Aug. 30 AI brief: Sony Music and Warner sue Anthropic, Nvidia moves beyond the GPU

Coverage window: Aug. 29 through the morning of Aug. 30, 2026 (Asia/Dhaka). Two developments put pressure on different parts of the AI business: music publishers are asking a court to impose billions in copyright damages on Anthropic, while Nvidia is selling more of the machinery that keeps large AI systems moving efficiently.

Sony Music Publishing, Warner Chappell, and other music publishers filed the case in the U.S. District Court for the Northern District of California. The complaint names Anthropic, Dario Amodei, and Benjamin Mann as defendants. The publishers allege that Anthropic used torrenting, web scraping, and downloads to obtain copyrighted works for Claude's training data. Anthropic told TechCrunch that it disagrees with the claims and plans to defend itself in court. 1

The potential damages make the filing larger than a dispute over a few unlicensed songs. The Verge reports that the plaintiffs seek up to $150,000 per copyrighted work and up to $25,000 for each instance in which identifiable copyright information was removed. The complaint covers tens of thousands of works, so a maximum award could reach several billion dollars. Those figures describe the publishers' requested damages; the case remains at the filing stage. 2

The filing also connects music rights to Anthropic's earlier book-training dispute. TechCrunch reports that Anthropic was ordered to pay $1.5 billion in the earlier Bartz case after a judge found that acquiring copyrighted works through piracy was unlawful, even though the judge allowed the use of copyrighted works in training under the circumstances described there. The new plaintiffs allege a broader version of the same acquisition problem, including books that contain lyrics and sheet music. 1

The next checkpoint is the complaint's treatment by the court and Anthropic's first substantive response. Developers and model buyers should also watch whether licensing terms become more explicit for training corpora that combine books, lyrics, and other commercial works.

Nvidia's advantage has traditionally been associated with GPUs. TechCrunch's Aug. 29 analysis describes a broader product stack in Nvidia's Vera Rubin rollout: the Rubin GPU sits alongside the Vera CPU, the Groq 3 LPX inference accelerator, and racks for storage and networking. The surrounding components handle how data reaches the accelerator and how the larger installation shares memory and work. 3

The bottleneck matters because a large model request spends time moving data as well as calculating tokens. Jason Hardy, Nvidia's vice president of storage technology, told TechCrunch that the Vera CPU produced up to a 3x improvement in the operations under discussion by helping Nvidia use flash storage without the same data-flow bottleneck. The claim comes from Nvidia's executive in a reported conversation, and the article does not present it as a common benchmark across models or vendors. 3

OpenAI's Jalapeño chip takes a different route: OpenAI said its design minimizes data movement by keeping a workload inside one connected system. TechCrunch presents the two approaches as responses to the same engineering constraint—reduce the time and energy spent moving data—while Nvidia spreads the work across a larger set of specialized components. 3

The shift changes what a serious infrastructure comparison has to include. A buyer evaluating an AI cluster will need figures for accelerator throughput, memory access, storage traffic, network coordination, power use, and the software that joins those pieces. Nvidia still faces competition from hyperscalers and chipmakers building their own components, so the next useful evidence is system-level measurement under comparable workloads rather than another isolated chip claim.