History?
Perhaps it's just remembering the 2020 whiplash: huge demand, then tons of supply, then demand vanished and profits tanked in physical goods. Or maybe it's just knowing how the chip industry works in cycles. No doubt Nvidia is about to put some gaudy numbers up from revenue to profits. However, while we at Avory & Co. believe that generative AI and accelerated computing are here to stay, nobody knows what steady-state demand will look like. How much of today's experimentation will stick and become consistent workloads? How long will we be in the experimentation phase, which most likely requires excess computing? These are important questions for anyone blindly moving into investments driven by AI.
To begin, here is a chart that displays the history of huge swings in revenue growth within the semiconductor space. That is the starting point, so let us dig in.
AI's Rise Fuels GPU Demand, BUT When Will We See A BullWhip?
The boom in AI has ignited a surge in demand for high-performance GPUs, especially those made by Nvidia, AMD, and now SuperMicro on the hardware side has entered the ring. Counter to some, the upward surge in equity values is somewhat logical to us at Avory & Co. given the somewhat sudden demand for this technology. Now, if we know investor history, the pendulum swings far from one extreme to the other, and in some cases, blinders are put on during the ride. So today, I want to remove the blinders and explore the potential other side of what we consider to be an important technological evolution.
Imagine a chain whip – a slight movement at the handle results in amplified flicks at the tip. In supply chains, this translates to minor fluctuations in consumer demand (for AI applications) leading to exaggerated swings in upstream demand (for GPU production). Predicting these fluctuations is challenging, especially in AI's evolving landscape. Below shows a bullwhip effect within retail. The principles here exist for semiconductors. Customers want compute power, therefore application vendors seek compute-ready hardware, hardware vendors then seek the processing chips required to meet this demand, which flows to the chip makers to ensure they can meet all this demand. In each phase, the order book grows to ensure you can meet the demand.
AI to GPU Relationship
The illustration above is important today given the demand for AI. AI and GPUs fuel each other's growth, right now you cannot have one without the other. AI requires computational power and as of today, Nvidia is one of the few companies to meet this demand. Its GPUs, designed for parallel processing, power countless AI applications, from image recognition to natural language processing. All the tools being used today claiming AI are likely being trained using massive GPU clusters from Nvidia. All parts of the technology stack are competing to win AI business. From applications to on-prem server vendors to hyperscale cloud vendors.
A Demand Peak is Elusive
Now the question is, when will we hit peak medium-term demand? Because this will send the whip violently in the other direction. Think of how autos several years ago couldn't make enough cars, only to now be underwhelmed with less than anticipated demand, leading to falling margins. For AI, we are tracking alternative data to gauge interest. So far, AI adoption shows no signs of stopping, Google Search trends on AI remain elevated, companies discussing it have moved up, AI app downloads remain firm, and product announcements all include some level of AI. Below we share trends over the last 2 years that show prompt engineering and AI services in demand.
Margin vs Market Share?
When demand moves this far this fast, not only are analysts who cover the area making highly unreliable predictions but the companies themselves are set up to make forecasting mistakes.
They can either:
1) Use recent trends to forecast future demand ensuring they capture any demand that comes their way, ie revenue
2) or they apply conservatism to recent trends to attempt to capture as much demand as possible while ensuring they do not oversupply the market.
These are two very hard things to get right because at any point you are either giving up market share or giving up margins down the line. We witnessed this during COVID for retailers. We do not know the medium-term peak demand, but that day will come. So to prevent over-estimating growth, it is best to look at what the size of the market could look like.
How Big is the Market for Accelerated Computing
The accelerated computing market, encompassing high-performance CPUs and GPUs, boasts an estimated annual revenue potential of $100 billion to over $400 billion. This figure combines market size estimations for CPUs and current GPU sales, with additional forecasts from various firms. However, the market remains fluid and its long-term trajectory is uncertain.
While NVIDIA is currently a dominant player with projected revenue of $90 billion in 2024, representing roughly a quarter of the potential mid-term market size, it's crucial to consider various segments and future trends:
1. Gaming: This is the largest market for GPUs, accounting for approximately 60% of total sales. Gamers prioritize high-performance GPUs for games. Nvidia owns this space, and the computing required in gaming is now required for AI. This is how Nvidia found itself in the leading position.
2. Professional Visualization: This segment includes fields like graphic design, animation, video editing, and engineering workstations. These applications require GPUs for complex 3D rendering, video processing, and simulations, representing around 15-20% of the market according to most research.
3. Artificial Intelligence (AI): This is the purpose of this piece but the rapid rise of AI has propelled the use of GPUs for deep learning workloads like image recognition, natural language processing, and more.
4. Cryptocurrency Mining: While its prominence has fluctuated, cryptocurrency mining previously consumed a considerable portion of GPUs.
5. Data Centers: Hyperscale data centers utilize GPUs. Now this is for another piece but hyperscalers are now in the process of creating their chips. Google's TPU, Amazon Trainium, Meta and Microsoft are also doing this. This is a longer-term risk.
6. Embedded Systems: GPUs are increasingly finding their way into embedded systems like self-driving cars, autonomous robots, and edge computing devices. This emerging market holds future potential for GPU adoption.
When Could We See the Air Pocket, that Leads to Bull-Whip?
The answer to this question is likely 6-18 months out where we could hit a bookings wall. This means all the demand for 2-3 years of experimentation and implementation today is accounted for. At this point, companies like Nvidia and AMD will be serving their orders, but order books will be drawn down. We saw this for server sales for companies like HPE and Dell over the last 2-3 years.
Valuation
NVIDIA's current sub-30x forward earnings multiple seems attractive considering its growth story and margins. However, this valuation assumes a sustained demand spike, which might not be realistic. Chip cycles tend to be sharp, and we thin eventually this could be the same. So exercise some caution.
Overall
No doubt Nvidia is in the driver's seat to take advantage of the demand for AI services. The risk is the air pocket in growth, the question is when? We know next quarter will be great, as will the next few quarters at least. All we are saying is that at these levels some caution is likely warranted.
About Avory & Co.
Investing where the world is headed. Avory is a high-conviction equity asset manager focused on Secular Growth and Transformation Stories. Allowing data to tell our stories. Avory invests in valuable companies led by exceptional teams. Avory & Co. is located in Miami, Florida offering SMA's for institutions and high-net-worth individuals.
📺 YouTube Investment Research: https://www.youtube.com/@avoryco
🎙️ Inside Scoop Podcast: Apple Podcast
🌐Avory Website: https://www.avory.xyz/
📝Avory Annual Investment Letter: Read
⚡Coming Soon⚡
Avory will be launching an Active ETF mimicking Our Premier + Strategy. Get Excited. Reach out to find out more. team@avoryco.com
Disclosure: Not a recommendation to invest in any companies mentioned. Investment involves risk.