Server Acceleration: Altera FPGAs Vs. Nvidia GPUs

| About: Intel Corporation (INTC)
This article is now exclusive for PRO subscribers.

Summary

Server acceleration is a rapidly developing profit opportunity.

Both GPUs and FPGAs are capable of performing this function.

Altera, using FPGAs, and Nvidia, using GPUs, are best positioned to compete for this business.

Competition will be on performance, ease of use and power consumption.

If Altera succeeds in garnering a substantial share of this market, there is considerable upside from recent prices in the $37 area.

Altera (NASDAQ:ALTR) CEO John Daane has been talking about a potential growth opportunity arising from the use of FPGAs in server acceleration for some time.

FPGAs (Field Programmable Gate Arrays) are a type of PLD (Programmable Logic Device). They have been used for many years in applications that may need to be modified after they are shipped or as an expense saving where the expected demand isn't great enough to support the engineering required to make a special purpose chip.

HPC (high performance computing), big data, search applications, video streaming and analytics for unstructured data all place extreme demands on server speed. For some time, it has been known that FPGAs, as well as GPUs (graphic processing units) can be paired with CPUs in ways that accelerate performance. As the demand for speed has increased, server acceleration is becoming important to meeting consumer and business expectations.

During the Q3 2014 earnings conference call (transcript), Daane estimated the long-term opportunity at $1 billion, which would represent an approximate 50% increase in revenues. Until recently, GPUs, such as Nvidia's (NASDAQ:NVDA) Tesla, have enjoyed a lead role in acceleration. And Nvidia CEO Jen-Hsun Huang has talked about using virtual GPUs to supplant an installed base of some 500 million units.

On the Q3 2015 earnings conference call (transcript), Huang was asked about competition from FPGAs in acceleration, and dismissed the idea, citing his own company's superior size and larger engineering workforce. He doesn't like the idea of Altera raining on his victory parade. Daane, for his part, asserts that the FPGA has an order of magnitude advantage in power consumption, and adds that power is the largest expense at data centers.

Huang claims his GPUs are easier to program, while Daane says that the availability of Altera's FPGA-based OpenCL compiler makes their competing products easy-to-use for the relevant developers.

For a well-written and informative article on the possible growth in FPGA server acceleration, I suggest "Will 2015 Be the Year of the FPGA?" by Nicole Hemsoth, published in HPCwire.

Dueling Duopolies

The FPGA market is dominated by Xilinx (NASDAQ:XLNX) and Altera, which effectively constitute a duopoly. Similarly, Nvidia and AMD (NASDAQ:AMD) share a duopoly in GPUs, although Nvidia has an estimated 76% of the business. Searching recent conference calls on "accelera," it appears that management and analysts are much more interested in the topic at Altera and Nvidia. That's what the word counts are showing.

My take: Nvidia has been looking forward to a lead role in the rapidly growing field of server acceleration, and Altera is now arriving at the scene with a competitive technology. As an example of Altera gaining traction, Microsoft will be using Altera FPGAs for Bing.

Both IBM (NYSE:IBM) and Intel (NASDAQ:INTC) have announced (link 1, link 2) products that combine an Altera FPGA with their server chips. Not to leave anybody out, Altera recently announced that its Stratix 10 SoC devices will incorporate a high-performance, quad-core 64-bit ARM Cortex™-A53 processor system.

Business Models Suggest Orderly Competition

As a general rule, companies that invest heavily in R&D price their products to develop high margins. It's a matter of economics - there has to be an ongoing flow of income sufficient to fund R&D going forward. R&D expenses are incurred before the product ships, and are expected to be recovered over the life cycle of the product, which may be relatively short due to the pace of technological innovation.

Under the circumstances, I expect the two adversaries to share the server acceleration market, with wins and losses driven by customer performance and price considerations for their specific applications.

Quantifying the Implications for Altera

Daane's $1 billion is a nice round number, and a long-term one at that. Much of the R&D expense has been incurred, so a large part of any revenue that is generated will make its way to the bottom line.

Here's how I do the math: Assuming normal revenue growth of 3% annually, I add $1 billion at the end of five years. To the resulting revenue, I apply 37% operating margin, equal to the 2nd highest level of the past 10 years. I deduct taxes at 12.5%, which will be less if research credits continue available as they were in the past. I assume share counts decrease 1.7% annually, the ten year average, which would fully utilize the 22 million shares remaining on the buyback authorization as of 9/30/2014.

Developing EPS of $3.97, I apply a five-year average P/E of 20.2 (per Morningstar) and arrive at a share price of $80, five years hence. From recent prices in the $37 area, capital appreciation would be 16.7% annually, to which one could add the dividend, currently yielding 1.7%.

A Digression on Long-term Projections

When a projection relies on the cumulative effect of multiple assumptions over a long period of time, the odds of an accurate prediction decrease at an exponential clip. It's tempting to round all the estimates down, just to be on the safe side. Of course the result is mediocre.

If a projection relies on the successive application of ten parameters, and you round them all down, you are guaranteed a low-ball estimate.

After working with situations of this type for many years, I advocate using one's best guess at every step along the way, then examining the result for credibility.

So, I take the 18.4% (including the dividend) projected return, and I cut it in half. 9.2% looks fairly good, compared to what one could logically expect from a stock market that appears very fully valued. And, noting that all the underlying assumptions have a historical precedent, there is a non-zero probability that the higher figure will be closer to the ultimate reality.

Strategy and Tactics

I still consider Altera suitable for dividend growth or buy, hold and monitor investors who are willing to place an emphasis on capital appreciation.

Altera has shown a tendency to make substantial moves based on news that is not directly relevant to its future earnings potential. It dipped markedly when Microchip Technology (NASDAQ:MCHP) made a groundless call for a down cycle in semiconductors, and also dipped on an earnings miss from Xilinx, which was driven by unrealistic guidance. Both moves represented short-term buying opportunities.

The company has a substantial repurchase authorization in place and states that the activity will be pursued on an opportunistic basis. Checking the most recent quarter, they paid an average $34.33, somewhat less than the market, which was $34.81 based on daily closings. I anticipate that there will be support around that level, in the event shares decline.

Buy the dip should be workable here. I have a nice profit on a position taken early this year with deep in the money LEAPS, and regard the strategy as attractive here in that it provides low cost leverage on a situation that has a large but ill-defined upward potential.

If shares make their way back down to the $34 area, I plan to do something speculative with a bullish risk reversal.

Disclosure: The author is long ALTR, INTC.

The author wrote this article themselves, and it expresses their own opinions. The author is not receiving compensation for it. The author has no business relationship with any company whose stock is mentioned in this article.