By any standard, Nvidia's (NVDA) - Get Reportshares have been on an epic run this year, more than tripling on the back of multiple blowout earnings reports and general enthusiasm about the company's gaming, data center and automotive efforts.
After seeing such massive gains in a stock that has rarely seen such giant moves since its 1999 IPO, it's understandable that many investors would heed a call from a research firm to take profits. Especially when that firm, though not having a perfect track record, has made a slew of impressive bearish calls on equities over the last few years, notably one onValeant Pharmaceuticals (VRX) .
All the same, the arguments that Citron Research, led by Andrew Left, make for selling Nvidia feel very much like a mixed bag. Some of them involve legitimate concerns for a stock that was trading at 43 times its consensus fiscal 2018 (ends in January 2018) EPS estimate when Citron made its case. Others, however, are on shakier ground and/or involve issues that have been well-understood by investors for some time.
On Wednesday, Citron -- while providing the caveat that it has long been a fan of Nvidia -- " it thinks markets are ignoring, and predicted the GPU giant's shares will head back to $90 in 2017. Since then, Nvidia has fallen about 8% to the $110 range, putting it on its way to hitting Citron's target.
Here's a look at each of the "risks" cited by Citron and a closer look at each claim.
Risk #1: Much of Nvidia's growth "has come from gaming and at AMD's expense," rather than from "total new addressable markets."
There's a grain of truth to this: The fact that Nvidia has faced little high-end gaming GPU competition from AMD (AMD) - Get Reportof late is one reason why its gaming revenue -- buoyed by sales of the company's Pascal-architecture GPUs, which launched earlier this year -- rose 63% annually in the October quarter to $1.24 billion (62% of total revenue). And as noted in one of my recent tech predictions columns, the upcoming launch of AMD's Vega-architecture GPUs will likely spell tougher high-end competition; Nvidia is expected to counter later in the year via its Volta GPU architecture.
However, Nvidia is also benefiting from the fact that the PC gaming hardware market has been growing in size, bucking broader PC weakness as gamers invest in systems capable of playing demanding new titles at high resolutions (up to 4K) and frame rates. And October quarter results were also boosted by sales increases of 193% and 61%, respectively, in Nvidia's data center and automotive revenue, which in turn benefited from growing addressable markets for server GPUs and infotainment system processors.
Risk #2:Nvidia faces "significant competition" in the data center from "existing and emerging players" such as Intel, AMD and Xilinx.
This is generally true. Intel (INTC) - Get Reportstepped up its efforts to battle Nvidia in the server accelerator card market this year by launching its Knights Landing Xeon Phi processors, and also bought AI/deep learning ASIC developer Nervana Systems. In 2017, Intel plans to gradually roll out a Nervana-based chip codenamed Lake Crest, and also launch a Knights Landing successor known as Knights Mill that promises major deep learning performance improvements.
For its part, AMD just unveiled its Radeon Instinct server GPUs, while highlighting their support for open-source AI software tools and libraries; they're expected to be available in Q2 2017. And Xilinx (XLNX) - Get Report has made some noise about the ability to use its field programmable chips (FPGAs) to handle deep learning algorithms.
At the same time, Nvidia's Tesla server GPUs are still seen by many as the gold standard for accelerating high-performance computing (HPC) and especially AI workloads. Nvidia has poured a lot of resources into both optimizing its GPUs for deep learning, and building out the Tesla line's software ecosystem. And judging by the uptake for the Pascal-based Tesla GPUs that launched this year, those investments are paying off well.
Also, compared with ASICs and FPGAs, GPUs arguably remain in the sweet spot for AI, particularly when it comes to training neural networks how to handle a given task. ASICs can deliver great performance for specific algorithms, but offer less programmability. FPGAs offer unmatched programmability, but less performance.
Risk #3: Intel has access to Nvidia's intellectual property through a licensing deal, calling into question the notion that Nvidia deserves a premium valuation due to its IP.
Intel and Nvidia's cross-licensing dealdoes give Intel the right to use all of Nvidia's patents that are filed on or before March 31, 2017. It also gives Nvidia access to certain Intel patents filed on or before that date.
But if this cross-licensing deal made it easy for Intel to develop GPUs that can challenge Nvidia's, one would think the company would have long ago done so. Instead, Intel has focused on creating integrated GPUs for its processors that don't come close to matching Nvidia (or AMD's) most powerful offerings. The engineering resources Nvidia pours into GPU development, together with the software tools and developer ecosystem it has created, still matter a lot.
Risk #4: The pending loss of patent-licensing revenue from Intel will have a big impact on Nvidia's 2017 bottom line.
Nvidia has been steadily recording $66 million in quarterly licensing revenue from Intel -- virtually all of which is pure gross profit -- and that revenue stream will indeed disappear after March. But this has been known by investors for years, and thus presumably priced in.
Risk #5: The arrival of new competition from Intel in mid-2017 will hurt Nvidia's gross margins.
This argument revolves around the fact that Intel manufactures its own chips, whereas Nvidia relies on a third party -- Taiwan Semiconductor (TSM) - Get Report-- that records about a 50% gross margin (GM) for itself. Citron thinks that Nvidia, which had a 59.2% company-wide GM last quarter, gets about an 80% GM on the TSMC-manufactured chips that Intel will take aim at, leaving Intel able to "charge 37% less for its chips and still generate 80% gross margins."
But this argument assumes Intel's manufacturing costs for relatively low-volume Lake Crest or Xeon Phi chips are no different from the costs TSMC, which specializes in mass-producing chips for "fabless" chipmakers such as Nvidia, incurs while making Tesla GPUs. And that assumption seems questionable, particularly given Intel had a company-wide GM of just 64.8% in Q3.
Moreover, there's little evidence that Intel's vertical integration has allowed it to exact severe margin pressure on other fabless rivals that rely on TSMC, such as Qualcomm (QCOM) - Get Report, Broadcom (AVGO) - Get Report and Marvell (MRVL) - Get Report. It's possible that Intel will price Lake Crest and Knights Mill aggressively to take share from Nvidia, but if it does so, the company will likely take a big margin hit of its own.
Risk #6: Competition from Nvidia customers developing custom chips and "GPU workarounds" is a threat.
Here, Citron cites Alphabet/Google's (GOOGL) - Get Report Tensor Processing Unit (TPU), an ASIC used by the search giant for certain AI workloads. It also mentions Apple (AAPL) - Get Report, which runs AI algorithms it has developed on user devices rather than on cloud servers, citing privacy considerations.
To an extent, this is a real long-term concern. In addition to Google and Apple, Citron could have cited Microsoft (MSFT) - Get Report and Baidu (BIDU) - Get Report, which have used FPGAs in AI projects. However, Apple, Microsoft and Baidu have thus far only turned to non-GPU processing for some of their AI work, not all of it. And although Google has been tight-lipped on the matter, there's speculation that the same holds true for Google's TPU, given the programmability limitations of ASICs.
While there are several aspects of Citron's bear case that can be questioned, it's worth keeping in mind the company's $90 target is still 173% above where Nvidia traded to start the year. Compared with many of Citron's other calls, this one is pretty tame. And considering the heights Nvidia has risen to this year, some of the concerns raised by the firm do bear monitoring, even if one shouldn't buy into all of them.