Editor's Note: This is the third in a three-part series on the colossal changes taking place in the IT world.
The tech world is continually undergoing extensive changes, with dominant incumbents ceding share to more-specialized niche players.
This dynamic recently played out for NVIDIA (NVDA 0.35%), which became one of tech's best-performing stocks over the past few years after its graphics processing units (GPUs) offered a more efficient way for companies of all types to do their computing.
But even well-loved NVIDIA is now under attack. The stock fell more than 50% in the final three months of 2018, with weak fourth-quarter guidance or slowing demand in China being possible reasons for Wall Street analysts tapering down their expectations.
I agree that NVIDIA's troubles will continue, but for a different reason. One of NVIDIA's most important growth markets is the cloud computing data center. Yet I believe field programmable gate arrays (FPGAs) will win out over GPUs in an industry that is quickly becoming a must-win battleground for NVIDIA. If that happens, the company's best growth days may now be behind it.
But what could take its place?
I believe Xilinx (XLNX) might be the stock to own in the tech space today because the company is extremely well positioned to capitalize on the industry's recent shifts. And that could mean big returns for investors.
Check out the latest Xilinx earnings call transcript.
The age of the FPGA
FPGAs are chips that can have their logic programming continually changed, meaning they are adaptable to changing software requirements. This makes them a perfect fit for machine learning, which is constantly training and retraining its artificial intelligence (AI) algorithms in order to optimize performance.
Xilinx invented FPGA technology three decades ago, and it remains one of only two large companies focusing on it today. The other is Intel (INTC 0.02%), which bought its way into FPGAs through its 2015 acquisition of Altera for $16.7 billion.
A price tag that hefty should immediately catch your attention. When the world's second-largest chipmaker shells out that kind of cash for a technology outside of its core competency, it's typically a sign that it is important.
Intel is a massive company with a huge R&D budget, and one might even think that it could become the most formidable player in FPGAs. But the computing behemoth also has several other product lines competing to win share in data centers. It's quite likely that Altera simply isn't large enough, even internally within Intel, to hold its own against the marketing muscle of Intel's other products.
On the contrary, Xilinx is laser-focused on FPGAs. It has designed and optimized the hardware, and now it's working on developing an ecosystem to make its FPGAs easily usable by developers.
Crossing the chasm
The success of that ecosystem -- called the Adaptive Compute Acceleration Platform (ACAP) -- will be incredibly important. Xilinx is working to perfect the fit between applications and hardware through what it describes as "the most dynamic processing technology in the industry." For large cloud vendors like Amazon.com (AMZN 0.23%), with its Web Services division (a Xilinx customer), that could mean maximizing application performance and minimizing data-center power costs.
The ACAP will also be the key for FPGAs gaining mass-market adoption. Software developers don't care what's going on behind the scenes with Amazon's cloud computing hardware. But they do want a lower monthly bill to rent processing and storage, as well as to have an easy-to-use platform that will allow their apps to work with any of the cloud vendors.
Technically, this won't be easy. There's a ton of programming that will be needed to build out Xilinx's invisible layer of abstraction, which is what would allow those developers to focus exclusively on the software and not on the hardware.
But the optimized performance of FPGAs might make it worth it. FPGAs can be significantly more efficient and are more flexible than GPUs, so the long-term data center opportunity is really Xilinx's for the taking. If it can successfully execute on its strategy, it could transform itself from being a niche hardware company to a mass-market optimizer of cloud computing performance.
Let's take a moment to quantify what that data center opportunity could be worth.
Data centers aren't even close to being Xilinx's largest market today. They currently account for around $600 million of annual sales, representing around a fifth of the overall top line.
Xilinx sees the combination of all of the markets it directly serves growing at 13% annually, rising from $14 billion in fiscal 2019 to $23 billion in 2023. That is respectable growth, but not outstanding.
But it believes the opportunity within data centers -- specifically programmable logic hardware devices used in public clouds -- will increase much more significantly. IHS Markit (an information services firm) and Xilinx together estimate this niche market will grow at 67% per year, from $600 million in fiscal 2019 to $4.6 billion in 2023.
Xilinx will be a prime contender to take advantage of this massive industry tailwind. Its "data center first" strategy looks to displace GPUs and application-specific integrated circuits at the largest cloud vendors.
As noted above, FPGAs are an oligopoly, and Xilinx reportedly had 59% global market share at the end of fiscal 2017 (Intel held the other 41%). If that market share split remains true in data centers, the segment could contribute $2.7 billion to Xilinx's fiscal 2023 revenue.
And to put that number into context, Xilinx did $2.5 billion of total revenue in all of its markets combined last year. Data center revenue alone could be greater than that within the next four years.
That estimate might initially sound overzealous, but the company is already off to a great start. Xilinx recently reported that its core product sales in data centers (excluding cryptocurrency mining) have already nearly doubled over last year. There appears to be a long growth runway, and Xilinx is accelerating on it.
The bottom line
Cloud and AI services are both still in their extremely early stages. Companies are learning more every day about how to use new software languages to write AI algorithms. The businesses that come to understand and embrace AI will mine insights and develop strong competitive advantages over their peers. They'll become the trailblazers of a new AI era.
But behind the scenes, the true workhorses of this AI movement will be the data center hardware chips, which are tirelessly crunching all of the data used to mine those insights. They'll need to process instantaneously and with extremely low latency to provide real-time decision making. That will initially appeal to consumer-facing applications like Alexa, where users want immediate responses without waiting for a lag. But if we play this trend forward, we'll soon begin to see more-demanding and higher-dollar applications move to the cloud. The advantages of FPGAs will find even more opportunities.
Xilinx is the pure-play enabler of this trend, and is perfectly poised to capitalize on the adoption of FPGAs into data centers. I believe that Xilinx's stock has a long growth runway, and that forward-thinking investors should consider placing it on their radar.