NVIDIA Roadshow: Revenue Approaching $100 Billion, Rubin Ultra on Schedule

Introduction

NVIDIA recently addressed several key investor concerns during a specialized roadshow hosted by Morgan Stanley in California: whether the Rubin Ultra roadmap has been delayed, whether custom ASICs are eroding GPU market share, and whether the company can sustain growth as quarterly revenue approaches $100 billion.

According to roadshow notes reported by Wall Street Journal, NVIDIA’s response was straightforward: growth has not peaked, and management believes it is accelerating.

CEO Jensen Huang, CFO Colette Kress, and Vice President of Investor Relations and Strategic Finance Toshiya Hari jointly participated in the meeting with institutional investors. The presence of this executive lineup made the event distinct from regular investor meetings, allowing the company to directly address concerns about product cadence, competitive dynamics, supply constraints, and the sustainability of AI infrastructure spending.

Morgan Stanley analyst Joseph Moore gave a positive assessment of the meeting’s tone. The firm maintains NVIDIA as its top semiconductor pick with an "Overweight" rating.

Rubin Ultra Delay Rumors? NVIDIA: Roadmap Unchanged

Before the roadshow, market rumors suggested Rubin Ultra might be delayed to 2028.

According to reports, Jensen Huang directly refuted this interpretation during the meeting. Morgan Stanley’s notes indicate Rubin Ultra is still expected to ship in 2027.

Some adjustments are being made to the rack architecture. The original Kyber design is reportedly being replaced by a configuration NVIDIA considers superior, potentially enabling larger compute domains. However, the roadshow framed this as an architectural optimization, not a timeline adjustment.

Key infrastructure work remains on track:

• 800-volt DC power delivery
• Rack-level liquid cooling
• Inter-rack optical connectivity
• Higher-density compute domains
• Rubin Ultra system-level optimization

NVIDIA’s public roadmap also supports the 2027 timeline. Its official infrastructure materials describe Kyber as a rack generation designed to house 576 Rubin Ultra GPUs, targeting 2027. NVIDIA has also publicly discussed transitioning to 800-volt DC power as a prerequisite for supporting that rack density level.

This distinction is critical: redesigning the rack changes the system form factor but does not delay the underlying product generation. According to the roadshow notes, NVIDIA positions this adjustment as an optimization of deployment architecture rather than a reset of the Rubin Ultra roadmap.

ASIC-Focused Customer Shifts to Nearly 50% NVIDIA Compute

One of the most closely watched details from the roadshow involved an NVIDIA AI lab customer.

Morgan Stanley estimates AI labs currently account for roughly 20% of NVIDIA’s total demand. Most frontier model developers already rely heavily on NVIDIA infrastructure, but one major customer had previously developed models primarily on custom ASIC hardware, with minimal NVIDIA involvement.

According to the roadshow report, NVIDIA’s share of that customer’s compute infrastructure has now risen to nearly 50%.

Morgan Stanley did not publicly disclose the customer’s identity. The source article suggests the description could point to Anthropic, given its close ties with Amazon and use of AWS Trainium chips. This remains market speculation, not confirmed disclosure from NVIDIA, Morgan Stanley, Amazon, or Anthropic.

Even without naming the customer, this shift addresses a key investor concern: custom chips do not necessarily directly replace NVIDIA’s GPUs.

Hyperscalers or AI labs can develop their own ASICs while continuing to purchase NVIDIA infrastructure in large volumes. The final decision depends on the economics of the entire workload, not the price of a single accelerator.

Total Cost per Token Matters More Than Chip Price

Morgan Stanley believes customers ultimately compare the total cost of generating tokens.

This calculation goes beyond chip purchase price:

• Training and inference performance
• Memory bandwidth
• Network efficiency
• Cluster utilization
• Software maturity
• Developer productivity
• Power consumption
• Deployment time
• Large-scale reliability

According to the investment bank’s industry research, NVIDIA continues to deliver a lower cost per token across many workloads. This helps explain why custom ASIC adoption and NVIDIA market share growth can occur simultaneously.

Moore also argued that despite major cloud companies expanding custom chip programs, NVIDIA’s overall share of AI computing is rising from 2024 through 2026.

This does not mean ASIC competition is irrelevant. It suggests the market is growing fast enough and workloads are diverse enough to allow multiple architectures to expand simultaneously while NVIDIA maintains a strong position.

Growth Becoming More Diversified

The roadshow divided NVIDIA’s demand into three major growth engines.

1. AI Labs

Morgan Stanley estimates AI labs account for roughly 20% of total demand.

Large frontier model developers remain heavily dependent on the NVIDIA platform. Meanwhile, customers that previously relied more on custom ASICs appear to be increasing GPU deployment.

This segment is driven by:

• Frontier model pre-training
• Post-training and reinforcement learning
• Long-context inference
• Agentic AI workloads
• Research experimentation
• Rapid scaling of production inference

Demand patterns are becoming broader than a few training clusters. Inference and agentic workloads require continuous capacity expansion as model usage grows.

2. Traditional Hyperscalers

In the roadshow framework, traditional hyperscalers account for roughly half of NVIDIA’s revenue.

Microsoft, Meta, Amazon, and Google remain the largest customer base. Their expansion is increasingly constrained by physical infrastructure rather than insufficient demand.

Major constraints include:

• Power availability
• Grid connections
• Land
• Data center construction
• Cooling
• Memory supply
• Networking equipment
• Deployment lead times

NVIDIA’s opportunity within this group has also expanded beyond GPUs. It is now selling broader systems including CPUs, networking, interconnects, storage acceleration, and rack-level architecture.

3. New AI Clouds, Sovereign AI, Industrial and Enterprise Customers

The third growth engine includes new AI cloud providers, sovereign AI projects, industrial companies, and enterprise customers.

Morgan Stanley expects this customer group to grow faster than traditional hyperscalers in certain periods, as these buyers tend to prefer purchasing complete, integrated infrastructure rather than building each layer themselves.

Their purchasing decisions are influenced by:

• Limited power and data center space
• Need for faster deployment
• Data sovereignty requirements
• National security and industrial policy considerations
• Preference for validated full-stack systems
• Limited experience operating hyperscale AI clusters internally

Sovereign AI is particularly important. Governments and local industries are building domestic compute capacity, national datasets, and country-specific models to maintain control over sensitive information and strategic technologies.

These projects are often less susceptible to competition from hyperscale-designed ASICs, as the buyers do not necessarily have in-house chip programs.

CPU and Networking Expand NVIDIA’s Market

NVIDIA no longer positions itself as a standalone GPU vendor.

The company is building an AI infrastructure platform spanning:

• GPUs
• CPUs
• NVLink scaling interconnect
• Ethernet and InfiniBand networking
• Data processing units
• Storage acceleration
• Rack architecture
• System software

This broader platform increases the revenue NVIDIA captures from each AI factory.

Vera CPU Could Become a Significant Business

The roadshow reportedly reaffirmed NVIDIA’s CPU business target of approximately $20 billion in the current fiscal year.

Morgan Stanley indicates nearly half of this revenue could come from standalone CPU racks, not just CPUs used as host processors within GPU systems.

This represents a significant expansion of the product’s role.

The Vera CPU is designed for control-intensive and latency-sensitive work around AI models, including:

• Code execution
• Tool use
• Sandbox isolation
• Analytics
• Data pipelines
• Reinforcement learning environments
• Workflow orchestration

NVIDIA’s official specifications describe Vera with 88 custom Olympus cores supporting 176 threads. Its architecture prioritizes high single-thread performance and memory efficiency to handle agentic and AI factory workloads.

If Vera succeeds both as a host CPU and as an independent data center processor, NVIDIA could capture a larger share of the server market.

As Cluster Scales Grow, Network Value Increases

Networking is another major source of growth.

As AI clusters expand, data transfer between GPUs becomes a critical bottleneck. Expensive accelerators cannot operate efficiently if they spend too much time waiting for data or synchronization.

NVIDIA is addressing this through the following technologies:

• NVLink for expanded connectivity
• Spectrum-X and Spectrum-6 Ethernet
• Quantum InfiniBand
• ConnectX SuperNICs
• BlueField data processing units
• Optical networking technology

The company's first-quarter fiscal 2027 results show rapid growth in this area. NVIDIA's data center networking revenue reached $14.8 billion, up 199% year-over-year and 35% quarter-over-quarter.

This supports the core thesis of the roadshow: NVIDIA's growth is no longer limited to selling more GPU chips but increasingly depends on selling complete systems.

NVIDIA Begins Attracting Value Investors

Morgan Stanley also noted a shift in NVIDIA's investor communication approach.

The company has historically been heavily held by growth funds. For some institutions, NVIDIA has already reached or is near internal holding limits, making it difficult for these investors to increase exposure.

The next wave of buyers may include value investors who focus more on metrics such as:

• Free cash flow
• Share buybacks
• Dividends
• Capital returns
• Earnings sustainability
• Long-term market structure

Moore expects that NVIDIA could eventually allocate over half of its cash flow to buybacks and shareholder returns.

NVIDIA's recent capital allocation decisions lend credibility to this argument. In May 2026, the company announced an additional $80 billion in share repurchase authorization and raised its quarterly cash dividend.

The value investment thesis does not replace the growth thesis. Instead, NVIDIA is attempting to present both aspects simultaneously:

1. Revenue and earnings are still expanding rapidly.
2. The cash flow generated by the company is sufficient to return more capital to shareholders.

This combination could broaden the stock's investor base.

Quarterly Revenue Is Approaching $100 Billion

The notion of "approaching $100 billion" is no longer theoretical.

NVIDIA reported first-quarter fiscal 2027 revenue of $81.6 billion, up 20% from the previous quarter and 85% year-over-year. The company's official outlook for the next quarter is $91 billion, plus or minus 2%.

Morgan Stanley's roadshow report states that management continues to position the business as accelerating growth at this scale.

This is a key point behind the investment logic. NVIDIA is not just adding billions of dollars in quarterly revenue; its expansion rate remains high as multiple markets grow simultaneously:

• Hyperscale AI factories
• Frontier model labs
• Production inference
• New GPU clouds
• Sovereign AI
• Enterprise and industrial AI
• CPU
• Networking
• Storage and system infrastructure

The challenge lies in converting demand into deliverable systems. Revenue depends on the availability of memory, networking components, power, cooling, data center space, and complete racks.

Strong Growth Expectations, but Valuation and Supply Still Matter

Morgan Stanley maintained an "overweight" rating and a $288 price target in the roadshow report.

The original article used NVIDIA's closing price of $202.78 on July 9, implying approximately 42% upside relative to that target. It also cited a market cap of around $4.97 trillion at the time.

These market data points are time-sensitive and should be viewed as a snapshot from the original publication date.

Moore's projected revenue growth is reportedly:

Period	Morgan Stanley Growth Expectation
Fiscal 2026	82%
Fiscal 2027	52.4%

These are analyst estimates, not NVIDIA guidance.

The investment thesis is based on several assumptions:

• Generative AI continues to drive cloud capital expenditures.
• Blackwell remains the primary platform for current workloads.
• Vera Rubin and Rubin Ultra extend NVIDIA's performance leadership.
• NVIDIA maintains strong share despite custom ASIC adoption.
• CPU and networking revenue continue to expand.
• Supply chain capacity keeps pace with demand without excessive inventory.

Key Risks Have Not Disappeared

Morgan Stanley also highlighted risks that could weaken the outlook.

These include:

• Supply catching up to demand faster than expected
• A sharp decline in AI development costs
• More competitive products from rival chipmakers
• Faster deployment of customer-customized hardware
• Constraints in power, memory, networking, and construction
• Valuation pressure if growth slows

A particularly important risk is the transition from scarcity to balance.

When demand far exceeds supply, manufacturers have strong pricing power and customers compete for capacity. If supply expands too quickly, data center growth could slow more sharply than investors expect.

The custom chip risk also remains real. Google, Amazon, Microsoft, Meta, and other large buyers have strong incentives to develop hardware optimized for their own workloads.

NVIDIA's defense is not just faster GPUs.

It lies in its software ecosystem, networking, CPU, rack design, deployment speed, and overall total cost per token economics.

The Real Constraint Is Delivering Complete AI Systems

The broader message conveyed in the roadshow is that demand for AI infrastructure remains strong.

NVIDIA's current problem is not convincing customers that AI computing is useful, but rather converting enormous demand into operable systems under multiple physical constraints.

Current limiting factors include:

• High-bandwidth memory
• Network capacity
• Power generation and distribution
• Cooling infrastructure
• Data center construction
• Optical components
• Rack integration
• Customer deployment readiness

This is why NVIDIA continues to invest in overall architecture.

A GPU must be packaged into a working system, connected to memory and networking, supplied with adequate power, and installed in a ready data center to generate revenue.

Rubin Ultra's 800 VDC and Kyber architecture reflect this shift. The next phase of AI computing is not just a semiconductor problem but an infrastructure engineering problem.

Frequently Asked Questions

Has NVIDIA's Rubin Ultra been delayed?

According to Morgan Stanley's roadshow report, NVIDIA denied claims that Rubin Ultra has been delayed to 2028 and maintained a 2027 shipment timeline. NVIDIA's public infrastructure roadmap also describes the Kyber system featuring Rubin Ultra GPUs as arriving in 2027.

What is the difference between Vera Rubin and Rubin Ultra?

Vera Rubin is NVIDIA's current next-generation AI platform, combining Rubin GPU, Vera CPU, NVLink, networking, and storage infrastructure. Rubin Ultra is a planned higher-density follow-up product built on the Kyber rack architecture and 800 VDC power system.

Why is NVIDIA's quarterly revenue approaching $100 billion?

NVIDIA reported first-quarter fiscal 2027 revenue of $81.6 billion and expects next-quarter revenue of $91 billion. Growth comes from hyperscale cloud providers, AI labs, new cloud service providers, sovereign AI projects, networking, CPUs, and complete AI factory systems.

Will custom ASICs replace NVIDIA GPUs?

Custom ASICs are expanding, but it is not a simple replacement. Customers may use their own accelerators for specific workloads while continuing to deploy NVIDIA systems, particularly where software maturity, flexibility, networking, or total cost per token offer advantages.

Which customer raised NVIDIA's compute share to nearly 50%?

Morgan Stanley did not name the customer in the roadshow notes. The original article speculated it could be Anthropic, but this judgment is based solely on guesswork and has not been confirmed by the companies involved.

What is sovereign AI?

Sovereign AI refers to a country's ability to build and operate AI using its own infrastructure, data, talent, and domestic ecosystem. Such projects may generate demand for local AI factories, national models, and secure computing power.

Why is NVIDIA expanding into CPU and networking?

Large AI clusters require more than just accelerators. CPUs manage data pipelines, tools, orchestration, and agent workloads, while high-speed networking enables thousands of GPUs to stay synchronized and fully utilized.

Is this article investment advice?

No. The analyst ratings, price targets, and forecasts discussed in the text are used to illustrate the original report.

Markets involve risk. Readers should assess their own objectives, financial situation, and risk tolerance before making investment decisions.

Related Tools

• NVIDIA Vera Rubin Platform: NVIDIA's full-stack platform designed for large-scale training, inference, agentic AI, and scientific computing.
• NVIDIA Vera CPU: A data center CPU purpose-built for agentic workloads, reinforcement learning, orchestration, and AI factory operations.
• NVIDIA Spectrum-X: An Ethernet networking platform for scaling distributed AI infrastructure.
• NVIDIA NVLink: NVIDIA's high-bandwidth interconnect technology for scaling GPU and CPU connectivity.
• NVIDIA BlueField: Data processing and storage infrastructure for networking, security, isolation, and data transfer.

Related Links

• NVIDIA Financial Results for First Quarter Fiscal 2027: Official revenue, data center, networking, outlook, and capital return data.
• NVIDIA Introduces Rubin Platform: Official overview of Rubin GPU, Vera CPU, NVLink 6, BlueField-4, and Spectrum-6 architecture.
• NVIDIA Vera Rubin Platform Overview: Official product information on Vera Rubin systems and related infrastructure.
• NVIDIA Kyber and 800 VDC Infrastructure: NVIDIA's explanation of Kyber racks, 800 VDC power, and the 2027 Rubin Ultra roadmap.
• NVIDIA 800 VDC Technical Architecture: Technical details on power delivery for next-generation high-density AI factories.
• NVIDIA Financial Reports: Official archive of NVIDIA's financial results and investor disclosures.
• NVIDIA Sovereign AI Overview: NVIDIA's definition of sovereign AI and national AI infrastructure.

Summary

Morgan Stanley's NVIDIA roadshow recap indicates that the company's growth is diversifying rather than converging. AI labs, hyperscalers, emerging AI clouds, sovereign AI projects, CPU, and networking businesses are collectively driving the next phase of expansion.

Rumors of a delay for Rubin Ultra have been denied, supporting NVIDIA's existing 2027 roadmap. Meanwhile, the shift toward larger racks, 800 VDC power, optical interconnects, and complete AI systems suggests that future growth relies on infrastructure engineering as much as chip design.

Custom ASICs remain a significant competitive risk, but the roadshow indicates their adoption is not a simple replacement cycle. Customers are evaluating the economics of entire workloads, including software, networking, utilization, and per-token costs.

NVIDIA's core challenge is no longer proving demand for AI exists, but translating that demand into complete systems that can be powered, cooled, interconnected, delivered, and operated at scale.