Nvidia: Ignore The Noise And Buy The Dip
Summary
Nvidia's growth outlook is strong, driven by the transition to inferencing, leading to the need for significant increases to compute capacity across the cloud services providers.
Despite market concerns over Microsoft's data center lease cancellations, other hyperscalers have not followed suit, suggesting that this news may not have a material impact on growth in compute capacity.
Nvidia began ramping up volume shipments of Blackwell GPUs, contributing $11b in sales in q4'24.
I'm raising my rating to a STRONG BUY given the share price pullback and growth potential as AI expands across the digital to physical world.
Oversized AI symbol surrounded by people
J Studios
NVIDIA (NASDAQ:NVDA) (TSX:NVDA:CA) may have a larger growth outlay than initially anticipated as AI transforms from a proof-of-concept to production AI applications. With the growing demand across enterprises as well as the driving factors for Industry 4.0, I believe compute capacity will need to expand significantly over the next few years to support the growing industry needs. With no indication of slowing capital investments for compute capacity and data centers, net of the recent news from Microsoft (MSFT), I suspect investments may remain durable for CY25, driving growth for data center infrastructure. I believe the recent selling pressure for NVDA shares has created a significant opportunity for investors to build their positions. Given these factors, I am upgrading my rating to a STRONG BUY with a price target of $235/share at 20.13x eFY27 price/sales.
Nvidia Operations
Corporate Reports
Corporate Reports
Jensen Huang recently appeared at the Nvidia GTC 2025 event with some appealing insights for the future of AI infrastructure. One driving point is that as AI models transition from training to inferencing and agentic AI with “reasoning” functionality, compute capacity will need to significantly expand from where it sits today. Mr. Huang suggested that it will need to be a hundred times more than what was expected last year. Because the “reasoning” function may take 10x the tokens when compared to a one-off answer, the “reasoning” model may require 10x the compute capacity to ensure a timely result.
This may mean shorter product cadences going forward. Mr. Huang announced the latest addition to the software ecosystem, cuDSS, to help accelerate the development of accelerators. cuDSS is a library of GPU-accelerated linear solvers with sparse matrices that can be used for real-time applications like autonomous driving and process simulation.
Nvidia Blackwell turned to full production as Hopper peaked in sales, bringing in the next cycle of GPUs. Accordingly, Blackwell delivered $11b in revenue in q4’25. Clusters of 100,000+ GPUs are being shipped to major infrastructure customers, allowing for post-training and mode customization as enterprises turn to production AI applications. CSPs accounted for roughly half of Nvidia’s data center revenue in q4’25.
Blackwell is said to provide 40x the performance when compared to Hopper when using a reasoning model. Accordingly, the Grace Blackwell NVLink 72 rack has the computation capacity of 570 terabytes per second. Driving efficiency at the data center, Mr. Huang announced the next-generation AI framework, Nvidia Dynamo. Nvidia Dynamo was designed for deploying GenAI and reasoning models in a large-scale distributed environment. Accordingly, this framework boosts inference performance on the GB200 NVL72 by 30x.
Following Blackwell will be the release of Vera Rubin in 2h26. Rubin will come with a new networking SmartNIC, a new NVLink, and HBM 4 (high bandwidth memory). Rubin will scale up to 15 exaflops and run 4.6 petabytes per second.
Mr. Huang also discussed the emergence of AI in robotics. As part of this, Teradyne (TER) had previously announced that it had partnered with Nvidia to bring AI into its industrial robotics. This could be a major shift in how robotics is viewed in an industrial setting from a pre-programmed machine to a multi-functional, adaptive machine. To help train industrial robots, Nvidia revealed the Mega Omniverse Blueprint for developing, testing, and optimizing physical AI and robot fleets. As part of this, Nvidia introduced Nvidia Isaac Groot N1, a foundational model for generalized skills and reasoning for humanoid robots.
Mr. Huang addressed quantum computing in the GTC presentation, suggesting that Nvidia is assisting with quantum architecture research and algorithms.
In the GTC presentation, Mr. Huang announced Nvidia’s new partnership with General Motors (GM) in developing its AI ecosystem to enable self-driving vehicles. This will include multiple phases in GM’s environment, including bringing Industry 4.0 to GM. This means that GM may bring more autonomous operations into its manufacturing process with robotics. Nvidia will also bring AI into the enterprise to accelerate the firm’s operations, including vehicle design and enabling AI in the vehicles.
In q4’24, Nvidia realized 103% year-over-year growth in its automotive segment, driven by the ramp-up in autonomous vehicles and robotaxis. In addition to GM, Nvidia was qualified for Toyota’s (TM) next-generation vehicles on Nvidia Orin.
Bloomberg recently reported some updates to the OpenAI Stargate project with the first AI factory housing as many as 400,000 Nvidia GB200 GPUs, integrated with Spectrum-X. The facility will have the capacity of 1.2GW of power. The facility is expected to be completed in mid-2026. Accordingly, the total investment outlay for the project is upwards of $500b, potentially allowing for an expansion of upwards of 10 additional facilities.
Nvidia Financial Position
Corporate Reports
Corporate Reports
Looking ahead to eq1’26, I’m forecasting Nvidia to generate $44.4b in net revenue with an adjusted EPS of $0.95/share. Management guided a 30% year-over-year increase in adjusted operating expenses for eFY25, suggesting that adjusted operating margins may remain relatively stable when compared to FY24 figures. I’m forecasting the eFY25 operating margin to come in at 66.58%.
Risks Related to Nvidia
Bull Case
The investment outlay continues to expand as Blackwell shipment volumes reach full production. With the high-capacity demand across the CSPs as they develop their competitive moat for AI inferencing, I suspect investments will remain elevated. Though a small component to total revenue, Nvidia is realizing substantial growth in its automotive market segment as OEMs turn to Nvidia for autonomous vehicles, and in Tesla’s case, Robotaxis.
Despite the recent news that Microsoft canceled some data center leases, I remain adamant about my $313b capital investment outlay forecast across the major hyperscalers.
Bear Case
Despite my bullish outlook, the market has reacted negatively to the news of the release of the DeepSeek-V3 models as well as Microsoft’s lease cancellations. Given that CYq1’25 is nearing its end, I believe more information relating to capital investments for compute capacity will be revealed for the duration of CY25. If Microsoft, as well as other hyperscalers, were to pull back on their respective capital investments for compute capacity and data centers, NVDA shares may face significant selling pressure as data center growth may slow.
Valuation & Shareholder Value
Corporate Reports
Corporate Reports
NVDA shares are currently trading at 22.44x price/sales on a trailing twelve-month basis. Though shares trade at a significant premium over its peers in the GPU and XPU market, I believe NVDA shares’ high valuation is justified by its exceptionally high top-line growth rate.
Seeking Alpha
Seeking Alpha
The question is how the market will value NVDA shares going forward; whether shares will return to its high valuation or if shares will normalize at a lower premium, potentially around its current valuation of 22.44x price/sales.
At the high end, shares could potentially grow to $296/share at 25.41x eFY27 price/sales; however, I believe that the market will be seeking a more moderate premium given the macroeconomic uncertainty. Given these factors, I believe NVDA shares should be priced at the low-end in my model at 20.13x eFY27 price/sales at $235/share. Given the growth potential and the price dispersion brought forth by the recent market sell-off, I am upgrading my rating to a STRONG BUY for NVDA shares with a price target of $235/share. .
Corporate Reports
Corporate Reports
Using the model: the valuation table above references my financial forecast in the firm’s “financial position” section and ties it to the stock’s historical trading premiums. The trading premium array is derived through the normal operating cycle, with the blue-sky scenario being the stock’s peak multiple and the gray-sky scenario being the lowest point. The target multiple aims for the midpoint, or the most likely trading range for the company’s stock. The trading multiples from there are set to a probability factor based on the likelihood of the stock trading at that premium based on its historical presence. From there, the trading multiple is tied to the probability factor to derive its relative market cap and relative multiple.
#nvda #ai $SPDR S&P 500 ETF Trust(SPY)$ $Nasdaq 100 Trust(QQQ)$ $C3.ai, Inc.(AI)$
Comments