This episode covers the most significant AI developments from the past week, featuring analysis of major funding rounds, executive changes, and strategic positioning among leading AI companies.

The host frames these developments as 'the calm before the AGI storm' - a period where major labs are positioning for rapid transformation ahead of next-generation model releases.

Key stories include OpenAI's record fundraising and internal tensions, Anthropic's code leak and usage limit controversies, Google's breakthrough open source release, and global infrastructure challenges affecting data center expansion.

OpenAI's $122B Funding Success Masks Secondary Market Struggles

OpenAI closed a record $122 billion funding round at $852 billion valuation, adding $12 billion to the previously announced $110 billion from Amazon, NVIDIA, and SoftBank.

Despite strong primary fundraising, Bloomberg reports OpenAI stock struggles in secondary markets with 'hundreds of millions of dollars worth' finding no buyers according to Next Round Capital's Ken Smythe.

Institutional investors show $2 billion in cash ready for Anthropic shares, suggesting better risk-reward at Anthropic's $600 billion implied valuation versus OpenAI's $852 billion.

OpenAI now generates $2 billion in monthly revenue, up from $1.6 billion at year-end, growing at four times the pace of companies that defined the internet and mobile eras.

Executive Upheaval and IPO Strategy Tensions

CEO of AGI deployment Fiji Simo announced medical leave for chronic neuroimmune condition, with President Greg Brockman taking over product organization.

Multiple leadership changes include Brad Lightkap stepping down as COO for special projects and Denise Dresser adding COO role to her Chief Revenue Officer position.

Sam Altman and CFO Sarah Fryer reportedly clash over IPO timing, with Altman pushing for Q4 2025 launch while Fryer questions readiness and $600 billion infrastructure spending commitments.

Sources suggest Altman has excluded Fryer from financial planning conversations, creating 'noticeable and awkward' dynamics despite her role shepherding Square's successful 2015 IPO.

OpenAI Acquires TBPN Tech Talk Show for Media Strategy

OpenAI acquired tech talk show TBPN despite Fiji Simo's mandate to cut 'side quests,' with Simo herself reportedly pushing for the acquisition.

The deal includes strict editorial independence terms, creating tension between supporting OpenAI's focus and maintaining arm's length operations.

Critics argue TBPN indexes highly in tech demographics that aren't OpenAI's main communications challenge, with broader public perception being the bigger issue.

Anthropic's Code Leak Reveals Advanced Agent Features

Anthropic accidentally released 512,000 lines of Claude Code source code, revealing unreleased features including 'Kairos' always-on agent with dream mode and proactive capabilities.

The leak showed Claude Code contains five compression strategies, dozens of tools, caching optimizations for sub-agents, and highly configurable system prompts.

Anthropic's lawyers issued over 8,000 GitHub takedown notices, mostly targeting legitimate public forks, later apologizing and retracting all but one notice.

Usage Limits and Pricing Changes Frustrate Anthropic Users

Users complained about hitting usage limits within hours or minutes, with some claiming caps after first prompts when using Opus model.

Claude co-developer Lydia Halley explained 'peak hour limits are tighter, and million token contact sessions got bigger' but denied overcharging bugs.

Anthropic announced users can no longer use subscriptions for third-party tools like OpenClaw, requiring API usage on pay-per-token basis starting Saturday.

Author Daniel Jeffries argues the 'subsidy era is coming to an end' with true intelligence costs resembling 'full-time salary to people' rather than pennies.

Google's Gemma 4 Delivers Frontier Performance Locally

Google released Gemma 4 open source models including 2B and 4B edge models, plus 26B mixture of experts and 31B dense model ranked third on Arena AI leaderboard.

The models run locally on laptops with 140 language support, 256k context window, and performance better than models '20x its size' according to Greg Eisenberg.

This represents the first Western open source model competing at frontier level in years, with significant implications for local AI deployment.

China's AI Strategy Shifts and DeepSeek V4 Anticipation

Alibaba released three proprietary models in three days, culminating in Quen 3.6 Plus that costs 1/8th of Opus while lagging by only a few points on benchmarks.

The Quen team announced their new model served 'a trillion tokens on release day' and ranked number one on OpenRouter, with 100% token sales flowing to Alibaba.

DeepSeek V4 expected within weeks as tech giants order hundreds of thousands of new Huawei chips, with the model optimized specifically for domestic hardware.

Global Infrastructure Challenges Threaten Data Center Expansion

Iranian Revolutionary Guard declared 18 U.S. tech companies including NVIDIA, Apple, Microsoft, and Google as 'legitimate targets' for retaliation over AI-enhanced targeting.

Three Amazon data centers in Bahrain and UAE were hit by drones, potentially making further Middle East construction unviable.

More than half of U.S. data centers face delays or cancellation due to electrical equipment shortages, with transformers, switchgear, and batteries in short supply.

Asia's $800 billion in planned data projects through 2030 face scrutiny due to energy import dependence and infrastructure constraints.