Morgan Stanley: Nvidia management directly responds to market concerns; Rubin roadmap remains unchanged; ASIC competition impact is limited

Morgan Stanley recently organized a Nvidia non-deal roadshow, attended in person by CEO Jensen Huang, CFO Colette Kress, and the head of investor relations. This high-level lineup itself suggests the company wants to address directly several issues the market has been most focused on recently: the Rubin product roadmap, ASIC (custom AI chip) competition, and whether investment in AI infrastructure can be sustained. According to meeting notes, management released several important signals:
1. The company’s quarterly revenue has already entered the $100 billion range, and is expected to keep accelerating over the next few quarters;
2. Demand for the Vera Rubin platform remains very strong, and it is expected to become the most important growth driver over the next 12 months;
3. Management clearly stated that Rubin Ultra will ship as planned next year, denying recent market rumors about delaying it to 2028;
4. The Kyber rack that previously sparked controversy has not been canceled; it is evolving toward a better rack architecture to support larger-scale scale-up. The company emphasized that the roadmap will be flexibly optimized based on engineering progress to reduce the risk of ramp-up in mass production;
5. Key technologies such as an 800V power architecture and optical scale-up between racks are still progressing on the established schedule;
6. Management reiterated again that tight AI memory supply, such as HBM, is expected to continue for the next few years, and memory remains one of the most critical bottlenecks across the entire AI industry chain.
Notably, this roadshow can also be seen as an official response to recent deferment reports from multiple institutions regarding Rubin/Kyber.
Earlier, some research firms believed that due to engineering challenges such as backplane signal integrity and the maturity of CPO switches, Kyber-related solutions might be pushed to 2028. The market once worried that Nvidia’s product cadence could slow down. But Nvidia reiterated in this session that “Roadmap is intact,” while explaining that product form factor adjustments are more about engineering optimization rather than changes in demand or strategy.
On the other hand, ASIC competition remains a topic investors repeatedly ask about. However, based on this exchange, management did not show any obvious concern.
The market broadly believes that self-developed ASICs such as Google TPU, AWS Trainium, Meta MTIA, Microsoft Maia will continue to grow, but they will handle more internal specific workloads rather than fully replacing GPUs.
For cutting-edge AI that needs frequent iteration and supports multi-model training and inference, Nvidia still has a clear advantage by relying on the CUDA software ecosystem, NVLink, the full system approach, and a platform upgrade cadence of one generation per year. Therefore, most sell-side firms currently still judge that ASIC is more likely to be complementary to the GPU market rather than a comprehensive replacement for GPUs.