An Oversupply Of Compute Power Will See Margins Recede

Why NVIDIA’s Growth Curve May Flatten Out

• NVDA continues to outperform while QoQ growth levels off.
• Future company growth rates may not be as drastic due to the already high revenue base of $35B.
• The next catalysts to watch are capex guidance from customers, occurring mid Jan to early Feb.
• Maintaining AI customer demand and a value proposition of increased profitability may become more difficult.
• Management is known to be “2 steps” ahead and announcing a new chip may sustain enthusiasm.

Nvidia posted Q3'25 revenues at $35B up 94% YoY. Sequential growth remained elevated and increased to 16.8%. The last 4 quarters resulted in revenues of $113.3B, up 152% from the prior period. This is well on its way to reaching my 2028 estimate for $165B (derived as a 55% share of my estimated 300B SAM for NVDA). 

NVDA: Quarterly revenue performance. QoQ & YoY emphasis from author

The company guided (p. 12) for a QoQ growth of around 7% to $37.5B, implying some 70% growth YoY.

Assuming a sustained QoQ growth rate of around 10%, NVDA will be able to reach my $165B target for 2028 in about 1 year, indicating that NVDA is outperforming my expectations. However, despite investor enthusiasm, I think that extrapolating NVDA's QoQ growth from a $35B quarterly base may not be grounded with the data center capex renewal cycles from NVDA's largest customers. For example, in their last earnings call, TSLA has pointed out that they will be more restrained on their future capex spend (1, 2), including on GPUs.

What I’m Factoring Into NVDA’s Growth

Most of NVDA’s revenue comes from cloud providers, which already have dedicated a substantial portion of their capex to updating their GPU infrastructure, and this is why I expect them to be more conservative with spending going forward unless they see a high improvement in the value proposition.

Generally, the value proposition from NVDA has 2 aspects.

1. Profit via cost savings. This is done by improving performance and is likely to be even better with NVDA’s Grace-Blackwell GPUs. This is especially true if someone is upgrading pre-H100 architecture to Blackwell, and while incrementally improving upon H100 or H200 infrastructure, its performance benefits likely still create cost savings.
2. The other side of the equation is AI or general compute demand. My sense is that NVDA has a fall-back pitch to companies using their GPUs along the lines of: if AI demand goes down, you can still use the infrastructure for general compute. This is true, and compute inflation seems to be creating demand. However over-provisioning on cutting edge infrastructure may lead to scenarios where the customers find themselves paying a premium for the hardware, and experiencing a lower demand, leading to pricing cuts on cloud infrastructure. 

My current thinking is that some of these cloud providers may start postponing infrastructure upgrades as the benefits may not be as steep as the first introduction of new gen GPUs.

Consumer demand may also change in 2 ways:

• General stagnation in AI applications. I’m not implying that interest falters, but that enough of the world is already using AI and are content with the current service offering.
• Market consolidation to one provider i.e. if ChatGPT becomes the clear winner in consumer AI, then rivals like Amazon and Google will have a more difficult time pitching continued increases in GPU spend.

In contrast to these scenarios, we will take a look at how NVDA is working to maintain and grow its GPU demand.

In order to sustain growth, NVDA's management is looking to expand sales to governments with sovereign AI. They have noted that Japan, India, and Indonesia are building AI infrastructure, and Denmark has launched its largest AI supercomputer built with 1,528 NVIDIA H100 GPUs. I expect this direction to continue and even more countries to adopt the sovereign AI pitch.

Additionally, NVDA has been expanding its AI and GPU use-cases to sectors that can benefit from, but whose main industry is not technology. These sectors include healthcare, robotics, telecom, automotive, etc.

To summarize, a viable portion of the cloud infrastructure now enables AI services, and NVDA needs to show ways that customers can benefit while expanding the number of large customers (such as governments) it will be selling its infrastructure to.

What Are Some Possible Catalysts On The Horizon

Since the beginning of my narrative, I speculated a cycling-down of demand for AI. However, NVDA’s exceptional performance and general value proposition to cloud service providers made this scenario difficult to time, and it seems that management has executed on selling AI as a global phenomenon. 

Things are looking like they are about to start changing as compute capacity grows and customers are looking to develop their own solutions in order to avoid NVDA’s large premiums. 

I still expect that NVDA will overperform its sales estimates and that quarterly revenue growth will be around 10% instead of the 7% guide. However, the potential news of a “high single-digit” growth may be a catalyst that shakes things up in the market, and investors may change their approach on the stock. 

This will likely play out in the following 2-3 quarters, but its beginnings will be seen on the quarterly results commentary from NVDA’s key customers such as MSFT, META, AMZN, GOOGL, as well as downstream from TSM, etc., all of which are typically reporting their next results (mid Jan to early Feb) sooner than NVDA. This is why I expect any repricing to start occurring early in the next earnings season. 

This is only my scenario - ultimately, there are no ways to accurately predict markets and NVDA may find growth avenues that lead to continued outperformance. One way management could retain investor enthusiasm would be to announce the development of a new chip, even if it's in its early stages.

Valuation

I am extending my forward estimates to 2029, with a revenue growth of 14.4%, resulting in the increase from $165B in 2028 (previous) to $189B in 2029. I am retaining my expected net margin to 35%, and therefore estimate a net income of $66B.

Using a 35x PE for 2029, I’m upgrading my future value to $2.3T from $2T. 

Discounted back to today, using a 7.5% rate, I get a present value of 1.6T or $68 per share.

In my view, NVIDIA continues to be an exceptional company, trading at an asymmetrically high premium.