I have to admit that I share your scepticism about being able to model NVIDIA. There are simply too many factors that can skew the models.
Demand shift from Cloud Providers, competition, TSMC blow up risk (or even temporary blocks), recession leading to a general slowdown for a couple of years on many levels.
All in all, I simply won't touch this without a huge margin of safety.
Just found your substack and spent an and hour or two enjoyably ingesting this 4-part series on Nvidia. A really fine and thorough effort. You should be commended for the research and the obvious attention to detail you put into it. Really magnificent.
I am a very long term investor, since the early 2000s. By way of critique, I have only one on your perspective in investing in this company and that's of traditional valuations. The shortcoming of DCF models are pretty clear in that your projected 5 year $175 price target has basically been achieved in 6 weeks after you published it. I am not saying traditional valuations are bad and that they shouldn't be used. In fact, they should for most companies. All I am saying is they do have limitations, for example, if someone bought the day this article was published, and put in a limit sell just under your target, they'd be out of the position already. It's ironic that it took more time to get these 4 articles out than the 5 yr price target your model produced.
What your articles have done is elucidated a wonderful story on Nvidia. It's a story stock and has been for some time. It's a story of a founder with a vision, 3 decades in and still at the tiller. A guy driven. A bunch of near misses. A deft long term strategy and nimbleness to nurture, to pivot, and to persevere in the face of a dynamic and challenging environment. Nvidia has put it's hands around the opportunity of our lifetimes, the next technological revolution, similar to how Intel and Microsoft seized the PC marketplace in the 1990s, commonly referred to as "wintel," except Nvidia controls both the hardware and the software side of the equation. I'm not sure how that is quantified in a DCF model.
Having been invested for nearly two decades based primarily on my belief in the CEO, the one thing I know for sure is that this company's valuations will remain stretched beyond traditional valuations for a long time to come. Investors should just park what ever money they're comfortable with and not look at it for 5 or 10 years, a bottom drawer stock, forget about it. Jensen is incredibly smart and ruthless, and he has created incredibly high barriers for competitors.
Hey EC, thank you for reading and this thoughtful comment. I really appreciate this feedback. I wanted to take my time to respond.
Firstly I completely agree with you on the flaws in a traditional IRR/DCF approach. No valuation approach is perfect and that's why i am trying not to put too much weight on these valuations in my articles as it's highly subjective, especially for growth companies like Nvidia. I certainly hope people are not treating these articles as stock a stock picking service because really the thing that I am most focused and interested in is the business analysis. Thats why I spent three articles discussing the business in depth and a much shorter one on the valuation - that really isn't the focus for me. Actually one of they main reasons I even do a valuation model is that it helps me think through the business and understand the drivers a lot better.
Re valuation, I could be completely off in so many assumptions but the biggest one as you pointed out is what does the future multiple need to be. That is usually by far the most sensitive assumption. This is the key thing I tried to highlight, that if you believe the stock can continue to trade on 40-50x+ PE/FCF, then you'd make that investment all day. If you believe it should revert to a more normal market multiple as most stocks tend to do over time, then the results will be a lot lower. I've had bulls tell me 30x is too low and bears say it's ridiculously high, so there's a huge divergence in opinion. I've got no idea what the stock should will trade at it in the future, but this valuation exercise helps you understand what you need to believe in to earn a certain return. I expect readers will make up their own views.
In any case i completely agree with your points around how exceptional Nvidia is - I am long the stock after all and plan to hold it for a long time. I would hope that whatever numbers I put down are conservative, and if they can unlock other growth options like Auto, Robotics, Omniverse etc, just like they unlocked Data Center 4-5 years ago, they could easily continue to maintain these current high multiples.
Anyway thank you once again for reading the articles and taking the time to write this feedback. Having these sorts of exchanges is exactly the reason for why I put my work out there.
I have now included the model as an attachment at the end of the article. Feel free to download and reach out if you want to chat about it - DM on Twitter would be best
Hi Punch Card, just found your Substack and I find your viewpoints on companies like SEA interesting. Currently a student in a local Singapore university, trying to break into the PE space. I’m interested in hearing more about your experience. Any chance we could connect off this platform?
40% YoY in data center for next year seems aggressive no? I have been struggling to model out the best way to solve for data center outside of using these pie in the sky TAM CAGR projections. Do you think we need to go through a digestion phase? Any other ideas on how to model data center growth more accurately?
So FY23 for Nvidia is actually a Jan-2023 year end, so most of it is CY2022 of which we are more than half way through. I had a $14.9bn DC revenue forecast for FY23, and in the first 2 quarters they've done $7.6bn, so even if the next two quarters are flat they should readily achieve this full year number I think. Beyond that, typically the approach that analysts use is to try to comp it the revenue growth, or more accurately capex growth, of the hyperscale cloud providers (AWS, Azure etc), who are a large customers of Nvidia, which is what I tried to do as well. The tricky thing is that AWS and Azure capex is not broken out separately in AMZN and MSFT's financials, so some estimation is required (or potentially just assume that capex grows in line with revenue, which is reported and forecasted by analysts).
Overall I agree with you in that it's not an easy segment to forecast, which is why for conservativism i've assumed a reasonable moderation going forward, and hope to be surprised on the upside. Hope this helps.
for the $19bn 24 number though - don't you think there will be some digestion in data center / cloud capex? Are the only customers really the large cloud hyperscalers? What growth are you expecting for in the end markets, and then I assume you have NVDA grow faster with penetration of total DC/cloud capex. The CEO's explanation this Q for enterprise vs. datacenter trends was enterprise was renewals and DC was new builds, but wouldn't new builds be an easier thing to cut going forward with macro weakness?
One more: what does NVDA net on a sale of a consumer graphics card when taking out their partners share - thinking gaming segment? (they don't make $1k per card I assume)
Great questions. These are all fair concerns, although so far all the cloud providers are still reporting continued strong investment and growth, plus the Meta factor I mentioned earlier which i think could make up for any Enterprise weakness, but whether next year's growth is 30% or 25% or 20%, hard for me to say. I know all forecasts will be wrong, so my approach is to try be directionally right long-term and see if the the returns look decent under what I think could be broadly achievable (hence then using concepts like TAM to benchmark long-term projections, but that's obviously got it's own problems). But to give you a sense of the sensitivity, if my FY28 DC revenue is off by $5bn, it shaves about 1% off the IRR.
On the consumer graphics cards, what Nvidia sells to the OEMs is the GPU, which from some numbers i saw floating around could be somewhere between $150-200. The rest of the card is actually built by the OEM, which is what takes it up to the $1K mark including retail margin
Thanks for the article. Read all 4. Enjoyed it so much. Merry Christmas!
Thank you for reading and Merry Christmas!
Thanks for the work.
I have to admit that I share your scepticism about being able to model NVIDIA. There are simply too many factors that can skew the models.
Demand shift from Cloud Providers, competition, TSMC blow up risk (or even temporary blocks), recession leading to a general slowdown for a couple of years on many levels.
All in all, I simply won't touch this without a huge margin of safety.
Thanks for reading Atul! Very difficult to forecast indeed. Will be very interesting to watch how it all plays out.
PunchCard,
Just found your substack and spent an and hour or two enjoyably ingesting this 4-part series on Nvidia. A really fine and thorough effort. You should be commended for the research and the obvious attention to detail you put into it. Really magnificent.
I am a very long term investor, since the early 2000s. By way of critique, I have only one on your perspective in investing in this company and that's of traditional valuations. The shortcoming of DCF models are pretty clear in that your projected 5 year $175 price target has basically been achieved in 6 weeks after you published it. I am not saying traditional valuations are bad and that they shouldn't be used. In fact, they should for most companies. All I am saying is they do have limitations, for example, if someone bought the day this article was published, and put in a limit sell just under your target, they'd be out of the position already. It's ironic that it took more time to get these 4 articles out than the 5 yr price target your model produced.
What your articles have done is elucidated a wonderful story on Nvidia. It's a story stock and has been for some time. It's a story of a founder with a vision, 3 decades in and still at the tiller. A guy driven. A bunch of near misses. A deft long term strategy and nimbleness to nurture, to pivot, and to persevere in the face of a dynamic and challenging environment. Nvidia has put it's hands around the opportunity of our lifetimes, the next technological revolution, similar to how Intel and Microsoft seized the PC marketplace in the 1990s, commonly referred to as "wintel," except Nvidia controls both the hardware and the software side of the equation. I'm not sure how that is quantified in a DCF model.
Having been invested for nearly two decades based primarily on my belief in the CEO, the one thing I know for sure is that this company's valuations will remain stretched beyond traditional valuations for a long time to come. Investors should just park what ever money they're comfortable with and not look at it for 5 or 10 years, a bottom drawer stock, forget about it. Jensen is incredibly smart and ruthless, and he has created incredibly high barriers for competitors.
Hey EC, thank you for reading and this thoughtful comment. I really appreciate this feedback. I wanted to take my time to respond.
Firstly I completely agree with you on the flaws in a traditional IRR/DCF approach. No valuation approach is perfect and that's why i am trying not to put too much weight on these valuations in my articles as it's highly subjective, especially for growth companies like Nvidia. I certainly hope people are not treating these articles as stock a stock picking service because really the thing that I am most focused and interested in is the business analysis. Thats why I spent three articles discussing the business in depth and a much shorter one on the valuation - that really isn't the focus for me. Actually one of they main reasons I even do a valuation model is that it helps me think through the business and understand the drivers a lot better.
Re valuation, I could be completely off in so many assumptions but the biggest one as you pointed out is what does the future multiple need to be. That is usually by far the most sensitive assumption. This is the key thing I tried to highlight, that if you believe the stock can continue to trade on 40-50x+ PE/FCF, then you'd make that investment all day. If you believe it should revert to a more normal market multiple as most stocks tend to do over time, then the results will be a lot lower. I've had bulls tell me 30x is too low and bears say it's ridiculously high, so there's a huge divergence in opinion. I've got no idea what the stock should will trade at it in the future, but this valuation exercise helps you understand what you need to believe in to earn a certain return. I expect readers will make up their own views.
In any case i completely agree with your points around how exceptional Nvidia is - I am long the stock after all and plan to hold it for a long time. I would hope that whatever numbers I put down are conservative, and if they can unlock other growth options like Auto, Robotics, Omniverse etc, just like they unlocked Data Center 4-5 years ago, they could easily continue to maintain these current high multiples.
Anyway thank you once again for reading the articles and taking the time to write this feedback. Having these sorts of exchanges is exactly the reason for why I put my work out there.
Cheers
Hi punch card! Really enjoyed this write-up! Would also love to get a copy of the model to dive deeper into the numbers! Jc_1336@yahoo.com
Thanks!!
Due to so many requests, I have now included the model as an attachment at the end of the article. Feel free to download
Hi Punch Card, could you also share the model with me? My email is dgray16@hotmail.com. Much appreciated!
Hi Punch Card, can u share your model with me too? my email is irene_lye@hotmail.com.. thank you very much
sure - ill clean it up over the next few days and send it around
thank you..!
I'm wondering whether you have bought it or not. Standing at May 2024, what do you think?
Great analysis! Love the visuals & impressive model -- would love a chance to check out the model & talk more offline --> ramsey@babbl.dev
Cheers!
I have now included the model as an attachment at the end of the article. Feel free to download and reach out if you want to chat about it - DM on Twitter would be best
Hi Punch Card, just found your Substack and I find your viewpoints on companies like SEA interesting. Currently a student in a local Singapore university, trying to break into the PE space. I’m interested in hearing more about your experience. Any chance we could connect off this platform?
Hey Ryan, thanks for reading. Happy to chat. Why don’t you DM me on Twitter @punchcardinvest or leave me your email. Cheers
Can’t seem to dm you via twitter. My email is ry0001en@e.ntu.edu.sg
40% YoY in data center for next year seems aggressive no? I have been struggling to model out the best way to solve for data center outside of using these pie in the sky TAM CAGR projections. Do you think we need to go through a digestion phase? Any other ideas on how to model data center growth more accurately?
So FY23 for Nvidia is actually a Jan-2023 year end, so most of it is CY2022 of which we are more than half way through. I had a $14.9bn DC revenue forecast for FY23, and in the first 2 quarters they've done $7.6bn, so even if the next two quarters are flat they should readily achieve this full year number I think. Beyond that, typically the approach that analysts use is to try to comp it the revenue growth, or more accurately capex growth, of the hyperscale cloud providers (AWS, Azure etc), who are a large customers of Nvidia, which is what I tried to do as well. The tricky thing is that AWS and Azure capex is not broken out separately in AMZN and MSFT's financials, so some estimation is required (or potentially just assume that capex grows in line with revenue, which is reported and forecasted by analysts).
Also worth noting that Meta is becoming a large Nvidia customer which should provide further support for FY24 (ie. next year) and possibly beyond - https://twitter.com/punchcardinvest/status/1582665839634894848
Overall I agree with you in that it's not an easy segment to forecast, which is why for conservativism i've assumed a reasonable moderation going forward, and hope to be surprised on the upside. Hope this helps.
for the $19bn 24 number though - don't you think there will be some digestion in data center / cloud capex? Are the only customers really the large cloud hyperscalers? What growth are you expecting for in the end markets, and then I assume you have NVDA grow faster with penetration of total DC/cloud capex. The CEO's explanation this Q for enterprise vs. datacenter trends was enterprise was renewals and DC was new builds, but wouldn't new builds be an easier thing to cut going forward with macro weakness?
One more: what does NVDA net on a sale of a consumer graphics card when taking out their partners share - thinking gaming segment? (they don't make $1k per card I assume)
Great questions. These are all fair concerns, although so far all the cloud providers are still reporting continued strong investment and growth, plus the Meta factor I mentioned earlier which i think could make up for any Enterprise weakness, but whether next year's growth is 30% or 25% or 20%, hard for me to say. I know all forecasts will be wrong, so my approach is to try be directionally right long-term and see if the the returns look decent under what I think could be broadly achievable (hence then using concepts like TAM to benchmark long-term projections, but that's obviously got it's own problems). But to give you a sense of the sensitivity, if my FY28 DC revenue is off by $5bn, it shaves about 1% off the IRR.
On the consumer graphics cards, what Nvidia sells to the OEMs is the GPU, which from some numbers i saw floating around could be somewhere between $150-200. The rest of the card is actually built by the OEM, which is what takes it up to the $1K mark including retail margin
Thank you @punch card investor...appreciate !
Do you share the Excel models?
I sure can, will just need to clean it up a little. What is your email ?
jphan1990@hotmail.com
Thanks!
could you share with ngtomkwongchun@gmail.com too? thank you!