Nvidia Sees AI Demand Broaden Beyond Cloud Giants, Reaffirms Next-Gen Chip Roadmap

Nvidia Corp. (NVDA) has signaled that demand for its artificial intelligence infrastructure is broadening beyond its core hyperscale cloud customers, with its long-term product roadmap remaining firmly on schedule. The update, detailed in a research note from Citi following discussions with the company's investor relations team, addresses key investor questions about the sustainability of AI spending and Nvidia's execution on future products.

Demand Expands to New Customer Segments

While massive investments from large cloud providers initially drove the AI boom, Nvidia said demand is now strengthening across a wider range of clients. The company highlighted that AI laboratories, sovereign customers, and enterprise on-premise deployments have all gained significant momentum over the past two years.

Nvidia expects these segments will constitute a larger share of the market as the physical adoption of AI accelerates. The company characterized overall AI demand as remaining "very strong," though it declined to comment specifically on the expansion plans of Meta Platforms, a major customer.

Next-Generation Roadmap Intact

Nvidia provided assurances that its ambitious product pipeline is proceeding as planned, a critical factor for maintaining its market leadership.

Ad

• Rubin Platform: The company's next-generation Rubin Ultra roadmap remains "fully intact."
• Networking: The NVLink domain architecture revealed at the Computex conference has not changed. Co-packaged optics (CPO) are already in production with its Spectrum-X networking platform, seeing high customer adoption.
• Feynman Platform: Beginning in calendar year 2028, the Feynman platform will give customers the choice between NVLink implementations using CPO or traditional copper interconnects.

Financial Guidance and Efficiency

Company management reiterated its financial targets, including a gross margin in the mid-70% range and a plan to return 50% of cash flow to shareholders this year, noting that buybacks could increase over time. A recent $25 billion debt offering was intended to provide greater financial flexibility.

Nvidia also clarified recent comments from CEO Jensen Huang regarding power efficiency. A projection that $100 billion of compute could eventually require just one gigawatt of power was meant to reflect expected long-term gains in energy efficiency, not current infrastructure costs. The company estimates its current systems generate roughly $30 billion to $40 billion of compute revenue per gigawatt, with newer GPUs like Blackwell offering significant efficiency improvements over the Hopper generation.