AI News Roundup: OpenAI Kills Sora, White House AI Framework, Shopify Agentic Storefronts

OpenAI Kills Sora, Pivots to Robotics

OpenAI confirmed on March 24 that it is shutting down Sora, its AI video generation platform, just six months after launching the dedicated mobile app. The shutdown also scuttled a planned $1 billion partnership with Disney. CEO Sam Altman told staff the decision was driven by economics: Sora was burning through an estimated $15 million per day in inference costs while generating only $2.1 million in total lifetime revenue.

The compute freed up by Sora’s shutdown is being redirected toward OpenAI’s new robotics research initiative. The Sora research team will pivot to “world simulation for robotics” — using video-generation expertise to help robots understand and navigate physical environments. It’s a dramatic bet that the future of AI lies in embodied systems, not consumer video tools.

White House Releases National AI Policy Framework

On March 20, the White House published its National Policy Framework for Artificial Intelligence, a four-page set of legislative recommendations aimed at Congress. The headline move: the administration wants federal law to preempt the growing patchwork of state AI regulations, arguing that inconsistent rules across states are stifling innovation.

The framework covers seven priority areas including child safety requirements for AI platforms, sector-specific regulation through existing agencies (no new AI regulatory body), streamlined permitting for data centers, and letting courts resolve whether AI training on copyrighted works violates IP law. Notably, the framework would shield AI developers from liability for unlawful third-party use of their models — a provision that consumer advocacy groups have already pushed back on.

Shopify Agentic Storefronts Go Live in ChatGPT

Shopify flipped the switch on “Agentic Storefronts” on March 24, making products from over 5.6 million merchants discoverable inside ChatGPT, Microsoft Copilot, Google’s AI Mode, and the Gemini app. The feature launched by default — merchants don’t need to install apps or pay extra fees beyond standard processing rates.

The key architectural shift: purchases now complete on merchant storefronts, not inside the AI chat. This replaces OpenAI’s earlier “Instant Checkout” feature and gives merchants more control over the buying experience. For brands not on Shopify, a new Agentic plan lets them add products to Shopify’s catalog to reach shoppers across all supported AI channels.

Google Launches Gemini 3 Deep Think for Science and Engineering

Google released a major update to Gemini 3 Deep Think on March 26, a specialized reasoning mode built for hard scientific and engineering problems where data is messy and there’s no single correct answer. The mode is available to AI Ultra subscribers in the Gemini app, with early API access opening for researchers and enterprises.

Early results are promising. A Rutgers mathematician used Deep Think to review a highly technical paper and it identified a subtle logical flaw that had passed through human peer review. Duke University’s Wang Lab used it to optimize crystal growth fabrication methods for semiconductor materials discovery. Google also launched Lyria 3 models alongside Deep Think, expanding its multimodal strategy into music generation.

Harvey AI Hits $11 Billion Valuation

Legal AI startup Harvey raised $200 million at an $11 billion valuation on March 25, co-led by GIC and Sequoia Capital. The round closed just three months after Harvey raised at an $8 billion valuation in December — a 37.5% jump that underscores investor appetite for vertical AI applications over pure foundation model plays.

Harvey has now raised over $1 billion in total capital. The company has partnered with most of the 100 largest U.S. law firms, over 500 in-house legal teams, and 50 asset management firms across 60 countries. The funding will go toward expanding AI agents and the legal engineering teams embedded with customers.

Meta Unveils Four Generations of MTIA Chips on RISC-V

Meta announced the MTIA 300, 400, 450, and 500 chip lineup on March 13, all built on the open-source RISC-V architecture in partnership with Broadcom and fabricated by TSMC. The MTIA 300 is already in production, powering training for ranking and recommendation systems across Facebook and Instagram, with three more generations rolling out roughly every six months through late 2027.

The performance trajectory is aggressive: Meta reports a 4.5x increase in HBM bandwidth and a 25x increase in compute FLOPs from the MTIA 300 to the MTIA 500. By choosing RISC-V over Arm, Meta avoids licensing royalties and gains full flexibility over its instruction set — a strategic move to reduce dependence on NVIDIA for inference workloads.