Crypto Bubble Hit Nvidia Harder Than Expected
After the fiscal Q3'2019 report, the majority of the sell-side community was utterly shocked by the size of Nvidia's (NASDAQ:NVDA) miss. Nvidia has had a phenomenal track record of delivering beat-and-raise quarters.
- Revenue of $3.18 billion vs. consensus estimate of $3.24 billion
- EPS of $1.84 vs. consensus estimate of $1.71
- Revenue guide of $2.7 billion (+/- 2%) vs. consensus of $3.4 billion
- Gaming revenue of $1.76 billion vs. consensus of $1.89 billion
- Datacenter revenue of $792 million vs. consensus of $821 million
The real miss was on the gaming side, coming in nearly 7% lower than expectations. The cryptocurrency bubble continuously downplayed by management actually hit Nvidia harder than expected, with revenues falling far below guidance. Unfortunately for the bulls, because management overestimated the demand within the GeForce lineup, it overproduced. This created an inventory glut. This glut, according to management, will take multiple quarters to work through. While the entire market was focused on the effect a crypto bubble would have on AMD (NASDAQ:AMD), the market underestimated (partly in thanks to management's comments) the effect lowered crypto demand would have on sales. Now, despite the fact that Nvidia has come to market with its RTX graphics card lineup, it is going to have inventory issues, particularly with the sales of GTX 1060, 1070, and 1080 cards. These were some of the more popular GPU choices from Nvidia in the crypto mining realm.
The impact crypto had on both Nvidia and AMD was continuously downplayed by the management team of both companies. The real problem was the fact that the crypto bubble had burst in late December, causing profit margins in crypto mining to slide, with GPU demand drying up.
While this article is not meant to debate the validity of cryptocurrency, the plunge in crypto prices clearly affected Nvidia in an adverse manner. Despite this, the management teams at Nvidia and AMD downplayed crypto, more-so at Nvidia. AMD's stock was extremely volatile, while Nvidia remained in a general uptrend despite the bursting of the crypto bubble.
AMD shares had plunged from as high as $13-14 all the way down to $9.04, the 52-week low, primarily driven by the crypto bubble bursting and the results afterward. Meanwhile, Nvidia had a much smaller ~6% decline in shares, as the sell-side and management continued to reinforce the long-term Nvidia growth story.
The problem was, Nvidia's management made the mistake of continuing production of GeForce graphics cards despite the dissipation of crypto demand. The thinking was, as the gaming market continues to expand with hits like e-sports, and games like PUBG and Fortnite, any demand vacuum created by a lack of crypto will be filled with significant gaming demand.
We need to take a step back and look at why GPU demand took such a heavy hit in the crypto space. For a mining rig to be profitable, there are basic costs like electricity and the mining hardware/software itself. Electricity costs ramp as the size of the operations ramp. The COGS of a mining rig ascend linearly with increases in computing power. But similar to say an oil driller, the entire business is dependent on the price of the commodity, the cryptocurrency itself. So higher crypto prices drive solid gross margins on a mining rig, offsetting the costs of operating the rig, when crypto prices came crashing down the entire profitability of the mining rig was destroyed. When the price per barrel of oil drops, so will gross margins and overall profitability.
The problem is, profitability was hit so much that the economic interest in crypto mining vanished.
The second reason for the hit Nvidia took from crypto was the fact that more efficient processors were headed for the market. Even if crypto was a sustainable market, Nvidia and AMD had competition gunning for their GPUs. In late March, Susquehanna released a note slashing its price target on Nvidia from $215 to $200, citing increased competition in crypto. This competition came from newer, far more efficient processors specifically targeted for crypto mining. These processors, known as ASICs (application specific integrated circuits), are developed solely for a single task at hand. In this case, the ASIC is developed for crypto mining. ASICs, while often more expensive than GPUs, are faster in terms of performance, representing a more compelling buy for miners.
The majority of the revenue that Nvidia and AMD generated from GPU based mining was sourced specifically to Ethereum. Eventually, Ethereum miners, just like Bitcoin miners before it, were going to move to ASICs instead f GPUs. ASICs are just more powerful. Even if the crypto market didn't tank, ASICs would eventually eat into the sales of GPUs.
Nvidia's management made the mistake of believing that gaming demand from Nvidia's GeForce lineup would fill any crypto demand vacuum to the levels at which crypto demand was present previously. And Nvidia just rolled out a new series of gaming GPUs that would fill this demand vacuum.
For so long, the crypto bubble drove GeForce ASPs sky-high, effectively pricing gamers out of the market. These GPUs, while including the new ray-tracing technology, were much higher in terms of price, with basic specs not improving much. Unlike say an Apple (NASDAQ:AAPL), GPUs are bought for computational efficiency and strength. iPhone buyers don't necessarily buy an iPhone for higher clock rates. The performance boost in basic metrics was not great enough to warrant high demand in the gaming segment.
On top of this, there were a high number of reported product defects, and a limited rollout of ray-tracing compatible games, effectively destroying the RTX value proposition.
GPUs are considered valuable buys if the price accurately matches the performance. With the RTX, the price has been too much higher than the performance. The lack of demand for RTX has led to this 2-3 quarter inventory reset.
Strengthening Competition From AMD in Gaming/Datacenter GPUs
Like the CPU and memory space, the GPU space has become a space of limited competition. The only true competitor to Nvidia in this space is AMD. For years now, AMD, and ATI before it, struggled to truly compete with Nvidia in the high-end gaming GPU market.
AMD has an opportunity to add to its market share in gaming and the datacenter if it exploits certain strategies:
- Partnership with Intel (NADSAQ:INTC)
- Lower prices than Nvidia
- Bundling CPU and GPU offerings into one (for enterprise clients)
- Get ahead of Nvidia on 7nm
The first move AMD has made to take share from Nvidia is partnering with Intel to develop an APU with Radeon integrated graphics. While the development of these APUs has been limited, it could contribute to market share grabs by AMD over Nvidia. This move works for both Intel and AMD. Intel gets increased exposure to high-quality graphics, and AMD gets exposure to increased volume and exposure, taking share from Nvidia.
That being said, unless AMD and Intel continue with the partnership, launching new iterations year after year, the partnership will not make a significant dent on Nvidia's growth trajectory.
The second thing AMD can do to take market share on Nvidia is continue offering lower-priced solutions, while being competitive with or even surpassing the performance of Nvidia GPUs. Operating as a dominator of the budget graphics market would not necessarily be a bad thing. However, management has made clear that it would like to put premium products first, with budget prices second. That is why AMD's high-performance cards are becoming closer to GeForce cards in price.
Thirdly, AMD has an advantage in the datacenter that neither Nvidia nor Intel has, something that could lead to AMD gaining share in CPUs and GPUs. Out of the three companies, AMD is the only one that offers CPUs and GPUs in the datacenter.
Intel only makes CPU for its enterprise clients, while Nvidia only makes GPUs. AMD has the unique ability to bundle its EPYC CPU offerings with its Radeon Instinct GPU offerings, an advantage that none of its competition has. This could be a factor that allows AMD to hold some market share in the datacenter GPU market.
The next shift in GPU process technology is moving to a 7nm node. If AMD moves efficiently enough into this next arena, it has a good chance at continuing market share grabs, particularly within the datacenter.
AMD has already launched a machine-learning GPU in the 7nm node. This GPU, the 7nm MI60, operates at 30% higher performance than AMD's previous 14nm MI25 GPU.
Assuming AMD gets to scale this GPU, AMD's competitive standing in the datacenter could improve drastically. In my view, if AMD develops and scales this performance GPU, AMD could even become a leader in the market. That may seem like a crazy proposition, but Rosenblatt analyst Hans Mosesmann believes AMD's 7nm GPU offerings are six-twelve months ahead of Nvidia's 7nm offerings. While Nvidia will probably get back ahead of AMD, as Nvidia has far more resources than AMD, AMD could gain market share in the datacenter GPU market.
Datacenter GPUs Could Get Hit By Vertical Integration From Their "Partners"
While some may dismiss the AMD threat in the datacenter, the GPU has significant threats it terms of its ability to compete technologically with new offerings.
If we look at Nvidia's webpage, the company touts the divergence between transistor growth and Moore's Law, and the GPU performance wins over the CPU. The market, management, and sell-side are not focused on the very real and very legitimate competition coming from some of its datacenter "partners". Each member of the cloud Super 7 is, in some way shape or form, developing their own in-house datacenter processors.
- Google (NASDAQ:GOOG) (NASDAQ:GOOGL): Google has been arguably one of the fastest movers into the in-house chip design arena. It has launched three separate generations of its TPU, displaying Google's ambition in the market. Google CEO Sundar Pichai said that TPU 3.0 would be 8X more powerful than its second-generation predecessor. While TPU 2.0 ran at 11.5 petaflops, TPU 3.0 is expected to run at 100 petaflops. That being said, TPU 3.0 is power hungry and requires liquid cooling. On top of that, Google is only using TPU 3 to power Google's own applications and power the Google Cloud Platform to handle client workloads. So, for now, deployment of the TPU is limited. Assuming the TPU is what Google says it is, if/when Google begins sales of this chip, we could see erosion in Nvidia's datacenter growth rate.
- Amazon (NASDAQ:AMZN): A couple of weeks ago, AWS announced the launch of its own AI exclusive chip. The chip, known as Inferentia, is designed to operate at a high performance levels for low costs. The chip will support deep learning frameworks like TensorFlow, Apache MXNet, and PyTorch. Inferentia provides high throughput, low latency performance, at a low price. While it hasn't been proven that Inferentia will succeed in providing better solutions than Nvidia, increased competition could hit Nvidia's datacenter operations. These chips will become available in 2019.
- Facebook (NASDAQ:FB): Facebook is one of the most transparent hyper-scalers when it comes to publishing and documenting what kinds of hardware it uses and for what purpose. Facebook is said to be developing its own inferencing chip.
- Alibaba (NYSE:BABA): Launched its own AI accelerator; dubbed the Ali-NPU, it is said to have 10X better performance than mainstream chips based on a CPU or GPU architecture.
Tencent (OTCMKTS:OTCPK:TCEHY), Baidu (NASDAQ:BIDU), and Microsoft (NASDAQ:MSFT) are all invested into in-house chips, most of them being FGPAs.
That being said, I cannot say with certainty that in-house chips designed by cloud hyper-scalers will completely replace Nvidia GPUs. While I am not fluent in the technicalities of chip architecture, the major competitive threats out of the big hyper-scalers (Facebook, Alibaba, Amazon, and Google mostly) have said that they do not intend to replace their Nvidia GPUs with in-house processors. The entire bull thesis would crumble if Nvidia's "partners" moved exclusively to in-house processors.
Is Autonomous Driving Going To Make A Difference?
While management and the bulls on Nvidia have said that autonomous driving will drive future growth, I remain more skeptical. The autonomous driving mania, in my opinion, is most comparable to the bitcoin mania that began to fade at the end of 2017. The auto market is ripe for an autonomous driving disruption, according to the bulls, and Nvidia is helping lead the way by offering autonomous solutions. I'm skeptical on both the autonomous driving market and Nvidia's place in the market.
First of all, we should expect some sort of infrastructure overhaul to adapt for an autonomous driving disruption. While not as dramatic an overhaul as say the transformation from horses to automobiles, the update from automobiles to self-driving cars will require transportation infrastructure to complete a large update, integrating increased software capabilities. After all, without continuous communication between the ecosystem and all its inhabitants, autonomous driving will not work. A complete overhaul of the automobile transportation infrastructure is billions of dollars and likely many years away.
When Nvidia throws out these catchy statements like the one at the very bottom of the page, my skepticism grows. What makes Nvidia think that the TAM in autonomous driving is going to be that large. And of the companies listed here, many of them appear to be small companies or outright startups. They must be doing phenomenal work to garner Nvidia's attention on the Q3 earnings slides. As long as there is news about someone dying in an autonomous driving car crash, sentiment on the autonomous driving business in general should be negative. According to a recent survey, only 21% of total respondents say they are willing to ride in an autonomous car. It will take many years, if ever to garner public enthusiasm and trust over autonomous driving. For that reason, I believe autonomous driving will be a niche business, almost like a bell-and-whistle for a car. And when Nvidia prints this rosy and almost arbitrary picture about the TAM in autonomous driving, my skepticism festers.
Assuming however that the autonomous driving market is a booming market at that, Nvidia is not short of high powered competition. This competition comes most notably from Tesla (NASDAQ:TSLA). Tesla is going in-house to design a chip that is (allegedly) 10X faster than Nvidia's current offerings. The chip was talked about in depth by Tesla executive Pete Bannon on the Q3 earnings call. While Tesla is an ambitious company that gets ahead of itself, it should make Nvidia more cautious. If automakers go in-house with chip design, the autonomous driving thesis could very well crumble. Another threat is general competition from Intel and AMD. Intel, via Mobileye powers Waymo, and it is rumored that AMD may enter the autonomous driving scene. All in all, I would be skeptical of the rosy TAM estimates Nvidia is providing.
Still An Extreme Valuation
Nvidia's stock while being sliced in half since peaking still trades at an extreme valuation.
Nvidia trades at the highest forward P/S ratio out of the eight semiconductor stocks, and the second highest on the forward P/E side. Meanwhile, its revenue growth rate in 2019 is expected to be 5.5%, with EPS being slightly negative. Meanwhile, AMD, the richest valued on a forward P/E basis, is expected to grow revenues at 6% next year, but grow profits at ~35%. And that is with crypto acting as a headwind on AMD's 2019. While revenue growth should reaccelerate in calendar 2020 onward, the dynamics of the next 12 months are not even close to priced in.
The median forward P/S of these eight semiconductor stocks is ~2.85X. Applying this valuation to Nvidia's stock gives us a $57 stock. Then again, many of these companies are cyclical chipmakers, and don't have the potential upside reward that Nvidia has. That is an important risk when valuing stocks relative to peers. If the market for Nvidia's peers is irrational, basing your assumptions about the company on those peers is just as irrational, if not more so.
Because of this, I am using a DCF model to value Nvidia's stock.
Here are my assumptions for cost of equity:
Here are my assumptions for WACC as a whole:
Here are my figures for Nvidia's long-term business:
Nvidia will have 2-3 quarters of inventory reset that will put pressure on GeForce, adversely affecting revenues and profits. On top of that, the legitimacy of the bull argument regarding autonomous driving and the datacenter is not sturdy. Combine all this with a valuation well above peers, investors should tread very carefully with Nvidia.
Disclaimer: This is not financial advice. Do your own due diligence before making trades in these securities. The views expressed in this article are my personal opinion.