USA Trends and Developments Contributed by: Steven Lieberman and Nicole (Nikki) DeAbrantes, Rothwell, Figg, Ernst & Manbeck PC
In one example, the USPTO applied the eligibil- ity analysis to claims reciting AI-based methods for analysing speech signals and separating out desired speech from extraneous speech. It determined that the claims recited an abstract idea, but found at least one claim patent eligible because it included limitations “directed to cre- ating a new speech signal that no longer con- tains extraneous speech signals from unwanted sources.” See USPTO July 2024 Subject Mat- ter Eligibility Examples, p. 24. Those limitations turned the abstract idea into a practical applica- tion. The new examples not only help clarify how the eligibility analysis applies in the AI context but also provide insight into the types of AI- related inventions that will be more susceptible to eligibility challenges. With respect to litigation, U.S. district courts began to see the filing of patent infringement lawsuits involving AI-related technology. For example, at the beginning of 2024, FriendliAI Inc. (“FriendliAI”) filed a first amended complaint against Hugging Face, Inc., alleging infringe- ment of two patents directed to techniques for executing batches of requests on an inference system using a transformer neural network mod- el. (See FriendliAI Inc. v Hugging Face , Inc., No. 23-cv-816, Dkt. 27 (D. Del. 8 January 2024)). The patents relate to batching with iteration- level scheduling (also called dynamic batching) which allows an AI model to process multiple requests at once. FriendliAI alleges that Hugging Face’s Text Generation Inference, a server for large language models, contains the patented dynamic batching feature and infringes its pat- ents. The court held a claim construction hearing but before the court issued an opinion the par- ties settled the lawsuit – see FriendliAI Inc., Dkt. 136 (D. Del. 10 January 2025).
In June, ParTec AG and BF exaQC AG filed a patent infringement lawsuit against Microsoft Corporation, alleging that Microsoft’s Azure AI system infringes three patents directed to com- puter cluster-booster systems for processing computation tasks and methods for operating such systems – see ParTec AG et al v Micro - soft Corp ., No. 24-cv-00433 (E.D. Tex. 10 June 2024). In September, Xockets, Inc. filed a federal law- suit against NVIDIA Corporation, Microsoft Cor- poration, and RPX Corporation, alleging that the defendants violated antitrust laws and are infringing seven of its patents – see Xockets, Inc. v NVIDIA Corp ., No. 24-cv-00453, Dkt. 7 (W.D. Tex. 5 September 2024). The asserted patents are directed to semiconductor technology that uses a computing architecture in a data process- ing unit (DPU) to offload, accelerate and isolate data-intensive computing tasks associated with AI, such as those used in cloud infrastructure services and analytics applications used in train- ing large language models for AI. Xockets alleg- es that NVIDIA and Microsoft are directly and indirectly infringing its patents through their use of NVIDIA’s Hopper and Blackwell server com- puter systems which contain three DPUs – the BlueField, ConnectX, and NVLink Switch DPUs – for different data-intensive tasks. Xockets argues that NVIDIA’s cloud computing systems enable “its industry-leading, high-performance training of AI large language models, including by enabling accelerated computing and AI in data centers.” (See id. at 73.) Xockets seeks an injunction preventing NVIDIA from using or sell- ing its Blackwell systems and enjoining Micro- soft’s use of those systems for its generative AI platforms (id. at 186). These two pending lawsuits – both of which are in the early stages – will present important early
589 CHAMBERS.COM
Powered by FlippingBook