up from 2 million in March and approximately 550-700k on January 1, representing explosive adoption following the GPT-5.6 Sol launch. The growth trajectory now substantially outpaces Claude Code's last reported 2 million users and $2.5 billion ARR from February, signaling a potential shift in coding agent market leadership.

GPT-5.6 Sol emerges as practical frontier model for product work; custom AI harnesses outperform general-purpose agents

Claire's benchmark testing across five product categories—PRDs, prototypes, wireframes, debugging, and agentic voice—shows GPT-5.6 Sol winning on practical usability and functional output despite Claude Fable's theoretical intelligence advantage. Sol produced more opinionated designs and easier-to-collaborate writing, with a 70/30 weighted index (70% human taste, 30% machine judgment) strongly favoring Sol. Per-task winners included Sol for prototypes and Fable for specific engineering tasks, signaling a shift in frontier model differentiation from raw capability to practical usability for shipped product work. This pattern extends to custom harness design: Claire's Sentry bug-triage harness demonstrates that purpose-built code around agents—opinionated tool adapters, structured artifacts, permission flags—produces more consistent output than open prompts to Claude Code. The harness architecture (frontend TUI, Claude Agent SDK core, tool adapters, artifact store) enables specialized agents to outperform general-purpose orchestration, suggesting the future involves custom harnesses routing to specialized models rather than monolithic agents attempting every job.

Satya Nadella coins "Reverse Information Paradox" to describe how AI usage transfers enterprise competitive advantage to model providers

Microsoft CEO Satya Nadella articulated a critical economic concern: when organizations feed proprietary data, workflows, corrections, and feedback to AI models to improve performance, they inadvertently train providers on their institutional know-how and decision-making patterns, creating information asymmetry that concentrates value with foundational model owners rather than knowledge creators. This flips Kenneth Arrow's classical information paradox, where the risk now falls on buyers rather than sellers. Nadella outlines five practical principles for enterprises to maintain "AI sovereignty": Control (own memory, traces, feedback, and fine-tuning rights), Capability (build proprietary learning environments within secure boundaries), Choice (decouple orchestration from single providers), Cost (optimize across models efficiently), and Compound (create self-improving learning loops). The framing has resonated widely across tech leadership, emphasizing a broader shift from renting intelligence toward owning the processes that generate and refine it.

BrowserOS launches as open-source agentic browser with local AI-first design, competing with OpenClaw on automation

Y Combinator-backed BrowserOS is a Chromium fork enabling users to run AI agents locally for web automation tasks—research, price monitoring, form filling—without sending data to cloud services. The browser includes 53 built-in automation tools (clicks, forms, navigation, screenshots, tabs, bookmarks, history), supports scheduled background tasks, and integrates with local models via Ollama or LM Studio. BrowserOS emphasizes zero technical setup (download and open) with 40+ built-in app integrations (Gmail, Slack, Notion, GitHub, Linear, Jira, Figma, Salesforce, Stripe) connecting via OAuth or API key, contrasting with OpenClaw's terminal-based Node.js installation and skills-based plugin architecture. The comparison signals diverging approaches to agentic infrastructure: BrowserOS optimizes for browser-native automation and ease of use, while OpenClaw prioritizes always-on daemon operation, mobile companion apps, and agent-to-agent communication.

Anthropic raises $30 billion in Series G funding at $380 billion post-money valuation

Anthropic announced a $30 billion Series G funding round at $380 billion post-money valuation, led by GIC and Coatue. Main takeaways:

• Run-rate revenue reached $14 billion, growing 10x annually for three consecutive years; customers spending $100k+ annually grew 7x year-over-year
• Claude Code run-rate revenue exceeded $2.5 billion (more than doubled since start of 2026); weekly active users doubled since January 1; 4% of all GitHub public commits now authored by Claude Code
• Enterprise customers spending $1M+ annually grew from 12 to 500 in two years; eight of Fortune 10 are Claude customers
• Claude available across all three major cloud platforms (AWS Bedrock, Google Vertex AI, Microsoft Azure Foundry) and trained on diversified hardware (AWS Trainium, Google TPUs, NVIDIA GPUs)
• Opus 4.6 model launched as leading performer on GDPval-AA benchmark for economically valuable knowledge work tasks

6 months to live for open models

Interconnects published an essay on the policy trajectory threatening open-source AI models in the US. Main takeaways:

• White House discussions underway on executive order to restrict open models, likely targeting Chinese-origin models and government uses first, with potential expansion to frontier-capability open models within 6 months.
• Anthropic's distillation warnings constitute regulatory capture: the company is lobbying for bans on Chinese open models while simultaneously hosting queryable APIs that have been repeatedly jailbroken, undermining its security argument.
• A US-only ban on frontier open models would be ineffective without global coordination and would isolate the US from the open-source community; the only viable off-ramp is for US companies (Meta, Microsoft) to release comparably capable open models immediately to shift policy focus.