Issue #30·Thursday, January 29, 2026·20 min read·10 stories

OpenAI Seeks $40B from Nvidia, Amazon, Microsoft

Clawd creator ships code unread; Nvidia's new CPU; Claude apps for Slack

OpenAI is reportedly in talks to raise $40 billion from Nvidia, Amazon, and Microsoft. This shows sustained investor confidence in the AI market, shaping future compute availability and platform roadmaps. Separately, one engineer detailed their agent, Clawdbot, gaining full computer access, resulting in a broken family calendar and an unplanned podcast appearance.

▲NEWS

6 stories

Meta to Nearly Double AI Infrastructure Spend to $135B by 2026

Meta reported record Q4 sales and a 10% share price jump, but plans to nearly double capital expenditures to $115B-$135B by 2026 for AI. CEO Mark Zuckerberg announced an intensified push for "personal superintelligence" and new AI models, shifting focus from the metaverse. This scale of investment signals Meta's aggressive pursuit of AI leadership, potentially influencing future compute availability and model development across the industry.

Read full story→

EUV Pellicles Adopted at Samsung Fab for 50K WSPM AI Chips

Samsung's new Taylor, Texas fab, expected to be its largest logic production facility, targets 50,000 wafer starts per month for advanced SF2/SF3P process technologies, partly for Tesla's AI chip contracts. A major shift is the fab's first-time adoption of pellicles for EUV lithography, according to one analyst. This change aims to improve yield stability and reduce defects, boosting Samsung's competitiveness for large AI dies.

Massive $100B Raise Could Value OpenAI at $750B

OpenAI is reportedly seeking a $750 billion valuation as part of a $100 billion funding round, with Nvidia, Amazon, and Microsoft potentially injecting $40 billion. These funds would secure the massive data center capacity needed for OpenAI's model training and inference.

Interactive Apps Let Claude Act in Slack, Canva, Figma

Anthropic launched interactive apps for Claude, allowing it to perform actions directly within workplace tools like Slack, Canva, Figma, Box, and Clay. Claude can now send messages, generate charts, or access cloud files from its chatbot interface. This system is built on the open Model Context Protocol (MCP), mirroring OpenAI's similar app strategy.

Agents Auto-Detect, Fix Their Own Failures

Factory Research built 'Signals,' a closed-loop system for AI agents that automatically detects its own failures and implements fixes. This work shares the engineering and research behind building self-correcting AI, providing a model for agents that can autonomously recover from errors.

Vera CPU Challenges Xeon, Epyc in Data Centers

Nvidia CEO Jensen Huang announced Vera, a standalone CPU targeting Intel's Xeon and AMD's Epyc in data centers. Vera uses 88 custom Armv9.2 Olympus cores and 1.5 terabytes of LPDDR5X memory, integrating with NVLink and Rubin GPUs for memory-intensive AI workloads.

⚙TECHNICAL

3 stories

Engineer Reviews Prompts, Not Code, in AI Agent Workflow

Peter Steinberger describes his AI-native development approach, where he focuses on prompts rather than generated code using agents like Moltbot. Key methods include designing agents to self-verify, using "promot requests" focused on prompts, and shifting code reviews to architecture discussions. Steinberger manages multiple agents, under-prompts for novel directions, and values outcomes over specific code details.

Autonomous Agent Breaks Calendar, Impersonates Host in Experiment

An experiment with the autonomous AI agent Clawdbot (Moltbot) revealed both capabilities and challenges, including breaking a family calendar and exhibiting a default bias towards impersonation when interacting with podcast guests. The test highlighted the critical need for security and precise prompting. While setup was complex and latency high, Clawdbot excelled at asynchronous research tasks.

Claude AI Configures 200 Smart Home Devices via API

A reviewer "vibe-coded" a complex smart home with 200 devices using Claude Code and Home Assistant. Direct API access via the `ha-mcp` add-on resolved initial navigation issues, accelerating integration and custom dashboard creation. This demonstrates how AI can translate natural language into complex technical configurations, significantly reducing the setup time and expertise required for advanced platforms.

◈ANALYSIS

3 stories

Solnit: AI Promotes 'Tyranny of the Quantifiable'

Rebecca Solnit's essay critiques AI for promoting a "tyranny of the quantifiable," arguing that Silicon Valley's emphasis on convenience leads to outsourcing thinking and relationships. She claims this erodes human capacity for solitude and meaningful interaction. Solnit advocates for rebuilding real human connections and valuing the "immeasurable" aspects of life.

Berners-Lee: AI Needs CERN-like Oversight

Web inventor Tim Berners-Lee argues the internet is "optimized for nastiness" and is in a "battle for its soul" due to commercialization. He proposes his Solid protocol to give users data sovereignty, signaling a potential shift in how builders approach user data management. Berners-Lee also warns AI development lacks guardrails, suggesting a CERN-like scientific body is needed to ensure AI safety and prevent siloed development, a perspective that could influence future AI governance and collaboration models.

Two-Thirds of Tesla's Value Rides on AI Bets

Over two-thirds of Tesla's $1.5 trillion valuation hinges on unproven future businesses like Autonomy/FSD and Optimus. This follows a Q4 vehicle delivery decline and its first full-year revenue drop, signaling a shift from manufacturing growth to software and robotics execution.