Sora Sinks

OpenAI is discontinuing its Sora video generation product line—including the consumer app, a developer-facing version, and video functionality inside ChatGPT—as part of a broader strategic reorientation toward enterprise productivity and coding tools. The move is framed explicitly as a resource allocation decision ahead of a potential IPO as early as Q4 of this year. The core rationale, as the discussion covers, is computational: video generation is extraordinarily GPU-intensive, and every chip powering Sora was unavailable for what OpenAI now considers its primary competitive battleground—coding and enterprise software. The company is under pressure to close the gap with Anthropic, particularly around productizing coding capabilities for non-engineers, in the way Anthropic's Claude has demonstrated with its Cowork product. The strategic pivot centers on merging ChatGPT, the Codex coding tool, and the Atlas browser into a single desktop "super app." This consolidation is designed to unify internal priorities and enable AI agents at scale. The Sora shutdown is characterized as a rebuke to a prior strategy of sprawling product launches that created organizational complexity and competing internal priorities. The collateral damage is notable. Disney, which had committed to a $1 billion investment in OpenAI and a three-year licensing agreement covering more than 200 characters, was blindsided—a Disney team was reportedly mid-project when notified of the shutdown with 30 minutes' notice. The investment never formally closed and no money changed hands. Both parties are now exploring alternative partnership structures. Separately, OpenAI CEO Sam Altman announced a reorganization: safety functions move under the Research organization led by Mark Chen, security moves under Greg Brockman's Scaling organization, and Fiji Simo takes the role of CEO of AGI Deployment, overseeing the bulk of product and operational functions. Altman indicated he will focus on fundraising and data center development. The company's next model, internally codenamed "Spud," has completed pre-training and is moving into safety and security evaluation.

A New Mexico jury found Meta willfully violated the state's Unfair Practices Act by failing to protect minors from predators on its platforms, awarding $375 million in civil damages. The verdict followed a civil trial centered on allegations that Meta misled consumers about the safety of Facebook and Instagram. The state's attorney general had initiated the case after an undercover operation in which a fake profile of a 13-year-old was, according to the attorney general, immediately targeted by abusers. The jury's finding of willfulness is significant: it reflects a determination that Meta's conduct was not merely negligent. Internal communications surfaced during trial reportedly showed Meta employees discussing how CEO Mark Zuckerberg's 2019 decision to implement end-to-end encryption on Facebook Messenger by default would affect the company's ability to report child sexual abuse material to law enforcement—specifically, that some 7.5 million such reports might become undisclosable. Meta has stated it will appeal. A second phase of the trial, beginning May 4 before a judge rather than a jury, will determine whether Meta created a public nuisance and should fund remediation programs. The state is also seeking operational changes including effective age verification and removal of predators from encrypted communications. Simultaneously, China has barred the co-founders of Manus—the AI agent startup Meta acquired for $2 billion—from leaving the country while regulators review whether the transaction violated foreign direct investment rules. The founders were summoned to a meeting with China's National Development and Reform Commission and subsequently told they could not depart, though no formal charges have been filed. The review involves potential violations related to Manus's onshore Chinese entities following an ownership change, and separately, the deal remains under review by China's Ministry of Commerce for potential export control violations. Chinese regulators have expressed concern about what officials describe as "selling young crops to foreign buyers" in strategic AI sectors. An unwinding of the completed transaction is described as a possible but "messy" extreme outcome.

Two funding rounds illustrate the sustained investor appetite for AI companies with defined professional use cases. Granola, an AI meeting note-taking and summarization startup, raised $125 million led by Index Ventures at a $1.5 billion valuation. The company's CEO acknowledged that meeting transcription is already a commoditized category and positioned the raise as funding for an agentic AI roadmap—enabling users to take actions based on accumulated meeting data. Planned integrations include Anthropic's Claude Code. The investment thesis at Index was explicitly tied to agentic potential rather than the note-taking product itself. Harvey, a legal AI company, raised $200 million led by GIC and Sequoia at an $11 billion valuation—up from $8 billion just three months prior in December. The company reported more than 100,000 users across more than 1,300 organizations. The rapid valuation step-up in a single quarter reflects the premium being placed on vertical AI platforms with demonstrated enterprise adoption.

In a unanimous ruling, the Supreme Court held that Cox Communications could not be held liable for customers' online music piracy. The court established that an internet service provider is not liable for merely providing services knowing some users will infringe copyrights; liability attaches only if the provider intended its service to be used for infringement or actively encouraged it. The decision has implications for intermediary liability more broadly, though the court did not resolve all open questions about platform responsibility for user-generated content. --- Key takeaways: - OpenAI's Sora shutdown signals that compute scarcity is now a primary strategic constraint, forcing explicit trade-offs between consumer-facing experimentation and enterprise-focused revenue generation ahead of an IPO. - Meta's $375 million New Mexico verdict, combined with the ongoing Manus regulatory blockade in China, represents a compounding legal and geopolitical risk profile that will likely intensify scrutiny of the company's platform safety practices and international M&A activity. - China's detention of the Manus founders introduces a new category of deal risk for cross-border AI acquisitions involving companies with Chinese operational roots, regardless of where they are formally headquartered. - Vertical AI valuations are accelerating sharply—Harvey's step from $8 billion to $11 billion in three months—suggesting investors are pricing in winner-take-most dynamics in professional AI tooling before market consolidation occurs. - The Granola raise illustrates a strategic pattern: use a commoditizing product category as a distribution wedge, then pivot to agentic capabilities as the defensible long-term value proposition.