Mellanox is facing an onslaught from well-resourced competitors that is at a nascent phase. Most notably, Intel has committed significant resources to its Omni-path InfiniBand offering which competes directly with and is taking market share from MLNX’s highly profitable InfiniBand offering, especially in the high performance computing end markets (HPC). When Intel brings its Skylake platform with integrated Omni-path to market next year, pricing pressure, which is already intense, will likely get much worse. Separately, Cavium has become increasingly competitive in its Ethernet offerings, Mellanox’s other large product area. In the past few months, Cavium announced that they were acquiring QLogic.
Business and Financial Summary
In short, Mellanox provides interconnect products to move data between computing nodes and storage networks where performance is a key consideration. Think moving data around in high performance computing, data centers etc. Mellanox supplies both Ethernet and InfiniBand. Ethernet products represent ~40% of revenue up from ~20% a year ago. InfiniBand products represent ~55% of revenue down from ~70% a year ago. The balance of revenue does not warrant discussion. Ethernet products face greater competition, but are less critical to a short thesis. Mellanox had 85-90% market share in its InfiniBand offerings up until very recently (2016). InfiniBand’s attractiveness lies in its high bandwidth and low latency. Given that interconnects often serve as one of the largest bottlenecks, customers with very demanding environments and applications (HPC, cloud, big data etc) often opt for InfiniBand. These buzzword applications are one way in which Mellanox defines its opportunity set. One key feature, that is also a big point of debate when comparing MLNX to InfiniBand-like competitors, is the ability of its offerings to “off-load” processing. In short, processing that might otherwise be done by a CPU when using a different interconnect solution is off-loaded to the network hardware. Stated differently, Mellanox’s InfiniBand solutions allow remote direct memory access.
Note: Estimates in this table are consensus
Intel Competitive Threat
Why is Intel interested? Interconnects are one of the biggest if not the biggest bottlenecks in HPC and storage performance. As a result, those customers, who place a relatively higher value on performance over price, are willing to pay a premium to reduce interconnect bottlenecks. High end HPC clients, in particular, opt to pay a premium. In some implementations, interconnects can eat 30% of a total budget. This means less money for Intel processors. Intel has a long history of supporting the commoditization of hardware from other vendors. I believe that interconnects are high on Intel’s list of products for which it would like customers to have better price to performance options.
What is Intel doing? Intel acquired interconnect assets from QLogic and Cray 4-5 years ago. They built on top of these acquisitions to bring Omni-path to market earlier this year (Note: Skeptics are dismissive of the immediate threat because of this lag and Intel’s “history of missing target deadlines.”). INTC has been pricing its’ offerings at substantial discounts (30-50%) to Mellanox competitive products. Also, Intel has announced that new versions of its Skylake platform will have Omni-path integrated. These are schedule to ship Q2 2017. Additionally, there are rumors (https://semiaccurate.com/2016/11/17/intel-preferentially-offers-two-customers-skylake-xeon-cpus/) that Intel has agreed to ship these to Google and Facebook prior to a broad launch.
How has Intel done? Intel introduced their Omni-path offerings in February of 2016. On the Company’s Q2 2016 call, they claimed to have achieved 30% market share in 100G (higher end, higher bandwidth) where it is consensus that MLNX has a greater advantage. Intel jumped from 1.6% system share in the top 500 HPCs in June 2016 to 5.6% in November of 2016 according to statistics supplied by top500.org (https://www.top500.org/statistics/list/), a site that tracks statistics on the 500 most powerful computers that are known about in the public domain. I suspect an integrated delivery will accelerate competition.
Mellanox Response Mellanox has been dismissive of Omni-path for some time. The Company points to two advantages: 1. Introduction of 200G soon; 2. Have more messages off-loaded. The first is largely inconsequential as customers are more sensitive to latency not throughput. Regarding #2, Intel has taken a mixed approach to off vs on loading, although they acknowledge some off-loading is beneficial. Given the customer adoption and anecdotal performance comparison feedback, I think that this advantage is at best overstated. With regard to their public posturing, shortly after Omni-path’s introduction, MLNX began publishing articles and “studies” dismissing or outright trashing Omni-path. A MLNX employee, published an article (https://www.hpcwire.com/2016/04/12/interconnect-offloading-versus-onloading/) in HPC Wire (an industry rag) which included performance benchmark results “proving” MLNX was superior. Two weeks later, the same industry rag published an article (http://www.hpcwire.com/2016/04/26/omni-path-steadily-gaining-market-traction-says-intel/), by one of the editors, reporting that Omni-path was gaining traction. Within two days, there was a rebuttal (https://www.mellanox.com/blog/2016/04/a-look-at-the-latest-omni-path-claims/) on MLNX’s Company blog. Methinks thou dost protest too much.
Ethernet products account for a material proportion of MLNX’s revenue. However, I do not think that this segment warrants as much attention. 1. InfiniBand is higher margin (I have heard estimates that this product portfolio accounts for the vast majority of profitability). 2. Recent Ethernet results were flattered by a large Microsoft order where MLNX was the only qualified vendor. 3. Cavium just recently closed on its acquisition of QLogic, making that offering much more competitive going forward. 4. For the short to work, one doesn’t need the Ethernet portfolio to come under pressure, although such an outcome is not unthinkable.
Market Expectations and Valuation
It appears to me that both the buyside and sellside are relatively sanguine about the Company’s prospects. The short interest is a relatively modest (for a tech company in this competitive situation) ~5% of the float. The sellside acknowledges the threat that Intel poses, and has decelerating growth overall and, in some cases, declining revenue in HPC. Yet, in nearly all sellside forecasts, gross profit is expected to increase. I believe that MLNX will face a choice to either dramatically cut the price of or face substantial volume declines in its most profitable product offering, InfiniBand. Either way, I expect gross profits to decline. MLNX is trading at 19x forward EBIT making the following assumptions: 1. Current InfiniBand gross margins are 80%; 2. Future InfiniBand margins get cut to 55% (which could imply a ~30% price cut ceteris paribus); 3. Sellside revenue expectations are spot on at ~$970MM. I don’t think the implied multiple is appropriate for a business with these competitive headwinds. With a more appropriate 12x multiple, which wasn’t chosen by any rigorous methodology, the stock would have 35%+ downside. Once competition intensifies, I believe that the Company’s performance and stock will languish. Also, I believe that the range of assumptions with very bad outcomes for the stock is wide enough to drive a truck through.
Back of the Envelope Math
I do not hold a position with the issuer such as employment, directorship, or consultancy. I and/or others I advise do not hold a material investment in the issuer's securities.
Continued erosion of market share, price cuts, or both in the face of pressure from Intel.