Agentic Brew Daily

The civil trial kicked off April 28 in N.D. Cal., Judge Yvonne Gonzalez Rogers presiding, scheduled for about three weeks. Musk dropped his fraud claims days before trial — case is now narrowed to unjust enrichment and breach of charitable trust. Musk took the stand as the first witness; Altman is expected to testify for over two hours, Satya Nadella for about one. The jury verdict is only advisory; the judge makes the binding call.

Why it matters: Damages of $130-150B would flow to OpenAI's charitable arm, which already controls ~26% of OpenAI Group PBC. The economics are circular but the governance precedent is enormous, and Microsoft's ~27% stake means Redmond is exposed too.

April 27: Microsoft and OpenAI ended Microsoft's exclusive cloud access; the IP license now extends to 2032 on a non-exclusive basis. Less than 24 hours later, AWS launched a limited preview of GPT-5.5, GPT-5.4, Codex, and Bedrock Managed Agents on Amazon Bedrock. MSFT shares slid as much as ~5% intraday.

Why it matters: The buried lede is that the original AGI clause — under which Microsoft's rights would terminate when OpenAI's board declared AGI achieved — got replaced with a fixed 2032 termination date. That dissolves the most consequential subjective trigger in AI corporate governance, and it happened with almost no fanfare.

Google inked a classified agreement letting DoD run Gemini on classified networks for "any lawful government purpose" — reported $200M contract that obligates Google to adjust AI safety settings and content filters at the government's request. Joins OpenAI and xAI in similar deals after Anthropic refused and was branded a "supply chain risk." 580-600+ Google/DeepMind employees including 20+ directors and VPs signed a letter against it.

Why it matters: The "any lawful government purpose" standard inverts a decade of vendor-led AUP. Instead of vendors enumerating refusals, the customer asserts an open-ended right limited only by U.S. law. Google's specific agreement to loosen safety filters on request goes further than even OpenAI's deal.

WSJ reported April 28 that OpenAI fell short of internal user/revenue targets, including the goal of 1 billion weekly ChatGPT users by end of 2025. Cue an instant AI-equity selloff: Oracle, Nvidia, AMD, Broadcom, Arm, CoreWeave, and SoftBank all slid. CFO Sarah Friar and Sam Altman called the report "ridiculous" and "prime clickbait" — but Friar reportedly told colleagues internally she's unsure how OpenAI will pay for future contracts.

Why it matters: Altman has publicly disclosed roughly $1.4-1.5T in compute commitments over 8 years (Stargate plus a separate $300B+ Oracle deal). Current revenue is ~2% of those commitments. Friar's worry puts her at odds with Altman's reported push for an IPO at a rumored $1T valuation, and Gary Marcus is already comparing OpenAI to WeWork.

OpenAI is reportedly co-developing a custom smartphone processor with MediaTek and Qualcomm, with Luxshare Precision (yes, the iPhone assembler) as exclusive system co-design and manufacturing partner. The device is designed so AI agents replace traditional apps, with continuous real-time user-context tracking. Specs by late 2026 / Q1 2027, mass production targeted for 2028. Qualcomm popped ~12% intraday on April 27, Luxshare ~10%.

Why it matters: Ming-Chi Kuo's argument is that an agent-first device cannot be built on iOS or Android because the permission sandboxes deliberately constrain the persistent context capture an agent needs. Pair the $6.5B Jony Ive io Products acquisition with dual-sourced custom silicon and an exclusive iPhone-assembler partner — this is hardware program territory, not a side project.