Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 1 of 8 IN THE UNITED STATES DISTRICT COURT FOR THE WESTERN DISTRICT OF TEXAS MIDLAND/ODESSA DIVISION NEURAL AI, LLC, Misc. Case No. 7:26-mc-00324-LS Petitioner, Principal case pending in Western District of v. Texas, Civil Action No. 7:24-cv-00221-LS- DTG GOOGLE, LLC, Respondent. NEURAL AI’S REPLY IN SUPPORT OF ITS MOTION TO COMPEL COMPLIANCE WITH SUBPOENA SERVED ON THIRD-PARTY GOOGLE, LLC. i Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 2 of 8 Google attacks a discovery dispute of its own invention. It exaggerates the original subpoenas by reading them as broadly as possible, while ignoring NAI’s repeated efforts to narrow them. Over nearly a month, NAI identified a sufficient subset of information, supplied focused technical questions, proposed and revised a declaration in lieu of production and testimony, and ultimately gave Google two options: execute the declaration or produce documents sufficient to establish the same facts. Google chose neither. Having stonewalled every accommodation, Google now asks the Court to reward its stubbornness. Google cannot reject every narrowing proposal and then oppose the motion based on burdens those proposals would have eliminated. I. NAI SATISFIED LOCAL RULE CV-7(g) Google’s Rule CV-7(g) argument claims the parties’ many conferences about the subpoenas concerned only a declaration, not the twelve requests and five topics. That is wrong. The parties held multiple Zoom conferences and exchanged correspondence over nearly a month. See Ex. 12; Ex. 20. On July 24, NAI explained the twelve requests and five topics, the information sought, and why it was proportional. When Google raised scope and burden, NAI offered focused technical questions and a declaration capturing the same facts sought by the requests and topics in a shorter, less burdensome form. Google’s own August 6 email confirms the parties were negotiating the substance of the subpoenas. Ex. 20 at 5. The declaration was thus not an “ancillary” issue, but was NAI’s proposed compromise on the substance of the subpoenas. That was a sustained effort to resolve the dispute. Exs. 13–14, 20. NAI also addressed Google’s specific objections. On the first call, Rocco Magni identified a sufficient subset of information NAI needed, which was the NVIDIA applications and frameworks Google uses and documents sufficient to define its software stack. NAI later offered to narrow every request to documents sufficient to establish the facts in the declaration. Ex. 12 at 1; Ex. 20 at 1. When 1 Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 3 of 8 Google attempted to rewrite that meet and confer history on the eve of NAI filing, NAI corrected the record: “I’m not sure why you are trying to create a false record. You and I were on a zoom back in July where I talked through a subset of information that we were looking for and said if you provided that information … we would consider Google to have produced sufficient documents.” Opp. Ex. A at 1. Google tries to erase those discussions by denying them now. Google’s written objections also made a request-by-request ritual futile. Google objected to every document request and every deposition topic. By filing, it had identified no request it would answer, produced nothing, and designated no witness. Exs. 9–10; Laiche Decl. ¶ 26. NAI’s certificate is accurate, and identifies the conferences and correspondence addressing the issues raised in the motion. After a month of negotiations, filing was necessary to preserve NAI’s rights. Yet Google made no effort to resolve the dispute. II. THE SUBPOENAS SEEK RELEVANT DISCOVERY Google’s own evidence establishes relevance. Google’s declarant quotes its Form 10-K describing the “foundation” of Google’s “full-stack approach” as “AI-optimized infrastructure” that offers GPUs, powers Google products such as Search and YouTube, and supports Google Cloud customers. Jaffe Decl. ¶ 6. NVIDIA likewise states Google Cloud and NVIDIA have collaborated for more than a decade, “co-engineering a full-stack AI platform that spans every technology layer.” Ex. 15 at 1. Google Cloud offers NVIDIA GPU instances and deploys NVIDIA-accelerated solutions for its customers. Ex. 16 at 1–3. This is evidence of deep integration and deployment, not speculation based merely on purchasing GPUs. Google’s configurations go directly to whether customers use the accused hardware and software to perform the claimed GPU computations and data transfers. Ex. 7 at 2–9, 11–14. NVIDIA has made customer deployment central, contending that NAI must obtain evidence from customers 2 Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 4 of 8 and end users because NVIDIA does not know how they ultimately configure and operate the accused products. Ex. 8 ¶¶ 18–23, 87, 131, 171; Laiche Decl. ¶¶ 24–25. Only Google can supply Google’s internal deployment facts. Moreover, NAI does not need to prove infringement before it asks Google the question. That is not how discovery works. III. THE SUBPOENAS ARE TAILORED AND WERE REPEATEDLY NARROWED Google attacks the broadest imaginable reading of the subpoenas, not the discovery NAI actually sought. Requests 1–5 seek documents sufficient to identify or show the relevant software stack, NVIDIA software and sample code Google uses, how Google software interfaces with NVIDIA functionality, and the architecture and data flow of systems using NVIDIA GPUs. Ex. 5 at 12–14. “Documents sufficient to show” is an express limitation. NAI narrowed further by offering to accept only documents sufficient to establish the facts in the three-page revised declaration. Ex. 20. That does not require every document, source-code file, or historical software version. Requests 6–12 likewise seek operational facts, not Google’s legal conclusions. They ask what Google’s systems do, including whether neural-network computations occur; how pointers, buffers, inputs, outputs, and intermediate results move through memory; and how computations are scheduled, synchronized, and executed. Ex. 5 at 14–17. A fact witness explaining system operation need not offer patent-law opinions. The five deposition topics cover the same subjects and provide an alternative means of filling factual gaps. Id. at 24. Google’s burden evidence is entirely derivative of its inflated reading. Google’s “240 products” refrain is a red herring. The definition of “NVIDIA GPUs” tracks the accused products identified in NAI’s infringement contentions and thus the scope of NVIDIA’s alleged infringement. Ex. 5 at 6–8; Ex. 7 at 2–9. But Google need not investigate GPUs it does not use. It knows which NVIDIA GPUs it purchases and deploys, and its procurement and asset records should identify that 3 Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 5 of 8 subset readily. And Google offers no basis to suggest that each GPU model runs a unique, bespoke software stack. The same frameworks, libraries, and APIs are likely deployed across most—if not all—NVIDIA GPU models. Thus, identifying the relevant GPUs and shared software configurations does not transform one request into “over 240 requests” as Google speculates. The Jaffe declaration also does not substantiate burden. It is from outside counsel, not a Google technical employee or custodian, and identifies no search performed, technical personnel consulted, custodians, repositories, estimated hours, or costs. Instead, it infers from Alphabet’s employee count and broad use of AI that “any number of thousands” of employees might be involved and compliance might take months. Jaffe Decl. ¶¶ 5–8. It never actually says Google uses most of the 240 GPUs, thousands of employees possess responsive information, or NAI’s narrowed search would actually take months. Conjecture does not satisfy Google’s obligation to state specifically how burden relates to each request. See 611 Carpenter LLC v. Atlantic Casualty Ins. Co., 2024 WL 1977160, at *1 (W.D. Tex. Apr. 30, 2024) (“A party objecting to discovery must state with specificity the objection and how it relates to the particular request being opposed, and not merely that it is overly broad and burdensome.”). Google’s proprietary-information objection is equally abstract. To be clear, NAI does not need access to Google’s unique trade-secret code. NAI needs to know how Google is incorporating NVIDIA’s code, what functionalities it is invoking, and what modifications, if any, it is making. NAI thus seeks documents sufficient to show identified deployment of NVIDIA software and execution facts. The existing protective order and tailored protections can address confidential material, and Rule 45 permits production under specified conditions. Google identifies no specific trade secret that would be disclosed, no specific document that cannot be protected, or reason redaction would be inadequate. Ex. 5 at 3–5; Opp. at 8–10; Jaffe Decl. ¶ 8. 4 Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 6 of 8 Moreover, Google’s time-period objection lacks merit. The time period is tied to the case. September 13, 2018 begins the damages period for the infringement conduct at issue. Google offers no evidence that relevant records are inaccessible or that the narrowed “documents sufficient” standard creates burden. Most importantly, NAI took every reasonable step Rule 45 requires to minimize burden. It distilled the subpoenas into focused technical questions, offered to accept a declaration in lieu of any production or testimony, invited Google to revise the declaration for accuracy, revised it to address Google’s stated concern, and ultimately gave Google two options: Google could either “(1) execute[] a revised declaration addressing the issues identified in the [draft declaration] without material gaps, or (2) produce[] documents sufficient to establish each of those issues.” Ex. 20 at 1; Exs. 13–14, 20–21; Laiche Decl. ¶¶ 21–26. Google chose neither. It answered no question, proposed no language, produced no document, designated no witness, and offered no narrower alternative of its own. NAI repeatedly narrowed and Google categorically refused to comply. The Court can also enforce the narrowed relief NAI offered, which is “produc[ing] documents sufficient to establish each of [the] issues” in NAI’s “revised declaration.” Ex. 20 at 1. Only material gaps would require testimony on the corresponding topics. That sequence protects Google without permitting it to avoid relevant discovery by refusing every form of compliance. IV. CONCLUSION The Court should grant NAI’s Motion. At minimum, it should compel Google to produce documents sufficient to establish the facts addressed in NAI’s draft declaration and designate a witness to testify regarding any remaining gaps. Ex. 14. 5 Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 7 of 8 Dated: September 1, 2026 Respectfully submitted, /s/ Rocco Magni Max L. Tribble Texas State Bar 20213950 Brian D. Melton Texas State Bar 24010620 Rocco Magni Texas State Bar 24092745 Samuel Drezdzon Texas State Bar 24117374 SUSMAN GODFREY L.L.P. 1000 Louisiana Suite 5100 Houston, TX 77002 Telephone: (713) 651-9366 Facsimile: (713) 654-6666 mtribble@susmangodfrey.com bmelton@susmangodfrey.com rmagni@susmangodfrey.com sdrezdzon@susmangodfrey.com Tamar Lusztig NY State Bar 5125174 Emily Portuguese NY State Bar 5920327 One Manhattan West, 50th Floor New York, NY 10001 tlusztig@susmangodfrey.com eportuguese@susmangodfrey.com Tanner Laiche WA State Bar 60450 401 Union Street, Suite 3000 Seattle, WA 98101 tlaiche@susmangodfrey.com Attorneys for Petitioner Neural AI, LLC 6 Case 7:26-mc-00324-LS Document 8 Filed 09/01/26 Page 8 of 8 CERTIFICATE OF SERVICE I hereby certify that on September 1, 2026, I electronically filed the foregoing document with the Clerk of Court using the CM/ECF system, which will send notification of such filing to counsel for Google LLC. /s/ Rocco Magni Rocco Magni 7