The AI Dispatch

Cloud & Capital

Microsoft and OpenAI Tear Up Exclusivity — AGI Clause Gone, Deal Extended to 2032

Azure is no longer the only cloud that can run OpenAI’s products. A restructured agreement uncaps what OpenAI owes Microsoft and strips the provision that once froze the partnership if AGI arrived.

By AI Dispatch Desk · April 27, 2026 · Source: Microsoft Blog

Microsoft and OpenAI announced Monday a comprehensive restructuring of the partnership that has defined the AI industry since 2019, ending Azure’s exclusive cloud hold on OpenAI products while simultaneously extending the overall agreement through 2032. Under the new terms, OpenAI can now ship its models and services across any cloud provider it chooses — Azure’s position shifts from exclusive to “preferred,” with OpenAI products continuing to launch first on Azure before appearing elsewhere.

The most closely watched change involves the revenue share OpenAI has historically owed Microsoft. The new structure introduces a cap on that figure through 2030, giving OpenAI greater financial clarity as it scales toward a projected $100 billion in annualized revenue. The exact cap was not disclosed in the public announcement, but people familiar with the negotiations described it as “substantially below what the previous uncapped structure would have required.”

Equally significant is the removal of the AGI trigger clause — a provision in the original agreement that suspended the partnership’s financial terms if OpenAI’s own board determined it had achieved artificial general intelligence. The clause, long a source of legal and practical ambiguity, had become a flashpoint during OpenAI’s board crisis in late 2023. Its deletion signals both companies prefer a commercially predictable relationship over a structure tied to contested definitions of machine cognition. Microsoft retains its equity stake and preferred-partner status through 2032.

Courtroom

Musk v. Altman Trial Opens in Oakland; OpenAI Counsel: “He Quit, Saying They Would Fail for Sure”

A nine-person jury is seated at the Ronald V. Dellums US Courthouse as the most consequential AI litigation in history formally begins. Musk seeks $130 billion in damages and Altman’s removal from OpenAI.

By AI Dispatch Desk · April 27, 2026 · Sources: CNBC • CNBC Live

The trial of Musk v. Altman et al. officially began Monday before Judge Yvonne Gonzalez Rogers in the Ronald V. Dellums Federal Courthouse in Oakland, California, with jury selection completed and opening statements delivered. Nine jurors were empaneled after several days of voir dire over the preceding week. The case centers on Musk’s claim that OpenAI broke its founding charitable mission when it restructured as a for-profit entity, and that he was defrauded in the process.

OpenAI’s lead counsel delivered a blunt opening that set the adversarial tone for what could be a multi-week proceeding: “We are here because Mr. Musk didn’t get his way at OpenAI. He quit, saying they would fail for sure.” The argument is that Musk’s involvement was always contingent on controlling the organization, and that when that control was denied, he departed — before the company became the most valuable AI enterprise in history.

Musk’s legal team countered that OpenAI’s original nonprofit charter constituted a binding commitment to develop AI “for the benefit of humanity,” and that the for-profit conversion violated both that charter and Musk’s reasonable reliance on it when he donated roughly $44 million in the early years. Musk’s lawyers are seeking a court order compelling OpenAI to roll back its for-profit structure, oust Sam Altman as CEO, and pay approximately $130 billion in damages — a figure that roughly corresponds to their estimate of OpenAI’s current enterprise value.

Open-Source Milestone

DeepSeek V4-Pro (1.6T) and V4-Flash (284B) Land on Hugging Face Under MIT License

The largest open-weight MoE ever released, V4-Pro delivers frontier-class long-context performance. Both variants are available for self-hosting, fine-tuning, and auditing — no licensing restrictions.

By AI Dispatch Desk · April 27, 2026 · Source: WinBuzzer

DeepSeek posted the full open weights for both V4-Pro (1.6T total parameters, Mixture-of-Experts architecture) and V4-Flash (284B) to Hugging Face on Monday, three days after a preview window that allowed limited API access. The MIT license terms are unrestricted: researchers, enterprises, and developers can self-host, fine-tune, fork, and audit the weights without royalties or usage constraints — the same permissive approach DeepSeek took with V3.

The release places the most capable open-weight, long-context MoE architecture ever publicly available into OSS hands. V4-Pro’s parameter count surpasses any previously open-licensed model by a substantial margin, and its MoE design — activating only a fraction of parameters per token — makes inference tractable on well-resourced consumer or cloud hardware. The V4-Flash variant targets latency-sensitive deployment scenarios. Community benchmarking is expected to begin immediately, with comparisons against GPT-5.5 and Gemini 3 Flash already underway in the research community.