Chip giant NVIDIA has been one of the main financial beneficiaries in the artificial intelligence boom.
Revenue surged due to high demand for its AI-learning chips and data center services, and the end doesn’t appear to be in sight.
Besides selling the most sought-after hardware, NVIDIA is also developing its own models, including NeMo Megatron models. These were trained using NVIDIA’s own hardware and with help from large text libraries, much like other tech giants do.
Authors Sue NVIDIA for Copyright Infringement
This includes authors, who, in various lawsuits, accused tech companies of training their models on pirated books. In early 2024, for example, several authors, including Abdi Nazemian, sued NVIDIA over alleged copyright infringement.
Through the class action lawsuit, they claimed that the company’s AI models were trained on the Books3 dataset that included copyrighted works taken from the ‘pirate’ site Bibliotik.
As the case progressed, the authors also brought up NVIDIA’s contacts with Anna’s Archive, inquiring about “high-speed access” to the shadow library’s massive collection of pirated books.
NVIDIA Wants Case Dismissed
In January, NVIDIA fired back with a comprehensive motion to dismiss, calling the authors’ allegations speculative, vague, and legally insufficient. At the California federal court, NVIDIA argues that the authors’ complaint is built on speculation rather than facts.
Specifically, the company asked the court to dismiss the direct copyright infringement claims linked to Bibliotik, Books3, and The Pile dataset.
In addition, the motion also targets the contributory copyright infringement allegations, which center on scripts and tools NVIDIA allegedly distributed so corporate customers could automatically download ‘The Pile,’ the dataset that contains Books3.
The chip giant initially asked the court to dismiss claims relating to Anna’s Archive, Z-Library, LibGen, Sci-Hub, and the Slimpajama dataset as well, but it withdrew this request in March, which substantially narrowed the dispute.
Scripts Have No Other Purpose than Infringement
In an order issued yesterday, U.S. District Judge Jon Tigar denied most of the dismissal request. Importantly, the contributory infringement claim survives, even after the Supreme Court’s Cox v. Sony ruling, which significantly impacts many copyright infringement cases.
NVIDIA argued that Cox tightened the standard, requiring “active encouragement through specific acts,” while stressing that the NeMo Megatron Framework as a whole has substantial non-infringing uses. Marketing or promoting this framework as a piracy tool was needed to prove this claim, NVIDIA argued.
Judge Tigar rejected the framing. Instead of analyzing the Megatron framework as a whole, he zeroed in on the specific scripts that NVIDIA distributed to clients so they could automatically download and preprocess The Pile dataset. Those scripts have no purpose other than enabling infringement, the court concluded.
“The scripts are alleged to have no other purpose than to speed up the process of infringement, unlike the digital video recorder systems at issue in Sony Corp. or the internet service provided in Cox,” Judge Tigar wrote.
This appears to be the first AI training case to apply the new Cox standard, and the result didn’t go the way NVIDIA hoped. The scripts it offered satisfied both the new ‘inducement’ and ‘tailored to infringement’ standards required for a contributory infringement finding.
BitTorrent Is ‘Merely a Tool’
Regarding the direct copyright infringement claims, NVIDIA also asked the court to dismiss “allegations concerning its ‘use of any [sic] BitTorrent Protocol.'”
The request was pretty thin, Judge Tigar noted, pointing out that the complaint contains exactly one reference to BitTorrent. That reference doesn’t point to any of NVIDIA’s alleged wrongdoing. It’s a descriptive line about Bibliotik distributing pirated works via the protocol.
Judge Tigar refused to dismiss all BitTorrent allegations, stressing that “BitTorrent is merely a tool, not a library or dataset.” He also offered a rather colorful analogy.
“Asking to dismiss allegations concerning BitTorrent is like asking to dismiss allegations concerning paintbrushes in a case about a dolphin painting,” the order reads, citing Folkens v. Wyland Worldwide, a copyright dispute over a painting of two dolphins crossing underwater.
NVIDIA’s interest in stripping BitTorrent from the case is easier to understand in light of Meta’s troubles in a parallel AI lawsuit. There, Meta’s BitTorrent seeding resulted in direct copyright infringement claims. NVIDIA appears to have wanted that door closed before discovery could open it.
Lawsuit Moves Forward
NVIDIA did get a small win as Judge Tigar dismissed the vicarious copyright infringement claim.
To state that claim, the authors needed to plausibly allege that NVIDIA had both the legal right to control the direct infringers and a direct financial interest in the infringement. Tigar found neither was adequately pleaded, but allowed the authors 21 days to address the deficiencies and refile.
For now, it is clear that this legal battle between the authors and NVIDIA is far from over.
The same also applies to a long list of other AI training lawsuits, which continue to grow every month. That includes a lawsuit filed against Meta and Mark Zuckerberg yesterday by major publishers, which, like many others, also accuses Meta of training on pirated books.
—
A copy of U.S. District Court Judge Jon Tigar’s order on NVIDIA’s motion to dismiss is available here (pdf).
