Top Stories
OpenAI and Microsoft Rewrite Their Partnership — Exclusivity Is Over
OpenAI and Microsoft announced a sweeping restructuring of their partnership on Sunday that ends Microsoft’s exclusive right to sell OpenAI’s models. OpenAI can now serve products to customers on any cloud provider, opening the door to deals with Amazon, Google, and others.
The practical effect: OpenAI is no longer locked to Azure. Microsoft remains the “preferred” cloud partner, and OpenAI products will ship on Azure first — unless Microsoft can’t or won’t support the necessary capabilities. Microsoft retains a license to OpenAI’s intellectual property for models and products through 2032 and keeps its stake as a major shareholder.
The financial terms shifted too. Microsoft will stop paying a revenue share to OpenAI. OpenAI will continue paying Microsoft a revenue share through 2030, but that’s now subject to a cap. The old AGI clause — which would have let OpenAI stop paying Microsoft entirely if it achieved artificial general intelligence — is gone. Both sides apparently decided that contract provision created more problems than it solved.
The restructuring also resolves a legal complication. Microsoft had faced potential challenges over OpenAI’s separate $50 billion Amazon cloud deal, which conflicted with the old exclusivity arrangement. With exclusivity removed, that tension disappears.
Read another way: OpenAI got its freedom to shop around, and Microsoft got certainty — a capped liability, continued IP access, and preferred-partner status without the open-ended financial exposure of the old deal.
Sources: Microsoft Blog · OpenAI · TechCrunch · CNBC · Bloomberg
China Vetoes Meta’s $2 Billion Acquisition of AI Agent Startup Manus
China’s National Development and Reform Commission ordered Meta to unwind its acquisition of Manus, the agentic AI startup that Meta bought for roughly $2 billion in December 2025. The decision comes after a months-long investigation into whether the deal violated Chinese export control and technology transfer laws.
Manus was founded by Chinese engineers who relocated to Singapore before the acquisition. Meta had planned to integrate the startup’s agent technology directly into Meta AI. China’s position is that the deal constituted an unauthorized transfer of Chinese-developed technology to a U.S. company, regardless of Manus’s Singapore incorporation.
Meta says the transaction “complied fully with applicable law” and expects “an appropriate resolution.” That resolution may not come easily. This is one of China’s most significant interventions in a cross-border AI deal, and it sets a precedent that extends beyond bilateral U.S.-China tensions. Any AI startup with Chinese founders or engineering teams now has to consider whether Beijing will treat their technology as subject to Chinese jurisdiction, even if the company was incorporated elsewhere.
For the broader market, it signals that cross-border AI M&A is becoming a national-security issue on both sides of the Pacific. The U.S. already restricts chip exports to China and reviews inbound Chinese investment. China is now applying equivalent scrutiny to outbound technology flows.
Sources: CNBC · Bloomberg · NPR · TechCrunch
FISA Section 702 Hits Deadline With AI Surveillance Reform Still Unresolved
Congress has two days to reach a deal on FISA Section 702 before the surveillance authority’s 10-day extension expires on April 30. The law authorizes warrantless electronic surveillance of foreign targets abroad, but civil liberties groups have raised alarms that AI is amplifying its impact on Americans whose communications get swept up incidentally.
A five-year reauthorization attempt failed on April 17 when 20 Republican dissenters joined most Democrats to block it. Congress passed a 10-day stopgap instead. House Speaker Mike Johnson then released the Foreign Intelligence Accountability Act, proposing a three-year reauthorization with reforms: monthly Civil Liberties Protection Officer reviews of U.S. person queries, attorney-level approval requirements replacing field supervisor sign-offs, and expanded congressional access to FISA Court proceedings.
Senator Ron Wyden used a Senate floor speech to warn specifically about AI-powered surveillance — the use of machine learning to mine the massive datasets collected under 702 and commercially available location and behavioral data. The concern isn’t hypothetical. AI can now cross-reference communications data with location tracking, purchase records, and social media activity at a scale that would have been impossible with manual analysis.
Whether the reform provisions survive the final deal is unclear. Intelligence agencies argue any additional oversight requirements will slow operations at a time of increased threats. Privacy advocates say the current system was built for an era when computers couldn’t do what AI can do now, and the law needs to catch up.
Sources: NPR · CNBC · GovPing · Sen. Wyden
Quick Hits
-
Bipartisan AI bill targets deepfakes and protects whistleblowers. Rep. Ted Lieu (D-Calif.) introduced legislation that would crack down on non-consensual AI-generated images, establish whistleblower protections for people reporting AI safety violations, and create a prize competition for AI research breakthroughs. The bill has Republican co-chair Rep. Jay Obernolte’s backing. CNBC
-
AI supercharging government surveillance capabilities. A bipartisan coalition on Capitol Hill is raising concerns that AI’s ability to search through communications data collected under Section 702 and commercially available datasets is making government surveillance far more invasive than the original legal framework anticipated. Google and Anthropic have both responded to lawmaker inquiries, with Anthropic noting its usage policy bans “unauthorized surveillance or tracking of individuals.” NBC News
-
DeepSeek slashes API prices. Following last week’s V4 preview launch, DeepSeek cut cache-hit pricing across its entire API to one-tenth of previous rates and slashed V4-Pro prices by 75%, intensifying the pricing war in China’s AI market and putting further pressure on U.S. providers charging 6-7x more for comparable performance. TechBriefly · The Next Web
Worth Watching
The OpenAI-Microsoft restructuring may be the most consequential AI business deal of 2026 so far. The old arrangement was essentially a marriage — Microsoft got exclusivity, OpenAI got infrastructure and capital. The new arrangement is more like a preferred supplier contract with an expiration date. OpenAI can now court Amazon, Google, Oracle, and anyone else willing to run its models. Microsoft keeps IP access through 2032 and shareholder upside, but loses the lock-in that made the original deal so valuable. The AGI clause removal is particularly telling — both companies apparently concluded that trying to contractually define the moment AI becomes “generally intelligent” was more trouble than the clause was worth. For the rest of the industry, the message is clear: the era of exclusive cloud-AI partnerships is ending. Every major AI lab will now be expected to run across multiple clouds.