Nvidia Q3 Review: Cannibalizing Their Own Growth
Contributor since 2023 / 3483 Followers
Summary:
- NVIDIA’s Q3 earnings beat expectations, but I remain bearish due to concerns about the long-term impact of Blackwell chips’ efficiency on revenue growth.
- Blackwell GPUs are 4 times more efficient than H100s, potentially reducing the number of units needed and shrinking NVIDIA’s market size.
- Despite strong demand for Blackwell, the efficiency gains may not offset the reduced volume, challenging NVIDIA’s ability to sustain high revenue growth.
- I still believe NVIDIA’s current valuation is too high; I believe shares should trade closer to the sector median at a 35x forward P/E, implying a potential 30.01% downside.
JHVEPhoto
Co-Authored By Noah Cox and Brock Heilig.
Investment Thesis
NVIDIA (NASDAQ:NVDA) (NEOE:NVDA:CA) shares are down roughly 3.1% since earnings were released on Wednesday after the bell, despite posting results that beat on the top and bottom line.
Wednesday’s Q3 earnings from the Silicon Valley AI chip giant still showed strong year-over-year growth, but that growth has clearly been slowing from earlier in the year. I am not bearish on Nvidia for slowing growth (their scale has now become immense, with TTM revenue of $113.27 billion).
At this scale (revenue over $113 billion) Nvidia’s biggest issue is not necessarily slowing growth, but rather how the company plans to push the bar higher in FY 2026. I’m still bearish on the company because most of Nvidia’s revenue is not recurring, its hardware sales (recurring software and support sales were at $1.5 billion on an annualized basis at the end of the quarter).
This is only slightly above 1% of the trailing 12-month revenue of $113.269 billion.
These sales have no promise of recurring next fiscal year. With this (and ironically), Nvidia’s new Blackwell chips will likely cannibalize a lot of potential H100 & H200 sales when they ramp up in production.
The cannibalization of older GPU sales is not what concerns me. What concerns me is that these new Blackwell chips are efficient and reduce the raw number of GPUs needed in order to effectively train, host, and do inferences with LLMs.
Demand for the Blackwell chips appears to be very strong among early hyperscaler customers, but I believe as we work into more use-cases for these new GPUs this data shows that customers who have less-intense computing needs won’t need as many GPUs. Blackwell appears to be roughly 4 times more efficient than H100s. The chips are not 4 times more expensive to offset Nvidia’s future revenue.
To be clear, I am a fan of Nvidia’s products, but at this point I am concerned these new chips are not going to help the company take revenue to the next level.
With this, I believe shares remain a strong sell.
Why I’m Doing Follow-Up Coverage
To anyone who is just reading my research for the first time, you’ll quickly see I have been consistently bearish on Nvidia by looking at my previous coverage updates on the chip giant over the last 12 months. While I have been wrong before (and I will fully admit that) I continue to have an issue with the GPU giant’s business model.
Driving my continued pessimism over the last 6 months in particular (and now in the light of albeit strong earnings results) has been my constant skepticism over the new Blackwell product. I have been previously concerned about supply (some of this ended up being warranted concerns). Earlier this week (as part of my earnings preview), I wrote on how the new Blackwell chips may require a special mount in server racks due to their risk of overheating.
With the new efficiency disclosure from the earnings call, I don’t believe this new Blackwell product will be the end all be all growth driver for the stock like many Nvidia bulls hope it will be.
The main reason I am doing follow-up coverage for Nvidia is that even though the company reported strong numbers in Q3, my concerns have long been geared towards looking into the future where Blackwell is the main driver.
Many of the things I saw in the recent earnings report confirm my concerns.
Q3 Review
Looking at actual FY 2025 Q3 earnings, Nvidia had a strong earnings report. The company posted an EPS of $0.81, which beat the analyst projection of $0.75 by $0.06. Adding to the EPS beat, the chip giant’s revenue numbers were also impressive. Nvidia hauled in $35.08 billion in revenue, which beat projections by a strong $1.95 billion. This revenue mark was 93.61% year-over-year growth for the company.
Colette Kress, Nvidia’s Executive Vice President and Chief Financial Officer, discussed the record Q3 on Wednesday’s earnings call.
Q3 was another record quarter, Kress said.
We continued to deliver incredible growth. Revenue of $35.1 billion was up 17% sequentially and up 94% year-on-year and, well above our outlook of $32.5 billion. All market platforms posted strong sequential and year-over-year growth, fueled by the adoption of NVIDIA accelerated computing and AI.
Kress also gave her thoughts on Blackwell and its technical benchmark performance. I think this was key.
Just 64 Blackwell GPUs are required to run the GPT-3 benchmark compared to 256 H100s or a 4 times reduction in cost, Kress said on the call.
NVIDIA Blackwell architecture with NVLINK Switch enables up to 30 times faster inference performance and a new level of inference scaling throughput and response time that is excellent for running new reasoning inference applications like OpenAI’s o1 model.
From the guidance front, Nvidia expects to generate roughly $37.5 billion in revenue in Q4 of FY 2025. This beat analyst expectations of $37.1 billion.
While the quarter was bullish on the surface with EPS and revenue numbers outpacing projections and management’s optimism surrounding Blackwell still being high, the big reason why I am still bearish on it is because of the devil in the details. New GPUs that are more efficient are good for customers. That doesn’t mean it’s good for revenue.
What I Saw That Concerned Me
The earlier quote from Kress about how 64 Blackwell GPUs can do the work of 256 H100s for an adjusted cost of four times less is what really concerns me. I think this is potentially shrinking the size of Nvidia’s market.
While Kress went on to say later in the earnings call that demand for Blackwell is really high, and Nvidia is struggling to keep up, this doesn’t mean they will match the same volume as the H100s.
Blackwell demand is staggering and we are racing to scale supply to meet the incredible demand customers are placing on us, Kress said.
Customers are gearing up to deploy Blackwell at scale. Oracle announced the world’s first ZettaScale AI Cloud computing clusters that can scale to over 131,000 Blackwell GPUs to help enterprises train and deploy some of the most demanding next-generation AI models.
While it’s potentially a good thing that Blackwell chips are more efficient than the former H100 and H200 chips, I think it’s obvious that this will lower the number of Blackwell chips that Nvidia customers need to do the same level of model development.
Kress noted that the demand for Blackwell chips is currently very high, but with Blackwell chips being roughly the same price as the current H100 series GPUs, the strong demand for Blackwell chips may not outweigh the efficiency difference. If the new Blackwell chips are four times more efficient than the current chips like Kress said, this will theoretically reduce the number of chips needed by customers by a factor of 4x.
While AI models are getting larger (and therefore need more compute) what we are essentially arguing here is that global demand for GPU compute will have to roughly 4x over the next 4 quarters in order for these 4x more efficient GPUs to sell at the same volume as the H100 series.
And that’s just to get to FY 2025 sales revenues. Keep in mind that revenue is supposed to grow 49.1% YoY in FY 2026. This would mean global GPU compute needs would need to be up to 6x higher than where they are today.
I just see that as highly unlikely.
Valuation
Compared to the sector median, as I mentioned in my previous coverage, Nvidia shares trade at quite a premium. The forward non-GAAP P/E ratio for Nvidia is currently at 50.01, which is a 100.73% premium to the sector median of 24.92. Using Seeking Alpha’s quant metrics system, Nvidia’s valuation grade comes out to a D. The Seeking Alpha quant system overall rates the stock as a hold. I am more bearish than this.
A big reason that shares trade at such a premium to their sector median is because the market continues to expect this strong forward growth from the company over the next 12 months (49.1% YoY revenue growth).
As I mentioned before, I’m personally concerned about Nvidia’s Blackwell sales cannibalizing the total opportunities they have because of the fact that their new GPUs are so much more efficient. As I mentioned before, global demand for GPU compute needs to multiply in order to support the volume of Blackwell chips Wall Street is estimating.
While this may be controversial, I think Nvidia should be trading closer to the sector median at 35 times forward earnings on a non-GAAP basis. I think this valuation makes far more sense than where the P/E is at today.
If we saw shares converge at a forward price-to-earnings ratio of 35 instead of 50.01 like it currently is, this would represent 30.01% downside for shares.
Bull Thesis
By all means, Nvidia’s Blackwell chip is a powerful new offering. I want to make sure I am clear that I understand that this new chip has a lot of potential for Nvidia’s customers. A big bull thesis here could be that Nvidia is able to find even more opportunities for customers to deploy Blackwell, simply because the chip is so efficient. On top of this, another key bull thesis is the opportunity for the GPU maker to dramatically increase their software sales.
I’ll talk first on the software sales because while I think there is opportunity for growth here, (the $1.5 billion in annualized software revenue will ramp to $2 billion by the end of the fiscal year), this is not enough to move the needle.
Even if new software revenue doubled to $4 billion next fiscal year, this would still represent less than 4% of overall trailing 12 month revenue. It’s just simply not enough.
On the GPU side, I think there are huge opportunities still left out there for GPUs to add incredible value to hyperscalers, so data centers become far more efficient. But this doesn’t mean that there will be enough opportunities. Nvidia is still struggling with intense revenue concentration, meaning they really need to diversify their customer base. At the end of Q3, 3 customers have represented a combined 36% of revenue (while a fourth one represents less than 10% of revenue).
Even the largest big tech companies in the US can become GPU-fatigued.
To this point (I talked about this in my pre-earnings coverage) OpenAI’s CEO, Sam Altman, said a few weeks ago that AGI can be achieved on ‘current’ hardware. At the time, this was the H100 & H200 series (so I assume this was the current hardware). Given this, the Blackwell GPUs can definitely support AGI by this measure.
But, by further extension, Blackwell GPUs are 4 times more efficient than H100 hardware. So to achieve AGI, we would need 1/4th as many Blackwell GPUs as we would need H100s.
In essence, I don’t see how that (again) leads to more Blackwell revenue next fiscal year compared to H100/H200 revenue this year.
Takeaway
To be clear, Nvidia sells a great product — one that I’m a fan of — but I just don’t think this will be enough to justify their current valuation.
Q3 was an objectively strong quarter for the GPU maker, with year-over-year revenue reaching nearly 100% growth. In the long run, I struggle to see how Nvidia will see such strong growth because of how efficient Blackwell is.
Although Nvidia is coming out with a much more efficient chip, I ironically think this will cannibalize some sales.
Diving deeper, we can see that the current capabilities of the H100/H200 series chips will likely be able to give us AGI (according to the CEO of the company that is likely closest to it). Blackwell can do the same too, but with 4x less units (since its 4x more efficient). I continue to be bearish on the outlook. With this, I think shares continue to be a strong sell.
Analyst’s Disclosure: I/we have no stock, option or similar derivative position in any of the companies mentioned, and no plans to initiate any such positions within the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.
Noah Cox (main account author) is the managing partner of Noah’s Arc Capital Management. His views in this article are not necessarily reflective of the firms. Nothing contained in this note is intended as investment advice. It is solely for informational purposes. Invest at your own risk.
Seeking Alpha’s Disclosure: Past performance is no guarantee of future results. No recommendation or advice is being given as to whether any investment is suitable for a particular investor. Any views or opinions expressed above may not reflect those of Seeking Alpha as a whole. Seeking Alpha is not a licensed securities dealer, broker or US investment adviser or investment bank. Our analysts are third party authors that include both professional investors and individual investors who may not be licensed or certified by any institute or regulatory body.