It is hard to make a profit selling hardware to supercomputing centers, hyperscalers, and cloud builders, all of whom demand the highest performance at the lowest prices. But in the first quarter of this year, network chip, adapter, switch, and cable supplier Mellanox Technologies – which has products aimed at all three of these segments – managed to do it.
And with activist investor, Starboard Value, pressing Mellanox to make the kinds of profits that other networking companies command, the swing to a very decent net income could not have come at a better time. Starboard has been on the case of the upper management at Mellanox since last fall, basically saying that the company, which sells a variety of InfiniBand and Ethernet switching elements aimed at those who demand high bandwidth and low latency, should be able to bring more money to the bottom line.
This is far easier said than done, particularly in the HPC, hyperscale, and cloud markets that Mellanox and its (relatively few) peers serve.
What Starboard doesn’t understand is the large amount of investment it takes to create products for these customers, and the fact that they are not high volume products that, quite frankly, have a longer shelf life and are sold to millions of enterprises that pay a much higher price per unit of capacity and therefore are more profitable. As we have pointed out before, it is hard to find a tougher set of customers and economic conditions than in the upper echelons of computing and networking. Ask IBM, Cray (eaten by Tera Computer), the former SGI (eaten by Hewlett Packard Enterprise after being eaten by Rackable Systems), and even Intel’s HPC division (itself a product the acquisition of interconnect businesses bought from QLogic and Cray plus some exotic Knights processors) and they will tell you all about it. Ask the myriad supercomputer makers who no longer exist how tough this market is – but you might need a Ouija board tuned to the corporate stations.
This is a very demanding part of the IT market, and one that we need someone to address so technologies will evolve and eventually be more broadly distributed. These companies represent an investment in the future, but it is far larger than can be gauged by share price or quarterly net income. We may all wish this to be otherwise, of course. But this is the reality of this part of the IT sector.
We think that Mellanox has done a remarkable job in growing its addressable market in the past decade. Were it not for its investment in Ethernet technologies, the company would be hard-pressed to live on its InfiniBand networking business alone, even with the increasing adoption of InfiniBand for database and storage clustering and for certain AI training workloads. The company’s expansive strategy has had risks, to be sure, but buying Voltaire, a rival in switching and customer of its InfiniBand switch ASICs, for $218 million back in November 2010 while at the same time getting into the crowded Ethernet arena has paid off. Spending another $129.5 million in July 2013 for IPtronics, an optical interconnect component supplier, and Kotura, a maker of photo detectors and modulators, has made its LinkX cable business, which is growing fast. The acquisition of EZchip Semiconductor in September 2015 for $811 million was expensive, but it gives Mellanox good massively parallel compute technology with its “Bluefield” processors that it can bring to bear on software-defined networking and further offload of functions from CPUs. Mellanox could have played it safe, but it keeps pushing into new and demanding markets and into a broader total addressable market.
In the first quarter ended in March, Mellanox posted sales of $251 million, an increase of 33 percent over the year ago period. This time last year, Mellanox booked a $12.2 million loss, and thanks in large part to the revenue boost as well as some changes to the way revenues are booked and some cost cutting, the company swung to a profit of $37.8 million. This profit level is consistent with past times when the company was reaping the benefits of past investments and new technologies were being acquired by customers. The company exited the quarter with $286 million in cash and investments, which gives it a cash cushion as well as a means to do acquisitions if it sees fit.
The company’s 100 Gb/sec EDR InfiniBand products accounted for $55.7 million in sales in the quarter, up 41.2 percent year on year. Total InfiniBand product sales came to $103.1 million, and we estimate from past trends that 56 Gb/sec FDR InfiniBand still accounted for $44.4 million, down 14 percent, and even the old 20 Gb/sec DDR InfiniBand brought in a few million bucks.
Ethernet products drove $136.9 million in sales, up a very impressive 70.2 percent, and one of the big drives was the adoption of 25 Gb/sec and 50 Gb/sec ConnectX-5 adapters by hyperscalers and cloud builders. Ethernet has been driving more revenues than InfiniBand since the fourth quarter of last year, and we think it will do this forever more at Mellanox.
We would go so far to say that eventually, despite the continuing use of InfiniBand for HPC, database, storage, and AI clusters where both low latency and high bandwidth are important, for some workloads bandwidth and compatibility with Ethernet are the driving issues and having four or five times lower the latency is not going to appreciably affect the application performance, and thus Ethernet is something Mellanox has to sell.
Mellanox makes InfiniBand and Ethernet switch chips as well as adapter chips for use in its own devices, but it also sells the chips raw to OEMs that want to make products to sell with their own brands on them and to ODMs that make products on behalf of cloud builders and hyperscalers that don’t want to buy OEM products because they are overstuffed with things they don’t need. In the first quarter, chip sales at Mellanox were down 35 percent to $28 million, but this is a very choppy business by the very nature who the customers are and the way they acquire technology. So is the Mellanox business of selling completed board products, such as adapter cards, and it just so happens there is an upgrade cycle going on here – one that Mellanox started investing in back in 2014 when it helps pushed the 25G Ethernet standard that the IEEE though it could ignore – and this business was up 83.9 percent to $118 million. The finished switch business, including both InfiniBand and Ethernet products, rose by 17.1 percent to $55.2 million, and that left other products, dominated by LinkX cables but also including fabric software and other items, with $50.2 million in sales, up 43.8 percent.
As the chart above shows, each of these segments has their own upgrade cycle, and sometimes they will all hit at once and Mellanox will rake in the dough, and sometimes they will all slow at once and it will not be as much fun. But that is the way networking for the world’s toughest customers goes, and in a sense, we all should be thankful that Mellanox plays the game at all. Just like we should be thankful for all of the investments that companies like IBM, Cray, Intel, HPE, Fujitsu, and a handful of others actually make in high performance networking, compute, and storage.
Someone has to do it.
One last thing. Eyal Waldman, co-founder and chief executive officer at Mellanox, who has been steering this company since it was founded in 1999, said on a call this week with Wall Street analysts going over the numbers that the company was projecting that sales would break through $1 billion this year, which is a little more than 20 percent growth at the midpoint of the forecast. This guidance is admittedly conservative because of the nature of those high-end customers that buy its products. But this will be a remarkable thing when it comes to pass, considering that Mellanox was nine times smaller than this as we were in the gaping maw of the Great Recession and InfiniBand was ten years old and a niche product very few had heard about outside of some HPC centers. It is not a coincidence that Mellanox stock has nearly doubled in the past two years and its market capitalization is in excess of $4 billion. It is hard to do better than this, all things considered.