TECHNOLOGY —Nvidia has denied market speculation that its Rubin Ultra platform has been delayed until 2028, according to a Morgan Stanley report following recent investor meetings with the chipmaker.

Nvidia CEO Jensen Huang, Chief Financial Officer Colette Kress, and other senior executives participated in a non-deal roadshow in California last week, discussing product timelines, ASIC competition, and demand for AI infrastructure with institutional investors.

Morgan Stanley analyst Joseph Moore said some rack designs for the Rubin system are being adjusted, but development of 800-volt power systems and optical interconnects remains on schedule. Rubin Ultra is still expected to begin shipping next year.

The report also said AI laboratories currently account for about 20% of Nvidia's total demand. One major frontier model customer that previously relied primarily on ASICs has increased Nvidia's share of its computing capacity to nearly 50%.

Morgan Stanley did not identify the customer, and its identity has not been officially confirmed.

Moore said the expansion of custom ASICs by cloud companies does not necessarily mean Nvidia's market share will decline at the same rate. AI customers also evaluate overall system performance and computing cost per token when selecting infrastructure.

Nvidia's AI demand is increasingly spread across AI laboratories, major cloud providers, emerging AI cloud companies, sovereign AI projects, and enterprise customers.

Large cloud companies including Microsoft, Meta, Amazon, and Google remain important customers, although data center expansion is increasingly constrained by power availability, land, and construction timelines. Sovereign AI projects are also drawing attention as governments invest in domestic computing infrastructure and AI models.

Beyond GPUs, Nvidia is expanding its CPU and networking businesses. The company reiterated a target of approximately $20 billion in CPU revenue for the current fiscal year, with its Vera CPU expected to address a broader range of server workloads.

As AI clusters grow, networking and data transfer between GPUs are becoming increasingly important. Nvidia has expanded its business beyond standalone GPUs into CPUs, networking, and integrated AI infrastructure systems.

Morgan Stanley maintained an Overweight rating on Nvidia and a $288 price target. The firm also identified several risks, including faster supply growth, declining AI development costs, stronger competing products, and increased deployment of custom chips by major customers.

The report said one of Nvidia's key challenges is converting AI computing demand into deliverable system revenue amid constraints involving memory, networking, electricity, and data center capacity.