Description

I believe nVidia is a short. Investors have mistaken significant short-term growth in revenue for a secular revenue growth rate. Most people here are probably familiar with recent tech hype in AI, autonomous driving, and crypto currency. nVidia has become the benefactor from much of this hype due to its graphics processing units (GPU) being used in all these end markets. I believe the most significant near-term driver that has significantly boosted revenue has been use of nVidia’s GPUs in mining Ethereum and associated coins where an ASIC is not available, as it is with Bitcoin. For a similar story, please reference the write-up on AAOI where investors mistakenly believed AAOI was benefiting from long-term generational growth in data center when they were really benefitting from short-term use of 40GB optical transceivers in a transition to 100GB. Similar to AAOI, nVidia is in the right place at the right time but I believe that is about to change.

Most people are aware of the crypto currency craze that drove prices up in the thousands of percent over the last year. nVidia was a significant beneficiary of this craze as GPUs were the only efficient and effective technology to mine Etherium and it’s associated alt coins. Similar to most technology pricing, GPUs typically decline in price over time so a lagging edge nVidia GPU will typically cost 80% of it’s MSRP at introduction 1-2 years after it is introduced. There is also a market for used GPUs where PC gamers can buy lagging edge GPUs at even higher discounts to update their PCs for better gaming performance. With the massive increase in the price of Ethereum, the price of both lagging edge GPUs and used GPUs went from heavy discounts to trading at premiums as high as 3-4x MSRP.

Even with the rise in GPU prices, the incentive to mine Ethereum was still significant as the ROI was still there. With the recent decline to $370 in March the payback period for somebody with low-to-reasonable power rates (it matters significantly if you are mining in Japan versus North America) and an optimized mining rig increased to about ten years. While the recent move in price above $600 will obviously improve the ROI, it is nothing like it was in December and January. It is also important to remember that with the increase in miners and the passage of time, the degree of difficulty in mining Ethereum increases. With the drop in price, and assuming a 20% per month increase in the degree of difficulty to mine, many miners find the payoff too insignificant to matter. No doubt the drop in price will lead to uncertainty and typical deer caught in the headlights syndrome.

Why does this matter to nVidia? Obviously as the largest GPU company they have benefited significantly from the growth in mining activity. The Wells Fargo analyst is one of a very few skeptics and he believes that nVidia’s mining revenue would be in the $300-500 million range in calendar Q1. Given the price momentum all of last year but especially November-January my guess is a great deal of nVidia’s mining GPU benefit came in fiscal Q4 ending in January 2018. The most detailed analysis comes from an article https://cryptocurrencyfare.com/attn-cramer-nvidia-the-ultimate-short-thesis/ that uses the Ethereum network hashrate to calculate that 6.2 million GPU cards were sold in 2017 and another 4.4 million cards in Q1 2018 with a total value of $2.65 billion. The writer doesn’t detail ASP used to make the calculation but even assuming lower historical MSRPs and the use of used cards the number would still lead to a significant benefit to nVidia. Admittedly, AMD has benefited too but given the significant differences in revenue between the two companies I am going to assume nVidia has benefited more.

There are there ultimate effects of this on nVidia. First, the significant benefit from mining revenue that nVidia has received in the last few quarters will lead to a slowing growth rate in revenue over the next few quarters as less people deploy mining rigs for Ethereum. Second, there should be a significant market for used cards as former miners exit the market and try to reclaim capital invested in mining. The buyers of the used cards would be PC gamers who have stayed out of the market as prices rose and it become uneconomic to build your own gaming rig versus buying one new. While the used market may not be incredibly robust as gamers concern themselves with the card quality after heavy mining use it is still bound to have an impact. Finally, the ability for nVidia (direct distribution through Founders Edition) and their distributors to sell cards at stable prices versus a discount to MSRP will diminish and impact revenue and margins. A cursory search on Amazon or newegg reveals that card prices have already dropped 30%+ in the last month and there are stories in Reddit of increasing availability of cards at MSRP. In fact, you can find a post from someone in Japan able to buy cards at MSRP in the last week. As many of you know, an estimated 70% of crypto mining came from Japan and Korea so this is not insignificant. Finally, last week TSMC guided revenue growth down from 10-15% to 10% citing weakness in iPhone and crypto. I would also note that as recently as February, nVidia management said they expected GPU prices to continue to climb through the second half of 2018 only to see them drop 30% in March. Clearly the early signs of a slowdown are evident.

There is another issue that is much more speculative that could affect demand for crypto GPUs. Bitmain, a Chinese miner, has developed an ASIC for Ethereum mining the E3. The cost of the ASIC is about $800 and has similar power efficiency as a GPU. It’s effectiveness, however, is estimated to be the equivalent of 4-6 GPUs that would cost 2-3x as much diminishing ROI. While this would have obvious impacts on demand for GPUs there is speculation that the introduction of the ASIC will lead to a hard fork as Bitmain’s use of their own ASIC will allow them to monopolize Ethereum. This would be reinforced by small miners using Bitmain’s ASIC. You can speculate on the outcome here but all I will say is that with the decline in ROI there is incentive to build a lower cost rig. That doesn’t mean the Ethereum community won’t manage to shut them out though. I would note that Bitcoin did convert to ASIC mining impacting AMD but that was in a more nascent market. Finally, there is speculation (so I won’t be going in to detail) that Ethereum will convert from a Proof of Work market to a Proof of Stake market which would likely inhibit large scale miners from participating which could inhibit demand for GPUs as they exit the market.

Another area that Nvidia has been benefiting from recent hype is autonomous driving. I would note that this is 5% of their revenue and did not grow in their last quarter. It is also mostly infotainment related so no matter the narrative it won’t be contributing significantly to near term growth. Longer term maybe, but given that an autonomous car currently consumes about 1000 GBs of data autonomous cars are unlikely to be a significant driver until third generation 5G or whatever 6G is.

The core gaming GPU business is a strong franchise and I expect it to grow in the future. I would note that UBS estimates the PC gamer market was flat in 2H 2018 (negative growth in Q4 FY2018) and the growth benefit likely came from the Nintendo Switch which sold greater than 14 million units and an estimated revenue contribution of $1.4 billion in 2017. Growth is likely peaking for the Switch and off a zero comp in fiscal 2017 made a meaningful contribution in fiscal 2018/ calendar 2017. As noted above, contribution from PC gamers is likely going to be hindered over the next year by supply in the used GPU market from miners who have thrown in the towel and are trying to recoup capital. Finally, AMD is also introducing a new processor at the 7 nm node. AMD has made limited disclosure about their plans and this chip may not even make it in to a GPU as it’s currently positioned for high performance compute so I’m not going to speculate on the outcome to nVidia’s core gaming business except to say there is new competition on the short-term horizon. At $5.5 billion in revenue, gaming is more than half of nVidia’s revenue. Many analysts believe that gaming will grow in the teens-to-twenty percent range in FY2019 even though there is a lack of clarity on how much crypto contributed, the Switch is peaking and there is the threat of a large used GPU market developing. I believe the core gaming business will be down in FY2019 and lead to an overall decline in nVidia’s revenue.

The final business that is growing at a very high rate and is projected to continue to grow is the data center business. This business represents about $1.5 billion in revenue and grew 50% last year with projections that it will grow at a 40% rate through the early 2020s. The growth in data center is coming from GPUs being used for machine learning and deep learning. I have to admit that growth in AI is not just luck as nVidia did invest heavily in CUDA, their AI development platform, so they were in the leading position when development dollars were pushed in to this area. And management is being awkwardly promotional about the opportunity framing it as a $50 billion TAM which is probably closer to the whole data center hardware TAM. Nevertheless, cudas to the investment and I do believe the business will continue to grow in the near term. The big question is how much?

In Saturday’s Barrons there is an article “The Looming Battle over AI Chips”. In the article they reference an article in Bloomberg speculating that Facebook is in the process of developing their own chips. They also reference Facebook’s head of machine learning Yann Lecun as saying “The amount of infrastructure if we use the current type of CPU [central processing unit] is just going to be overwhelming.” While he only mentions CPUs you can assume that the existing AI architecture includes both a CPU and GPU. While it is not clear that Facebook is developing a proprietary processor, it is not a glaring assumption to believe that anything that reduces their compute cost will take precedent especially when it is evident that their operating costs are about to rise, possibly significantly.

They do have a playbook to follow in that Google has already beat them to the mark. Google Cloud’s AI/ ML technology includes TensorFlow, an open source machine learning framework for machine/ deep learning that allows AI researchers a much more simplified development platform than CUDA. https://techcrunch.com/2018/03/15/the-red-hot-ai-chip-space-gets-even-hotter-with-56m-for-a-startup-called-sambanova/ This article gives a very detailed description of the competitive environment in AI hardware with some pretty significant remarks about the threat to nVidia’s position. If you want greater detail on the future of AI hardware competition I would reference this article as I don’t want o make this write-up unnecessarily long. Current customers of TensorFlow include Airbnb, Uber, Snap, Intel and more. Beyond TensorFlow, Google has APIs for Video, Vision, Speech, Natural Language and Translation. Along with this, Google currently has a second generation TensorFlow Processor in beta that will power search and the APIs listed above. While they already have the large scale customers listed above on TensorFlow it makes sense that with the new generation they will probably open this newer version to more customers including small-to-mid size researchers. UBS published a piece on January 24^th entitled “Getting started with deep learning on TensorFlow” if you would like more detail on its capabilities. I would note that JP Morgan has estimated that the next generation TensorFlow Processor will provide a $1-2 billion dollar revenue benefit to Broadcom who will make the chip. This is similar in revenue scale to nVidia’s current data center business which, if correct, is likely to significantly dampen nVidia’s presumed 40% growth rate going forward.

Beyond Google and Facebook, I have little doubt that Amazon is working on something similar while there are many rumors that Intel/ Altera are working on an FPGA for AI. Beyond the behemoths there are many start-ups including one that was involved in developing the first generation TensorFlow.

To summarize, I believe estimates significantly under-estimate the impact to earnings and revenue from crypto mining for nVidia. This will have knock on effects in the sales of the gaming segment as miners off load used GPUs and they are bought by PC gamers. This will also impact pricing on lagging generation nVidia gaming GPUs. Also, contribution from Nintendo switch has likely peaked or at the very least will face tougher comps. The automotive segment, at 5% of revenue and currently not growing, is not significant enough to matter. While nVidia has done a good job in positioning themselves for growth in AI, there is some significant competition imminent and further out that will hurt their ability to grow in this segment leading to damage to the AI narrative and slower than assumed growth rates. I am a long only PM (although I do have the ability to short in the portfolios, I have never done so) and feel it’s a bit of a mugs game to give a target price on a short. nVidia was a $30-50 before the crypto craze but this was also before the contribution from their well positioned AI framework. For the stock to maintain a level above $100, AI will have to continue to make a large growth contribution at rates higher than 20%. My guess is when the effects of the decline in crypto happens, people will become more skeptical of AI leading to a lower revenue multiple for the AI segment. Time will tell.

Catalyst

Decline in crypto mining revenue

Increase in perceived threats from other AI suppliers.