Jensen Huang took the stage personally at a Morgan Stanley investor roadshow, delivering a core message: growth is not only not peaking but is actually accelerating.
This week, Morgan Stanley hosted a non-deal roadshow (NDR) for NVIDIA in California. NVIDIA CEO Jensen Huang, CFO Colette Kress, and Head of Investor Relations Toshiya Hari attended in person, meeting with numerous institutional investors.
The direct participation of the executive team itself is a signal—the company aims to directly address market concerns about product timelines, ASIC competition, and the sustainability of growth.
Morgan Stanley analyst Joseph Moore subsequently published a report stating the meeting had a "positive tone," with NVIDIA framing the current phase as one of "accelerating growth." He noted that even as quarterly revenue approaches $100 billion, the pace of growth is expected to continue increasing. The firm believes this growth narrative is attractive to both value and growth investors, reiterating NVIDIA as its top pick in the semiconductor sector and maintaining an "Overweight" rating.
Rubin Delay Rumors? NVIDIA Denies
Prior to the roadshow, market rumors suggested the Rubin Ultra might be delayed until 2028. Jensen Huang directly denied this claim during the meeting.
Analyst Joseph Moore stated the Rubin Ultra remains on track for shipment next year. While there have been some adjustments to the rack design for the Rubin system—with the original Kyber rack plan being replaced by a "better solution" that may support larger computational domains—these are optimizations at the system architecture level. The 800V power supply and optical interconnects between racks are proceeding as planned, with no material change to the product timeline.
A Single ASIC Customer's Compute Share Rises to 50%, Anthropic?
The most closely watched detail from the roadshow pertained to changes within the AI lab customer base.
Joseph Moore described in his report: "AI labs currently represent ~20% of total NVIDIA demand. Notably, one fairly representative frontier model that was previously developed primarily on ASICs with very little NVIDIA involvement has now moved to ~50% NVIDIA, while other frontier models continue to run primarily on NVIDIA."
The firm did not directly name the customer. However, based on the characteristics of being a "frontier model" and "primarily using ASICs," the market widely believes this points to Anthropic—whose backer, Amazon, is a major proponent of the Trainium chip.
This change directly addresses a core market concern: will the aggressive development of in-house ASICs by cloud providers erode NVIDIA's share?
Moore's assessment is that both things can happen simultaneously. Hyperscale cloud vendors can continue developing custom chips while NVIDIA maintains a high market share. The reason is that customers ultimately compare not the price of a single chip but the total cost per token. Citing industry checks, Moore noted that "NVIDIA solutions still offer a lower cost per token in many scenarios," keeping them competitive in both training and inference workloads.
Moore also pointed out that NVIDIA's overall share in AI compute is actually increasing from 2024 to 2026.
Diversified Growth Drivers: Three Lines Opening Simultaneously
Analysts outlined three growth vectors based on NVIDIA's new business classification:
First, AI labs (approximately 20% of total demand). Beyond leading models continuing to deeply utilize the NVIDIA platform, customers previously leaning towards ASICs are also increasing their GPU allocations.
Second, traditional hyperscale cloud vendors (approximately half of revenue). Microsoft, Meta, Amazon, and Google remain the largest customer group, but their expansion is increasingly constrained by power, land, and data center build-out speeds. NVIDIA's revenue from this segment is expanding from GPUs to include CPUs and networking equipment.
Third, new AI clouds, sovereign AI, industrial, and enterprise customers. Analyst Moore believes these customers, facing constraints in space, power, and geopolitics, are inclined to procure more integrated AI infrastructure solutions. Their future growth rate may surpass that of traditional hyperscale cloud vendors.
Sovereign AI is particularly noteworthy. Countries, motivated by data security and industrial autonomy, are building local models and compute infrastructure, with such projects being relatively less affected by in-house ASIC competition.
CPU & Networking: NVIDIA's Addressable Market Continues to Expand
NVIDIA reiterated during the roadshow that its CPU business target for this fiscal year is approximately $20 billion.
Moore noted that close to half of this may come from standalone CPU racks—meaning the Vera CPU is not just for management nodes within GPU servers but is entering the broader server market. The Vera chip is designed for single-threaded workloads, utilizing larger die sizes, fewer cores, and memory optimizations for AI scenarios.
The networking business is similarly expanding NVIDIA's revenue boundaries. As AI cluster sizes grow, data transfer between GPUs becomes more likely to be a bottleneck. NVIDIA is transitioning from a pure GPU supplier to an AI infrastructure platform encompassing GPUs, CPUs, networking interconnects, and system architecture.
Beginning to Court Value Investors
Analyst Moore pointed out that NVIDIA is actively working to broaden its investor base, making value-oriented investors a key focus of its communications.
The reason is that NVIDIA is already heavily owned by growth funds, with some institutional holdings approaching single-stock limits. Moore anticipates that NVIDIA may use over 50% of its future cash flow for buybacks and shareholder returns, giving it the cash flow characteristics of a value stock while maintaining rapid growth.
Moore forecasts NVIDIA's revenue to grow 82% in fiscal 2026 and 52.4% in fiscal 2027, with a $288 price target implying approximately 42% upside from current levels.
Strong Growth Outlook, But Valuation and Supply Remain Key Variables
Morgan Stanley maintains an "Overweight" rating on NVIDIA with a $288 price target. NVIDIA closed at $202.78 on July 9th, corresponding to a market capitalization of approximately $4.97 trillion.
The firm expects NVIDIA's revenue to grow 82% in fiscal 2026 and 52.4% in fiscal 2027. Its core thesis is that generative AI will continue driving growth in cloud computing capital expenditure, Blackwell will remain a key solution for generative AI workloads, and the subsequent Rubin products are expected to maintain the company's performance leadership.
However, Moore also noted that the risks facing NVIDIA have not disappeared. If supply catches up with demand faster than expected, growth in the data center business could slow significantly. Other risks include:
A significant reduction in AI development costs;
Competitors launching more competitive products;
And customers accelerating the deployment of in-house custom hardware.
Based on the information released from the current roadshow, NVIDIA's primary challenge is not whether AI demand exists, but rather how to convert that demand into deliverable systems revenue amid multiple constraints including memory, networking, power, and data center space.
Comments