After a decade of vendor consolidation that saw some of the world’s biggest IT firms acquire first-class HPC providers such as SGI, Cray, and Sun Microsystems, as well as smaller players like Penguin Computing, WhamCloud, Appro, and Isilon, it is natural to wonder who is next. Or maybe, more to the point, who is left?
As it turns out, there are still plenty of companies, large and small, that can fill critical holes in the product portfolios of HPC providers, or those who want to be HPC players. These niche acquisitions will be especially important to these same providers as they expand into HPC-adjacent markets such as artificial intelligence, data analytics and edge computing.
One company that can play into all of these markets is FPGA-maker Xilinx. Since Intel acquired Altera in 2015, Xilinx is the only standalone company of any size that makes reconfigurable logic devices. Give that, the natural buyer for Xilinx would be AMD, Intel’s arch-nemesis. AMD, of course, already has a highly competitive lineup of CPUs and GPUs to challenge its much larger rival, and the addition of an FPGA portfolio would open a third front. It would also provide AMD entry into a whole array of new application markets where FPGAs operate: ASIC prototyping, IoT, embedded aerospace/automotive, 5G communications, AI inference, database acceleration, and computational storage, to name a few.
The only problem is that Xilinx’s current market cap of around $25 billion, or about half the current market cap of AMD. And if you’re wondering about AMD’s piggy bank, the chipmaker has $1.2 billion cash on hand as of September 2019. Which means any deal would probably take the form of a merger rather than a straight acquisition. There’s nothing wrong with that, but a merger is a more complex decision and has greater ramifications for both parties. That’s why the rumors of a Xilinx acquisition have tended to center on larger semiconductor manufacturers that might be looking to diversify their offerings, like Broadcom or Qualcomm. Those acquisitions wouldn’t offer the HPC and AI technology synergies that AMD could provide, but they would likely be easier to execute.
Another area that continues to be ripe for acquisitions is the storage market. In HPC, Panasas and DataDirect Networks stand alone – well, stand together – as the two HPC specialists left in the market. And of those two, the more modest-sized Panasas would be easier to swallow. But most HPC OEMs, including the biggies like Hewlett Packard Enterprise, Dell Technologies, and Lenovo already have their own HPC storage and file system offerings of one sort or another, although Lenovo is probably most deficient in this regard. For what it’s worth though, Panasas, which has been around since 1999, has never attracted the kind of suitor willing to fold the company’s rather specialized parallel file system technologies into its own product portfolio. In all honesty, we don’t expect that to change.
The real storage action in the coming years in HPC, as well as in the enterprise and the cloud, is going to be in the software defined space, where companies like WekaIO, VAST Data, Excelero, and DataCore Software have built products that can virtualize all sorts of hardware. That’s because the way storage is being used and deployed in the datacenter these days is being transformed by cheaper capacity (disks) and cheaper IOPS (NVM-Express and other SSD devices), the availability of cloud storage, and the inverse trends of disaggregation and hyperconvergence.
As we noted last July: “While there are plenty of NAS and SAN appliances being sold into the enterprise to support legacy applications, modern storage tends to be either disaggregated – with compute and storage broken free of each other at the hardware level but glued together on the fly with software to look local – or hyperconverged – with the compute and block storage virtualized and running on the same physical server clusters and atop the same server virtualization hypervisors.”
Any of the aforementioned SDS companies, along with others, may find themselves courted by OEMs and storage-makers, and even cloud providers. DDN has been busy in that regard, having acquired software-defined storage maker Nexenta in May 2019. We expect to see more of such deals in the coming years. Besides DDN, other storage companies like NetApp should be looking hard at bringing more SDS in-house. The big cloud providers – Amazon, Microsoft, Google, and so on – will also be making some big investments in SDS technologies, even if they’re not buying such companies outright.
One market that is nowhere near the consolidation stage is quantum computing. However, that doesn’t mean companies won’t be looking to acquire some promising startups in this area, even at this early stage. While major tech firms such as IBM, Google, Intel, Fujitsu, Microsoft, and Baidu have already invested a lot on in-house development and are busy selecting technology partners, other companies have taken a more wait-and-see approach.
In the latter category, one that particularly stands out is HPE. In this case, the company is more focused on near-term R&D, like memristors or other memory-centric technologies. While there may be some logic in letting other companies spend their money figuring out the most promising approaches for quantum computing, and then swoop in and copy (or buy) whatever technology is most viable, there is also the risk of being left behind. That’s something HPE cannot afford.
That said, HPE has recently invested in IonQ, a promising quantum computing startup that has built workable prototype using ion trap technology. The investment was provided via Pathfinder, HPE’s investment arm. In an internal blog post on the subject penned by Abhishek Shukla, managing director of global venture investments, and Ray Beausoleil, Senior Fellow of large scale integrated photonics, the authors extol the virtues of IonQ’s technical approach:
“IonQ’s technology has already surpassed all other quantum computers now available, demonstrating the largest number of usable qubits in the market. Its gate fidelity, which measures the accuracy of logical operations, is greater than 98 percent for both one-qubit and two-qubit operations, meaning it can handle longer calculations than other commercial quantum computers. We believe IonQ’s qubits and methodology are of such high quality, they will be able to scale to 100 qubits (and 10,000 gate operations) without needing any error correction.”
As far as we can tell, HPE has no plans to acquire the company (and it shares investment in the startup with other companies, including Amazon, Google, and Samsung, among others). But if HPE is truly convinced IonQ is the path forward, it would make sense to pull the acquisition trigger sooner rather than later.
We have no illusions that any of this comes to pass in 2020 – or ever. As logical as the deals we have suggested seem to us, the world of acquisitions and mergers is a lot more mysterious and counterintuitive than we’d like to admit (cases in point: Intel buying Whamcloud or essentially buying Cloudera through such heavy investment). More certain is the fact that these deals will continue to reshape the HPC vendor landscape in the coming decade as companies go after new markets and consolidate their hold on old ones. If anything, the number of businesses bought and sold will increase as high performance computing, driven by AI and analytics, will extend into more application domains. Or as the Greeks put it more succinctly, “the only constant is change.”