Yakov Shulman - Chief Financial Officer, Principal Accounting Officer and Corporate Controller
Joseph Wolf - Barclays Capital, Research Division
Mellanox Technologies, Ltd. (MLNX) Barclays Global Technology, Media and Telecommunications Conference May 22, 2013 10:55 AM ET
Joseph Wolf - Barclays Capital, Research Division
[Audio Gap] Shulman, CFO of Mellanox.
Thank you, Joseph. Good morning. My name is Jacob Shulman, I'm the Chief Financial Officer from Mellanox. We have the presentation.
Okay. So we have the presentation. It's not working. Okay. All right. So what is Mellanox? Mellanox is a fabless semiconductor company. We design, manufacture and sell high-performance interconnect products that help to facilitate data transmission between servers, storage and systems. We're an Israel company, we duly [ph] headquartered. Our R&D operation is located in Israel. Our business departments, business people located in Sunnyvale, California. We have approximately 1,200 employees worldwide.
2012 was a great year for us. Our revenue increased 93% from 2011 and we reached the level of $500 million. Our Q1 '13 revenue was $83.1 million, and we guided Q2 for $92.5 million to $97.5 million.
So what do we do? Each compute and storage systems consists of 3 major blocks: CPU, memory and interconnect, so we do interconnect. That's a pipe that feeds CPU and memory with the data. Over years, CPU vendors introduced more and more powerful CPUs. Memory became faster and faster. That puts the focus on the interconnect, because if interconnect cannot feed data fast enough to CPU and memory, then the system is unbalanced and not utilized fully. So the name of the game is bandwidth and latency. Bandwidth is the amount of data we can move, and latency is how fast you can do that.
So when we first came into the market in 2002 we introduced 10 gigabit per second products then the market progressed to 20, 40 and in 2011, we launched our latest generation products, 52 gigabit per second products. Today, we're the only game in town for 53 gigabit per second InfiniBand products. We also offer 10, 40 and 56 gigabit per second Ethernet products. Majority of our revenues today come from InfiniBand.
Historically, Mellanox has been viewed as HPC player, but things have certainly changed since then. Today, we offer our products in Web 2.0, cloud, database and storage markets, as well as financial services. And there are different values that we had provide to the customers in those markets.
In high-performance computing, InfiniBand today is the most used interconnect on a TOP500 list. Every 6 months, a list of TOP500 computers is published, and InfiniBand today connect 224 systems of the TOP500, which is 45%.
Our latest generation FDR products today represent 45 systems of that, 224 systems connected is InfiniBand. And the great example of how our products used in high-performance computing is the Purdue University. So the university was able to build supercomputer with 187 teraflops using only 648 servers. This is the half server count and double performance from the prior cluster.
In Web 2.0 market, the multiple examples. We work today with 5 major Web 2.0 players. And the way we define Web 2.0 as opposed to cloud is when people build supercomputer to run one application, we call them Web 2.0 guys. When people build a supercomputer running multiple applications, that's cloud for us. So one of the great examples on how our products use in Web 2.0 market is Bing Maps. So when Microsoft was built in data center for the Bing Maps application, they were considering 2 possible options: one is using 10 gig Ethernet, and the other one using our 40 gig InfiniBand solution. At the end, they decided to go with InfiniBand. So they were able to achieve 10x performance from the prior cluster, but also they were able to do that at half cost compared to 10 gigabit Ethernet solution.
In the data center market, Oracle put us on the map. Today, Oracle uses our InfiniBand in all of their data, Exadata, Exa family. Really, every box that ship to corporate world today to top banks, top financial institutions, top corporations includes Mellanox and InfiniBand. So using our technology, Oracle customers able to reduce hardware costs by factor of 5x, and on the other hand, to improve performance by a factor of 10x.
In storage market, we think it's a greenfield for us. We really believe that InfiniBand and Ethernet will replace Fibre Channel in storage market. If you look today, we have a multiple design wins in that market. So at our Analyst Day that we held back in October of last year, the emcee said that we have half a dozen programs with them. Some of them went into production like Isilon, some of them expected to go into production in the late part of 2013, 2014 time frame. Teradata recently announced that they're using our InfiniBand for their future development. So, so far they introduced as their [ph] family products as well their 6700 series products using InfiniBand. The other examples of storage design wins like Oracle ZFS, IBM XIV and others. Storage is a big market, and we think we'll have a very good opportunity to play in that market.
The way we estimate our total available market is by the number of endpoints sold annually. We can connect servers, storage using both InfiniBand or Ethernet by protocol. So in total, we estimate that approximately $14.6 million of endpoints sold annually in the market, and that comprise of 10.6 million of servers, 3.4 million of endpoints in storage and 600 came in embedded, and then we apply our ASP per endpoint of $400. That's how we derived to $5.8 billion total available market for us in 2013. And the good thing is that we expect this market to grow. We expect that in 2015, approximately 16.8 million endpoints will be sold. And even if we assume 10% price erosion in our ASP per endpoint, we still get into approximately $6 billion total available market.
I think in the past, for 2012, we said that we shipped approximately 1.1 million devices. And excluding some of the switch points, assuming million devices for endpoint with $500 million revenues, our ASP today per endpoint is significantly higher than assumed in this -- than the assumed $400 per endpoint.
Who do we sell to? We primarily sell to big, large server OEMs, HP, IBM have historically been primary customers for us. In Q1 of 2013, they represented approximately 35% of the business. Oracle, Oracle has approximately slightly less than 10% ownership in the company, that's why we disclose also revenues from Oracle. Oracle, historically, has been representing mid single digits of the revenues. So we sell through OEMs. Sometimes we could -- we work with end customers, especially big end customers in Web and cloud markets sometime [ph], as you can see, to create demand, to design the infrastructure, to design the architecture, but the fulfillment would still go through OEMs. Today, OEMs represent approximately 90% of our revenues. The other 10% come from distribution. And we work with multiple software partners to ensure compatibility with different operating systems.
Really, we provide end-to-end solution. We're the only game in town for full end-to-end 56 gigabit per second InfiniBand. We're the only game in town for 40 and 56 gigabit Ethernet. Today, people have some switches at 40 gig Ethernet. No one has 40 gig endpoints on Ethernet.
And our gross margins, we sell different product categories here. We sell silicon, we sell adapter cards, switches, sell cables and kind of software, which is a performance booster. And our gross margins, dependent on the product mix. We garnish [ph] highest margins from selling silicon, followed by boards and switches and cables. So the more silicon we sell, the high gross margins. The more cables we sell, the more [ph] gross margins.
Last week we announced acquisition of Kotura. It's a very strategic acquisition for us. Kotura was founded in 2004 with first revenues in 2006. So as we move to higher speed, the need for having end-to-end solution in business [ph] really one -- every gig [ph] of second count, you really have to have full control of the end-to-end solution.
Also materials that used today will have some limitations in terms of -- just from the physics, from the physics perspective. For example, copper will have its limitations on their reach, both from InfiniBand or Ethernet. It looks like no one will be able to use copper for reaches beyond 3 meters. So new technologies would have to be introduced, and fiber optic is the technology that could take longer reach.
Today, there are 2 major technologies in fiber optics world. One, is VCSEL and the other one is silicon photonics. VCSEL is working today at 50 gigabit per second. It will work at 100 gig with challenges. We don't think that VCSEL could work beyond 100 gig per second, that's why silicon photonics is becoming very strategic to us. Some of our competitors have these capabilities. Cisco acquired Lightwire last year, and Intel demonstrated some silicon photonics capabilities developed in-house and now Mellanox also have these capabilities.
Also the second reason for the acquisition is, by buying -- by having this technology in-house, we'll be able to eliminate margin stacking. So as we move to higher speeds, the attach rate of our cables and modules is expected to increase. We see that for 56 gigabit per second, attach rate for our cables is much higher than for [ph] 40. And we expect that when we move to 100 gig per second, our attach rate will be even higher. So by elimination margins taken by buying directly from the fab, we will be able to add hundreds of dollars per endpoint to our bottom line, and that will become significant input on our gross margins. We will be able to improve materially gross margins for our cable product family.
In terms of the roadmap, as I said, we're the only game in town for 56 gigabit per second on InfiniBand side. We're currently working on the 100 gig per second solutions. We expect to introduce 100 gig solution in 2014, 2015 time frame. Our closest competition is Intel. Intel acquired 4 companies to compete with us. Intel acquired NetEffect for the mail with [ph] Ethernet, they acquired Fulcrum for Ethernet switching, they acquired QLogic InfiniBand assets, as well as Cray interconnect. To date, Intel offers 40 gig solution on InfiniBand side. We hear that they decided to skip 56 gigabit per second generation and will try to intercept us at 100 gig. So we expect to introduce our full end-to-end 100 gigabit solution sometime in 2014, 2015 time frame. We think that Intel is behind us. We estimate that they will launch their solution in 2015 time frame.
On a financial side, really, Mellanox is a growth story. 2012 was exceptional year for us. As I said, our revenues increased 93% year-over-year. But over 5 years, our CAGR was approximately 43%. People today look for growth stories, and Mellanox is definitely one of them. On a quarterly breakdown, really, 2012 was impacted by large unique deals, primarily associated with our own reverse [ph] cycle. So in Q1, we said that approximately $10 million were Romley-related; Q2, approximately $30 million; and Q3, $25 million Romley-related. So in total, 2012, $65 million pent-up demand associated with Romley refresh [ph].
Our revenue in Q4 was below the guidance we initially provided. We guided $145 million to $150 million. Our actual revenues were $122 million. We noted 2 issues that caused for our revenues to be below the estimations: one was the cabling issue, which was completely resolved in Q4; and the other one was the build-out -- one of the large OEMs -- buildout -- inventory build-out of one of the large OEMs, which we estimated at $30 million. In Q1, less than half of that inventory was [indiscernible] into the market, and we expect that it will take few more quarters for the inventory to be fully depleted. Our guidance for Q2 is $92.5 million to $97.5 million. We did say that we expect to grow quarter-o-quarter [ph] throughout the 2013.
This slide represents kind of transitions that the company experienced throughout last few years. Back in 2009, 2008, where we're mostly component vendor, and more than 80%, 86% of our revenues came from sales of silicon and boards. In 2011, we acquired Voltaire, and since then, the silicon and boards represent only 50% of our revenues. The other 50% come from selling switches and cables.
And on the data rate side, really, FDR is our newest generation, a product that was launched back in 2011 and ramped up much faster than QDR, our previous generation. Today, or in 2012, FDR represented 47%. In Q1 of '13, FDR represented approximately 50% of our revenues, and we believe that there is room to grow for FDR product as a portion of our revenues. Usually we will charge premium for the latest generation of products. So I think our premium on FDR as compared to QDR was 20% to 25% higher in ASP. We also garnished [ph] better gross margins on latest generation products.
We generate healthy amount of cash. The cash generated in 2012 was $182 million. We exited March quarter with $403 million in cash and investments.
In terms of the long-term model, obviously, 2012 was an exceptional year for us. We -- our results were well above our long-term model. We had a huge leverage from the top line. Our operating expenses increased 50%, but our revenue increased 93%, that's why we were able to achieve net income as 31% of the revenues. With the revenue reduction, obviously, our profitability significantly less than the long-term model, and long-term model for us is 2 to 3 years from now. So once we get back on track and increase our revenues quarter-over-quarter, we will see our profitability growing and we still expect to get to mid-20s in operating income within the next few years.
We did say that we want to grow our OpEx at much lower pace than we did in 2012. Really, we don't see any changes in the market. In opposite, we see our products growing in the mainstream markets. There are lot of programs that we're working on with our customers. We're always understaffed. We're always wanting there's [ph] more. We see this future opportunity, that's why we did not reduce our OpEx. But on opposite, we increase our OpEx. We just want to be responsible here and we don't want to grow too much. So if you look at the guidance for Q2, we guided our revenues to increase approximately 17%, we guided for OpEx increase of 10% to 11% and we expect that going forward, our OpEx would grow lower, at lower rate then [ph] the revenue increases.
So just to summarize. Interconnect is very strategic [indiscernible] compute and storage systems, and today, we're playing in multiple markets that emphasize importance of interconnect, HPC, Web 2.0, storage, database, cloud, Big Data markets. Anyone who needs to move data, anyone who needs to analyze data, anyone for whom data is a business needs more performance. That means, they need higher -- bigger parts to move the data, that means Mellanox.
Within silicon photonics win [ph] this group of vendors that have capabilities, and we look forward to introduction of our 100 gig products using this technology. We're uniquely positioned to continue within the market. Today, we provide best interconnect, we have strong partnership with our vendors, we have general relationships with our customers and we provide compelling benefits. We also demonstrated growth of 43% CAGR over the last 5 years, and we're targeting big market. $6 billion market is a huge market for us. Today, we'd penetrate at less than 10% of that market. We still have huge room for growth.
With that, I'll turn to questions.
Jacob, can you just give more [ph] inventory with the customer -- specific customer? And again the last year, if there's inventory issues in [ph] the other customers, and also what the company has done to make sure that you've got a better idea about how that inventories are leaving the customer and that it doesn't get build up again. And then, just as a second question so that I don't waste too much time, if you could go through -- you went through the product segments by speed and seeds [ph], but if you could go through, perhaps, by end market to show what the new penetration of new market is by HPC, cloud and storage.
Okay. So first, on the inventory. As I said, we do not sell to end customers. We fulfill through OEMs. So what happened is that, one of our large OEMs purchased inventory for one of its Web 2.0 guys, one of Web 2.0 customers, and then the Web 2.0 customer decided to stop the program. And that's what created this overcount of the inventory. It's not a unique inventory, it's general inventory that's, currently, that large OEM is selling to other customers. And the pace of which inventory is depleted, really, depends on the product mix within that inventory. Different customers have different architectures that they use in data centers. Therefore, the mix of products required for them is slightly different from that Web 2.0 guy. So the good thing for us is that large OEM is continuing buying from us. But on the other hand, the inventory has been depleted at lower pace. So usually, we thought that majority of the inventory will be depleted during Q1 with some spillovers into Q2. At the end of Q1, we realized that less than half was depleted, and we estimate today that it will take few more quarters for the inventory to be completely depleted. So with that, we introduced different processes, we established different processes to improve our visibility into channel. We have periodic calls with large OEMs to try to identify end customers that they're buying for. We also expand our business team that touch an [ph] end customers to be able to better feel the market, better feel the end customers. So we have better visibility towards the end customer. And on the second question, yes, we don't have hard numbers to share with you what portion of revenues each of this market segments represent. Really, some of the OEMs, they sell to both HPC and Web, and that's why it's very hard for us to exactly know where the products are going. But we have kind of high-level estimate. And in 2012, HPC was a significant portion for us. We said that it represented approximately 55% to 60% of the revenues. Web was 10% to 15% of the revenues, storage was 10% to 15% and database, 10% to 15%. Cloud and financial services were insignificant. We do think that in 2013, cloud will be a growth year for us. We'll grow into cloud, so revenues coming from cloud and market segment will increase. It take longer for cloud guys to implement our technology because of the different additional requirements in the cloud market as opposed to Web 2.0. Again, in Web 2.0, they just build one supercomputer for one application. Cloud is supercomputer for multiple applications. So different issues of security producing in a separate needs [ph] to be resolved, that's why it take longer for cloud.
Any other questions? We have [ph] the next question over there.
The 10 million that you identified, how much -- I'm still trying to figure, how much is that is just pure organic growth of the market? How much of that is just taking market share from Ethernet itself?
So let's take a look at the market segments, and let's kind of focus right now on the servers. 10.6 million endpoints sold in servers than HPC, so 1.9 million. HPC market has grown 8% CAGR. Today, if we look at TOP500 list of supercomputers, Mellanox has approximately 45% market share, all right? So we grew definitely faster in HPC market, taking share from Ethernet and other proprietary interconnect. In the Web market, we just -- we think our penetration is very low, and we expect to grow in that market higher than the currently estimated average market growth of 11%. In enterprise data center, approximately 5.9 million endpoints sold. That market is growing at 1%. Our penetration in that market is very low.
I understand that, but this $5.8 billion market already exist with Ethernet, and you have an advantage over that. So are you saying that lead is going to sustain, and that's how you come with the $5.8 billion? Or is that -- you're going to just take market share from the existing technology? I don't know. Where is that...
We're go -- so in order for us to grow more than 1%, we'll have to take share from existing technologies and technologies that we've taken share. So Ethernet products, our Ethernet products today, we think best products in the market, and our Ethernet products today grow into those markets. So we're increasing share of our Ethernet. We've also taken share of -- from Fibre Channel market, right? So it's mostly taking share from other technologies that we need to go.
Yes, that's what I'm still trying to figure out. How do you take technology, because this technology is already embedded in customer system? So you can only go for use cases where you have a lead, I mean, if...
So for new data centers that build, they use our technology, right? So let's take Web 2.0 guys. They would build data center or use our technology for one certain application. Once they feel comfortable, they would build a new data center using our technology, or expand our technology and implement it for additional applications. So instead of buying, let's say, from the competition, they would go with us. And that's why we see the penetration in those markets.
[indiscernible] on the Ethernet products, are they comparable?
On 40 gig, Ethernet product margin's comparable with InfiniBand. On 10 gigabit, the Ethernet margins are slightly lower than the corporate margins.
Any more questions? Okay. Yes.
I think you said Intel model [ph] 2015, can you just talk about that potential threat a little bit more, and if they do integrate it to the CPU?
All right. So obviously, we do see Intel as primary competition, and we're very concerned. For us, the game plan is to stay ahead of Intel. Today, we're leading by generation. As I said, they acquired 4 different technologies to compete with us, so it's very challenging to combine all of those technologies into one and come up with a product. But Intel has resources and very capable company, so we are concerned. For us, it's a matter of execution. I think we have better technology, and if we execute well, we could take significant market share. We do know that all -- we heard that Intel is planning to combine the interconnect on CPU. We don't know the time frame for that. We think that it would make sense probably to first to come with a discrete solution to see how the market adopts this solution and then combine on CPU. But even when Intel comes with a solution, with a 100 gig solution, we have so many sticky design wins. If you look at multiple design wins we have with Oracle, with Teradata, with IBM, with EMC, when Intel comes with their solution, these customers will have to rewrite some of the code they created around our products, and we'd have to qualify completely the code, right? So it's very sticky designs, and it will take some time for Intel to really become a viable competitor in those markets.
Could you talk maybe a little bit about what the ultimate end applications are at your products, particularly at the very high end you're going into? I mean, given the rate of speed that you're talking about, I mean, how much of that is essentially kind of rapid-fire Wall Street trading programs where very small fractions for a second are critical. How much is certain [ph] depends [ph] applications?
Right. So really, our products today offer highest bandwidth and lowest latencies. So for different applications, different features required. For financial trading, lowest latency is the game, right? As well as for Web 2.0 guys. That's why they buy our products because of the RDMA and some other features that help to them get [ph] the lowest latency, right? On the compute side, really, what matters is the amount of data you move or you could analyze. So we see our products going to HPC markets, where huge amounts of data need to be moved. And since we offer higher bandwidths, that's why our customers buy our products.
Some it's latency. For some, bandwidth is more important.
Is it 50-50 [ph] or is it [indiscernible]?
I don't have that numbers.
You talked -- sorry, you talked about, I guess, working with cloud customers but then fulfilling through your partners. A lot of what we hear in the market these days about the cloud customers is that they assembled these things themselves, they're going direct to sort of the ODMs instead of the OEMs that you referenced. Given you talked about low latency being really important, a lot of these customers are very focused on low latency. So could you just help us understand a little bit more about why you're fulfilling still for these OEMs, they don't seem like they're as big of a player anymore in that field? And could you potentially work with the ODMs, or directly with the cloud guys who are assembling this themselves? Because I think you referenced that you're not really in there right now.
Yes, we do work with ODMs and sometimes directly with end customers in the Web market, it's just very small kind of comparing to OEMs. But we do see that some of the Web guys, those who have better designing capabilities in house, they prefer to buy components and go to ODMs if they want and build their own machines, right? Some of the customers who have less designing capabilities in house, they go for full end-to-end solution. So for us, really, we would work with anyone, right? In different modes. We just, today, ODM model is not as significant as OEM.
Could you just talk a little bit about the storage market? How you see that evolving over the next 2 to 3 years, especially now that you are shipping, I guess, natively [ph] with EMC?
Yes. So the storage market is a big market for us. It's -- it takes very long time to get the design win in storage market. It's very small market. It's usually an HPC design win to take 2 to 3 quarters. Storage market could take up to 3 years or even more. Because if you lose a bit of data on the server, then you just deal [ph] with the server. If you lose a bit of data in storage market, then it's gone forever, right? That's why it's very slow, but it tends to be very sticky. So we have several design wins with EMC. Isilon is a public one. There are some others that are not public. Some of them will go into production in late part of '15, 2014 time frame. Teradata is also becoming a big customer for us. They just announced design wins in late 2013. They also introduced new product family in 2013. I think it's 6700 series. So they -- these customers like IBM XIV and others, they become a significant portion of our revenues, and we expect that to grow in 2014. What's good is it's less lumpy revenue as opposed to HPC and Web market that's [ph] kind of provided in [ph] steady base revenue going forward.
Maybe just -- I don't know, this is maybe sensitive, but [indiscernible] talk in the Israeli press [ph] at least about the shareholder vote and the split of the Chairman and the CEO title. Maybe if you could just give us a little overview of what's going on there?
Yes. So obviously, the dual role is up to the shareholders' vote. We postponed our AGM meetings twice, and the reason being because of the technical issue. Some of the shareholder's votes gets back in tabulation, so that was the reason we postponed it. And we feel that some of them -- our major shareholders do support the dual role for Chairman and the CEO, but it's tough to vote for shareholders.
So some votes back in tabulation. So shareholders thought they voted, but it didn't go through AST, our -- it didn't got through AST, our transfer agent.
Any more questions? All right. Thank you very much.
Copyright policy: All transcripts on this site are the copyright of Seeking Alpha. However, we view them as an important resource for bloggers and journalists, and are excited to contribute to the democratization of financial information on the Internet. (Until now investors have had to pay thousands of dollars in subscription fees for transcripts.) So our reproduction policy is as follows: You may quote up to 400 words of any transcript on the condition that you attribute the transcript to Seeking Alpha and either link to the original transcript or to www.SeekingAlpha.com. All other use is prohibited.
THE INFORMATION CONTAINED HERE IS A TEXTUAL REPRESENTATION OF THE APPLICABLE COMPANY'S CONFERENCE CALL, CONFERENCE PRESENTATION OR OTHER AUDIO PRESENTATION, AND WHILE EFFORTS ARE MADE TO PROVIDE AN ACCURATE TRANSCRIPTION, THERE MAY BE MATERIAL ERRORS, OMISSIONS, OR INACCURACIES IN THE REPORTING OF THE SUBSTANCE OF THE AUDIO PRESENTATIONS. IN NO WAY DOES SEEKING ALPHA ASSUME ANY RESPONSIBILITY FOR ANY INVESTMENT OR OTHER DECISIONS MADE BASED UPON THE INFORMATION PROVIDED ON THIS WEB SITE OR IN ANY TRANSCRIPT. USERS ARE ADVISED TO REVIEW THE APPLICABLE COMPANY'S AUDIO PRESENTATION ITSELF AND THE APPLICABLE COMPANY'S SEC FILINGS BEFORE MAKING ANY INVESTMENT OR OTHER DECISIONS.
If you have any additional questions about our online transcripts, please contact us at: email@example.com. Thank you!