NVIDIA (NASDAQ:NVDA) executives used a wide-ranging Q&A session to underscore what they described as accelerating demand for AI infrastructure, a faster-than-usual product cadence heading into the Vera Rubin platform, and expanding opportunities in areas such as autonomous vehicles, agentic AI, physical AI, networking, and storage.
Benchmarking, systems co-design, and the Blackwell step-up
In response to questions about AI performance testing and comparative results, NVIDIA CEO Jensen Huang argued that rigorous benchmarks are difficult to complete and suggested that not all participants ultimately submit results. Huang pointed to SemiAnalysis as a “living, breathing benchmark” he said he likes because it is continuously updated.
He also referenced a third-party analysis (he cited Signal65) that he said showed roughly a 10-to-1 improvement moving from Hopper to Blackwell along with a “10 to 1 reduction in cost,” despite only about a two-times increase in transistor count. Huang used that comparison to argue that gains now require broad architectural changes rather than incremental chip improvements, citing Amdahl’s Law and saying Moore’s Law is “completely over.”
Vera Rubin: “Full production,” timing, and ramp dynamics
Bank of America’s Vivek Arya asked for clarity on what “Vera Rubin in full production” means. NVIDIA confirmed that while the company had previously said the chips had taped out, it now describes Rubin as in full production and reiterated plans to bring it to market in the second half of the year. Huang noted cycle time is “nine months plus.”
Later, analysts pressed on how quickly Rubin can ramp and what that means for revenue recognition and product overlap. Huang said Rubin’s ramp “should be fast,” but described Rubin as exceptionally complex—“the only computer in history where literally every single chip is new,” including components such as HBM4 and other system elements that “never existed” before. He said the company worked on Rubin for several years, “something close to five years,” de-risking supply chain and technology elements in advance.
Huang also described system-level reliability improvements, saying NVLink networking in the Rubin generation is hot-swappable and designed to minimize downtime in large AI factories.
Autonomous vehicles, robotics, and “physical AI” as long-cycle bets
Asked about the scope and timing for autonomous driving and robotics opportunities, Huang traced NVIDIA’s autonomous vehicle (AV) efforts back eight years, describing an extended development and safety-architecture timeline culminating in production deployment with Mercedes. He said NVIDIA works with a broad set of automakers and AV developers, citing BYD, Geely, Xiaomi, Stellantis, and multiple robotaxi companies, and said “almost everybody who has a self-driving anything has NVIDIA in the data center.”
Huang characterized the AV business as “multi-billion” and said it is “just getting started,” adding that over the next decade it should become “a very large business,” with the company expecting significant scale by the end of the decade.
On “physical AI,” Huang said it benefits from multimodality and alignment work already done in large language models. He described a strategy of leveraging NVIDIA’s Nemotron models and vision/world-model training to create what he called Cosmo, which he said NVIDIA is providing broadly to lower the barrier for others to build physical AI systems, while still requiring fine-tuning and domain adaptation by end users.
DGX Cloud strategy, NeoClouds, and China supply comments
NVIDIA also addressed how it views cloud partnerships. Huang said DGX Cloud was “never intended to compete” with cloud service providers (CSPs). Instead, he described it as a strategy to help CSPs adopt AI infrastructure, create an “exemplar” NVIDIA cloud environment inside their platforms, attract AI-native developers, and distribute NVIDIA models into CSP ecosystems. He said he believes the “flywheel” is now established and suggested the need to use DGX Cloud as a forcing function has diminished, though NVIDIA still needs substantial compute capacity for its own model-building efforts.
On NeoClouds (GPU-focused cloud providers), Huang said NVIDIA supported them because AI technology and infrastructure requirements were moving quickly and because access to “land, power, and shell” was not assured. He cited examples including Nscale in Europe, Yotta in India, CoreWeave, Lambda, and G42 in the Middle East, and said he expects more regional infrastructure build-outs.
Separately, in response to a question about China, NVIDIA said it has sufficient supply to meet demand in other regions and indicated that any H20 supply for China would be specific to that market and not taken from existing allocations. The company also noted it was awaiting further clarity on licensing processes involving government approvals and acknowledged demand from China.
Networking, storage, and the role of BlueField
Goldman Sachs asked about a “context memory storage controller” NVIDIA announced. Huang framed it as part of an expansion beyond compute into networking and storage, stating NVIDIA is the largest networking company “in the world today” and expects it could become the largest “storage processor” company as well. He pointed to BlueField and the DOCA software stack, saying BlueField is widely adopted outside hyperscalers that build their own SmartNICs.
Huang argued that AI workloads create storage demands unlike traditional structured database use cases. He described AI key-value (KV) cache as “insanely heavyweight” and said traditional north-south storage approaches are inefficient for that workload. He said NVIDIA introduced a new tier of storage integrated into the computing fabric, describing it as a new market that will hold the working memory of the world’s AIs.
Throughout the session, Huang repeatedly returned to two themes: demand strength and the importance of integrated systems engineering. He said NVIDIA’s scale has allowed it to prepare upstream and downstream supply chains over several years, including direct purchasing of large amounts of DRAM and working with every major cloud. He also suggested that future token growth could accelerate beyond recent trends, attributing a prior 5x increase to reasoning models and predicting a larger potential increase from agentic systems involving tool use, planning, and simulation.
About NVIDIA (NASDAQ:NVDA)
NVIDIA Corporation, founded in 1993 and headquartered in Santa Clara, California, is a global technology company that designs and develops graphics processing units (GPUs) and system-on-chip (SoC) technologies. Co-founded by Jensen Huang, who serves as president and chief executive officer, along with Chris Malachowsky and Curtis Priem, NVIDIA has grown from a graphics-focused chipmaker into a broad provider of accelerated computing hardware and software for multiple industries.
The company’s product portfolio spans discrete GPUs for gaming and professional visualization (marketed under the GeForce and NVIDIA RTX lines), high-performance data center accelerators used for AI training and inference (including widely adopted platforms such as the A100 and H100 series), and Tegra SoCs for automotive and edge applications.
