The update comes as investors closely monitor Nvidia to determine whether the robust AI spending led by hyperscale cloud providers is expanding to enterprise customers, sovereign AI initiatives, and AI startups. Market participants are also watching for any signs that increasing competition could affect the company's product roadmap, supply chain, or profitability.

Following discussions with Nvidia's investor relations team, Citi reiterated that Nvidia remains its top large-cap data center semiconductor pick, citing the company's strong ability to secure constrained DRAM memory supply. Nvidia confirmed that its Rubin Ultra roadmap remains fully intact, with no changes to the NVLink domain architecture unveiled at Computex. The company also reaffirmed that co-packaged optics (CPO) for scale-out networking has entered volume production through Spectrum-X, with customer adoption continuing to accelerate.

Starting with the Feynman platform in 2028, customers will have the flexibility to choose between CPO-based and copper interconnect implementations for NVLink. Nvidia declined to comment on Meta Platforms' cloud expansion plans but emphasized that overall AI demand remains exceptionally strong. While hyperscale cloud providers initially accounted for the majority of AI infrastructure deployment, the company noted that AI research labs, sovereign customers, and enterprise on-premises deployments have gained significant momentum over the past two years. As physical AI continues to develop, these segments are expected to account for a growing share of the market.

Nvidia also highlighted that both open-source and proprietary AI models play essential roles within the AI ecosystem. Frontier models continue to drive performance breakthroughs, while open-source models enable enterprises and governments to deploy AI at scale. The company's in-house AI models, including Nem and Cosmos, are designed to support enterprise and sovereign AI adoption rather than compete with leading frontier model developers.

Management also clarified recent comments made by CEO Jensen Huang, who suggested that $100 billion worth of computing power could eventually operate on just one gigawatt of electricity. The company explained that the figure refers to expected improvements in energy efficiency rather than reductions in infrastructure costs. Nvidia estimates that current systems generate approximately $30 billion to $40 billion in computing revenue per gigawatt, while next-generation GPUs such as Blackwell already deliver substantially better energy efficiency than the Hopper architecture.

Market Insight:

Nvidia reaffirmed its commitment to returning 50% of annual free cash flow to shareholders and indicated that its share repurchase program could expand further over time. The company also maintained its gross margin guidance in the mid-70% range and stated that its recent $25 billion bond issuance was intended to provide greater financial flexibility.