During the CES tech event in Las Vegas, Nvidia provided an in-depth outline of Vera Rubin, its latest AI-focused computing platform designed specifically for data centers. Initially previewed earlier, the company now detailed the platform’s architecture, capabilities, and a timeline for its product rollout, which is scheduled for the second half of 2026.
Nvidia, well-known as a foundational player in the AI technology surge, briefly achieved a remarkable $5 trillion valuation last year, largely propelled by the extensive adoption of its AI chips and computing frameworks. Despite this ascent, the company is confronting concerns regarding a potential AI bubble driven by rapid investment inflows, coupled with growing competition from other technology firms striving to develop proprietary AI hardware to reduce dependence on Nvidia’s solutions.
Nvidia’s CEO Jensen Huang, appearing in his characteristic leather jacket, addressed the audience from the Fontainebleau theater in Las Vegas. He highlighted the ongoing reallocation of corporate budgets away from traditional computing research and development towards AI-centric initiatives. "People ask, where is the money coming from? That’s where the money is coming from," Huang stated, clarifying that the increased AI investment is sourced by pivoting existing tech expenditures.
The Vera Rubin platform is Nvidia’s response to the escalating computational requirements posed by sophisticated AI models. The company emphasized the platform’s potential to accommodate increasingly complex data processing tasks that current infrastructures struggle with. According to a company statement, Vera Rubin’s AI server rack, the NVL72, offers bandwidth capacities that surpass the entire internet, showcasing a groundbreaking leap in connectivity and data handling.
Nvidia detailed innovations within Vera Rubin, notably a novel storage system engineered to enable AI models to process context-rich and intricate information considerably faster. This development addresses bottlenecks faced by existing storage solutions and GPU memory architectures, which may become insufficient as AI applications evolve beyond simple chatbot interactions towards more advanced AI assistants, exemplified by projects from Google, OpenAI, and Anthropic.
During the presentation, Huang illustrated a transition from basic chatbots to intelligent agent models. A broadcasted demonstration featured a tabletop robot functioning as a personal AI assistant by integrating several AI models via the Nvidia DGX Spark desktop system. The robot performed tasks such as recalling its user's to-do list and instructing a dog to vacate a couch, showcasing the real-world utility of multi-step reasoning AI assistants. Huang remarked that building such applications would have been implausible just a few years ago and now is "utterly trivial" due to the availability of large language models superseding traditional programming methods.
The narrative highlighted a shift in bottlenecks within AI systems, as explained by Dion Harris, Nvidia’s senior director of high-performance computing and AI hyperscale solutions. He stressed that the limitations are moving away from raw computing power towards challenges in managing complex contexts effectively. Accordingly, Harris asserted that storage must no longer be an afterthought in AI infrastructure design, highlighting the necessity for comprehensive data solutions.
Further illustrating Nvidia’s strategic moves, the company recently established a licensing agreement with Groq, a specialist in AI inference technology, reinforcing Nvidia’s commitment to investing in inference as a critical AI process. Huang emphasized that AI inference is transitioning into a dynamic reasoning process rather than delivering one-off answers, reflecting the increasing sophistication of AI tasks.
Nvidia confirmed that leading cloud providers such as Microsoft, Amazon Web Services, Google Cloud, and CoreWeave will be early adopters of the Vera Rubin platform. Additionally, hardware integrators like Dell and Cisco are expected to incorporate Nvidia’s new chips within their data center solutions. Premier AI research laboratories at OpenAI, Anthropic, Meta, and xAI are also anticipated to utilize this technology to enhance model training and deliver more sophisticated AI responses.
Beyond data centers, Nvidia expanded its footprint in autonomous vehicle technologies with the introduction of new models termed Alpamayo and advancements in physical AI, the segment focusing on AI-powered robotics and machinery. These developments build upon strategic directions revealed during Nvidia’s GTC conference held in October.
Despite its advancements, Nvidia faces the ongoing challenge of meeting Wall Street’s demanding growth expectations while contending with concerns about the sustainability of AI infrastructure spending. Companies including Meta, Microsoft, and Amazon have collectively devoted tens of billions of dollars in capital expenditures this year. Analysts at McKinsey & Company predict near $7 trillion global investment in data center infrastructure by 2030. However, much of this capital appears concentrated within a relatively small network of tech firms engaging in what is described as circular funding, exchanging money and technology among themselves.
Simultaneously, competitors including Google and OpenAI are developing proprietary AI chips to better tailor hardware to their model specifications, signaling a diversification away from exclusive reliance on Nvidia. The chip maker also faces mounting competition from AMD and Qualcomm, with the latter recently entering the data center market. Technology research head Ben Barringer from investment firm Quilter Cheviot noted that many firms are eager to reduce dependence on Nvidia, seeking to broaden their chip vendor base amid this evolving competitive landscape.