Good morning. My name is Misty, and I will be your conference operator today. Welcome to the NVIDIA to acquire Mellanox conference call. At this time, all participants are in a listen-only mode. Later, we will conduct a question-and-answer session, and instructions will follow at this time. If you would like to ask a question during the time of the conference, please press star then the number one on your telephone keypad. If you would like to withdraw your question, press the pound key. I would now like to turn the call over to Simona Jankowski, Vice President of Investor Relations at NVIDIA. Please go ahead.
Thank you. Good morning, everyone, and good afternoon to those in Israel. With me on the call today from NVIDIA are Jensen Huang, President and Chief Executive Officer, Colette Kress, Executive Vice President and Chief Financial Officer. We're also joined by Eyal Waldman, President and Chief Executive Officer of Mellanox. I'd like to remind you that our call is being webcast live on NVIDIA's investor relations website. The webcast will be available for replay following this call. In addition, you can find a presentation on the proposed transaction as well as FAQs on the NVIDIA investor relations website. The content of today's call is NVIDIA's property. It can be reproduced or transcribed without our prior written consent. During this call, we may make forward-looking statements based on current expectations. These are subject to a number of significant risks and uncertainties. Our actual results may differ materially.
For a discussion of factors that could affect our results and business, please refer to our most recent Forms 10-K and 10-Q, the reports that we may file on Form 8-K with the Securities and Exchange Commission. All our statements are made as of today, March 11, 2019, based on information currently available to us. Except as required by law, we assume no obligation to update any such statements. With that, let me turn the call over to Jensen.
Thanks, Simona, thanks, everyone, for joining on short notice. Today, we're announcing our proposed acquisition of Mellanox for $6.9 billion. Mellanox is headquartered in Israel and is world-renowned for its high-performance data center interconnect technology. We have worked together for many years, have partnered to build many of the world's highest performance data center systems. We're combining two of the world's leading data center and high-performance computing technology companies today. Between us, we power over 250 of the world's top 500 supercomputers. We name every major computer maker and Cloud Service Provider as our customers. Let me tell you why this makes sense for NVIDIA and why I'm so excited about it. As you know, NVIDIA is focused on high-performance computing. Because of workloads such as artificial intelligence, scientific computing, and data analytics, performance demands on hyperscale and enterprise data centers are skyrocketing.
From a computer architecture point of view, two dynamics are particularly important. First, CPU performance advancement is slowing just as computing demand is skyrocketing. As a result, accelerated computing and accelerated networking, or otherwise known as CPU offloading, is a fundamental path forward. Second, as data and compute-intensive workloads like AI and data analytics are growing exponentially, they require thousands, tens of thousands, of server nodes working together. To support these new workloads, the computing nodes and the connecting fabric will form a data center-scale compute engine. Optimal design will translate directly to higher data center throughput and utilization and lower operating costs. To accelerate applications, NVIDIA optimizes across the entire application stack, from architecture to chips, to systems, to algorithms. In the future, we want to optimize data center-scale workloads, again, across the entire stack, from the compute node to networking to storage.
For this reason, Mellanox's system-to-system data center-scale interconnect technology is important to us. We believe our platform will be stronger and deliver the best possible performance for data center customers by innovating across the computing, networking, and storage stacks. Beyond the significant product synergies, we're excited about the tremendous cultural fit with Mellanox. Like NVIDIA, they're a performance-driven culture with a focus on innovation. I look forward to working even more closely with their talented employees and visionary leaders to invent the future of computing. Great to have you on the call, Eyal.
Thank you, Jensen. I'm delighted too. The vision that Jensen has set out for accelerated computing is one that we fully share at Mellanox. The future of computing will benefit from holistic innovation across the entire technology stack, from compute to networking to storage. Combining our two companies is a natural extension of our existing partnership and a great fit given our common performance-driven cultures. We have built Mellanox based on technical excellence, innovation, and staying nimble even as we've grown. NVIDIA is an excellent match for us. Combining our companies will be good for our customers who will benefit from accelerated innovation in high-performance computing, good for our partners, who will benefit from optimized solutions across the entire computing and networking stack, and good for our employees, who will have new growth opportunities and challenges.
Let me now turn it over to Colette to give more details on the proposed transaction. Colette?
Thanks, Eyal. Today we are announcing the acquisition of Mellanox for $125 per share in cash, implying a total enterprise value of $6.9 billion. We expect the acquisition to close by the end of calendar 2019, subject to regulatory approvals in various jurisdictions, as well as approval by Mellanox shareholders. The transaction has been approved by both NVIDIA's and Mellanox's board of directors. In addition to the strategic rationale Jensen outlined, we believe this is also a financially compelling transaction. The deal is expected to be immediately accretive to non-GAAP gross margins, non-GAAP earnings per share, and free cash flow, even before assuming any cost or revenue synergies. For those less familiar with the company, Mellanox generated approximately $1.1 billion in revenue in calendar 2018, up 26% from the prior year. Calendar 2018 GAAP gross margin was 64.3%, and non-GAAP gross margin was 69.2%.
GAAP operating income was $112 million, non-GAAP operating income was $270 million. Mellanox's customer base and channel partners have significant overlap with our own. Customers include large hyperscale and cloud service providers, as well as many of the world's largest supercomputing centers. Direct customers, including many of the same large IT infrastructure OEMs that we partner with for our go-to market. We believe that there will be opportunities to cross-sell and enhance our broader solution set over time, given each company's technology assets and customer bases in high-performance computing, enterprise, and hyperscale. We are conservatively assuming no initial revenue synergies. We are also not modeling any cost synergies at this time. It is a very strategic deal, and we highly value Mellanox employees. We will continue to invest to go after the large TAM we see for the combined company. Lastly, let me touch on capital allocation.
We view the proposed acquisition to be a terrific use of our cash, not because it's immediately accretive from a financial perspective, but most importantly because it is a strong strategic fit from a technology and cultural perspective. Further, this announcement does not change our capital return program for the year. Of the $3 billion we intend to return to shareholders by the end of fiscal 2020, we returned $700 million through share repurchases during the fourth quarter of fiscal 2019. We still plan to return the remaining $2.3 billion to shareholders through the end of fiscal 2020 in the form of dividends and buybacks. We will now open the call for questions. Operator, would you please pull for questions? Thank you.
At this time, as a reminder, if you would like to ask a question, press star one on your telephone keypad. Again, that is star one to ask a question, and we will pause just for a moment. Your first question is from C.J. Muse with Evercore.
Question. I guess, Jensen, first question. Can you talk a bit about the strategic rationale here, particularly around which end markets you're looking to target? HPC is obviously the biggest potential, but would also love to hear is public cloud the main focus here, or is it more of offering a platform solution a la DGX into the enterprise?
Yeah. C.J., that's a great question. The data center market today consists of high-performance computing or supercomputing, which is the initial target of our company's CUDA architecture. It also includes hyperscale for internal consumption of internet services, provisioning of internet services, cloud service providers offering them as cloud infrastructure, then enterprise computing. These are really the major segments, if you will, from the way we look at data centers. As you know, one of the most important drivers in the data center world today is a new type of workload that's emerged, it's a data-centric type of workload. It requires an enormous amount of data and a huge amount of computation to infer insight, to extract insight out of data. We call it artificial intelligence and machine learning and data analytics. This type of workload is just simply too big to fit on any one computer.
There's a couple of ways that you could solve that problem. The dynamics are basically related to, on the firsthand, CPU scaling has slowed. Without offloading the algorithm somehow and accelerating it using new accelerators like what we do with GPU computing and what Mellanox does with accelerated networking, it's really not possible to continue to scale computing as Moore's Law ends. The second is because the data size and the compute size is so large, it won't fit on any one computer, thousands of computers have to be networked together to work together. The dynamic that is happening here is that in the future, it won't just be server scale computing that people do, but it will be data center scale computing, where the network becomes an extension of the computing fabric. You see early trends of that already. Software-defined networks, intelligent networks.
Long-term, even some part of the computation will go into the network. Where the compute node starts and ends and where the networking starts and ends as part of the overall computing fabric will be very difficult to discern. We want to take a lead in architecting and developing solutions that innovate across all of these stacks. Now, of course, the solutions will have to be open because everybody's data centers are different, and everybody's amount of supporting traditional software stacks or new software stacks like containers and Kubernetes are going to be a little bit different. Data centers will be heterogeneous for a very long time because of that.
Our ability with Mellanox to be able to support supercomputing networks from InfiniBand all the way to hyperscale data center networks with Ethernet, and to be able to put acceleration and intelligence into the entire fabric is really, really powerful. I hope that answers your question. Basically, we would like to accelerate growth into our data centers by being able to architect across both the compute as well as the network fabric.
That's very helpful. If I could ask a quick follow-up. Colette, you talked about no cost synergies at this time. Are there any cost savings from the combination of two public companies, et cetera, that we should be thinking about?
Yeah. Thanks for the follow-up question, C.J. As we discussed, this acquisition's certainly about driving accelerated growth and improving our position in the data center. Mellanox is well-managed, and it already has quite a lean organization. We have many areas of operational leverage that we can accelerate for both of our development plans as we move forward. Over the long term, we certainly see opportunities for operating leverage of bringing these two companies together.
Thank you.
Your next question is from Aaron Rakers with Wells Fargo.
Congratulations on the transaction. One housekeeping item and then one architectural question. First of all, Colette, just any color on, is there a breakup fee, the size of the breakup fee, any details on that? On the architectural question, there's a lot of discussion around the role of InfiniBand in the context of HPC, and I understand what you just said around bringing intelligence into the networking. I'm interested, Jensen, how you see Mellanox's Ethernet switch business, particularly playing into the portfolio, given the traction that the company started to see over the last couple of quarters. Thank you.
Let me start first. We're going to concentrate over the rest of the calendar year to focus on closing this deal. We need regulatory approval in the U.S. and China and a couple other jurisdictions. This time, we think that's our right focus before closing.
First of all, the easiest way to think about that is that I think the world of their Ethernet solutions and the reason for that, as you know, we work with them in high-performance computing data centers all over the world, CSPs all over the world. We have our product called T4, which is about scale-out accelerated computing. Wherever we do scale-out accelerated computing, scale-out meaning small number of GPUs accelerators inside a large number inside each node, and many nodes are being connected together, built a little bit differently than DGX, for example. DGX is a scale-up architecture. T4 was designed to be a universal accelerator for hyperscale data centers, and it's intended for scale-out. In those kind of environments, we work with Mellanox a lot. The reason for that is because of the software stack.
It turns out that the best way to accelerate a data center, is to offload the CPU, to keep as little of the data movements from being copied around as possible. The RDMA software stack, and the RoCE software stack that Mellanox has created over the years, and the body of work that they've developed in accelerating the networking stack is just world-class. That's where you get the greatest value of extending the compute node, and the networking fabric to become an extended computing fabric. You want to make sure that as much of the software is optimized as possible. They do this really fantastically throughout. Now we're going to continue to work with switch companies all over the world, and the reason for that is, of course, a data center is heterogeneous by nature.
There are parts of the data center where certain partners are going to be particularly great at. Arista is great in the segments that they serve, and Cisco continues to be incredibly valuable. We have to work together as an industry, and our goal is to make sure that the computing fabric is as optimized as possible, and Mellanox plays a really important role in that. I love all of their product lines, and particularly I love their software stack. Their architecture, which the software stack runs on, is just extraordinary.
Very good. Thank you.
Our next question is from Timothy Arcuri from UBS.
Hi, Jensen. This is Pradeep Ramani calling for Timothy Arcuri. I think you just touched on the software stack aspect of it, and that's exactly what we wanted more details on. From a software perspective, I guess, how much more work do you have to do in terms of closely integrating RDMA and RoCE with NVIDIA products? I mean, and can you talk about the capabilities that you have in-house in terms of the software engineering capabilities to do that, and kind of just the roadmap in general?
Yeah, as you know, we do a lot of software inside the company, even for networking. The reason why we do it is because we do it for NVLink, which is a processor-to-processor, in-system interconnect. Mellanox develops system-to-system interconnect. NVLink and Mellanox are complementary, and the reason why we work so closely together is because of the complementary nature of these two interconnects. We have a great deal of expertise in both companies in how to manage networking and how to optimize the networking stack for the highest throughput and the ability to connect the largest number of computers. Both companies has a great deal of expertise in this area. The important thing is, at the highest level, we are two of the world's leading high-performance computing technology companies.
Between the two of us, we power over 250 of the world's top 500 computers. There's a great deal of expertise in both companies. I would love to have Eyal answer some of that question, too, just so that we could have him do some talking. Eyal's a visionary leader, and he's, as you guys know, built Mellanox from the ground up, and I love it if he would take part of it.
Thank you, Jensen, for the warm words. Yeah. I think there's a lot of synergy in the architecture between the GPUs that NVIDIA designs and sells and the interconnected Mellanox designs and sells. We can now move on to the next stage of integration and build more efficient data centers for wide number of applications and markets. The way we've done it now is through external connectivity. The next stage of building smarter, more efficient, and more productive data centers will be higher integration of mechanisms between the interconnect and the CPU, GPU that now we'll be able to do more intimately.
Okay, thank you.
Your next question is from Stacy Rasgon, Bernstein Research.
Hi, guys. Thanks for taking my questions. First, I was curious how long you've been exploring this purchase in earnest before the announcement today. Was it weeks or months? Like, how long have you actually been in this process before actually closing on the deal today?
Yeah, that's a great question, Stacy. Eyal knows I've loved Mellanox for years. I've loved Mellanox for years for several reasons. One, we collaborate. We've been working together for a very long time, and we built high-performance computers together. They are in every one of the DGX servers, the AI supercomputers that we make. In our own products, they're in every single one. They're all over our data center. The ones that we use to achieve and deliver the world's highest performances on MLPerf. We work hand-in-hand with their engineers, to achieve those results all the time. We've been working together for years. I've loved this company for a very long time.
I've believed for a very long time also that NVIDIA would go from a chip-level company to a server-level company to essentially a data center scale company, and that when we think about architecture from that perspective. I've loved the company for a long time. In terms of when the acquisition talks first happened, that's going to be in the proxy. Obviously it was more recent.
Okay. Just to follow up, maybe housekeeping, how are you going to account for Mellanox's revenues when you report? Are you going to leave them split out in their own segment? Are you going to put them into your data center segment?
Hey Stacy, this is Colette. I'll try and answer that question. At this time, we're not changing our guidance that we just provided about a month ago for the current fiscal year. When this deal actually closes, we'll be able to give you more color regarding what we see going forward in terms of guidance with the two companies together, we'll talk about the overall reporting at that time.
Okay. Thank you, guys.
Thanks, Stacy.
Your next question is from William Stein with SunTrust.
Thanks for taking my question and congratulations on this transaction. Jensen, I'm hoping you can put this acquisition in sort of an historical perspective for us relative to the other ones you've done. What lessons would you say you learned from the prior acquisitions that sort of encouraged this transaction? What lessons might you have learned from the prior deals that might provide a different approach or different tactics around this one? Thank you.
Yeah. From a scale perspective, this is surely the largest one. From a percentage or proportional size perspective, our integration, our acquisition of 3dfx was the largest at the time. At the core, it's really about several things. One, and just as a statement, almost every single 3dfx engineer and employee that joined us at that time are still here, and many of them are on NVIDIA's management team today, on E-staff today. We wouldn't be here if not for the incredible work that the people did that came to us. We are just such a different company result of that acquisition. I think this acquisition would also be transformative for us, changing us from a chip and a system-level company to a data center scale level company. Several things that I learned.
One, there has to be an alignment about the purpose of companies, the culture of the company, and just the core attitude, the principles that govern the companies. I've worked with Eyal for a long time. We've worked with Mellanox for a very long time. Our two companies are driven in very similar ways. We're both best in our class. We're singularly focused on high-performance computing. We're very performance-oriented. You just can't help but love the fact that this company is also based in Israel, so that the population of great technical excellence is high. I'm thrilled about that part of it. I think number one is the alignment of purpose and culture. The second is just alignment of vision of the future of the industry.
We both see it the same way, that this transaction is about uniting two of the leading technology companies in high-performance computing. It's about doubling down in data centers. It's about inventing the future of data centers, which is going to be workloads that run across the entire data center. Imagine one application that's split up in containers that's operating in servers all over the data center, and results are being sent around from one server to another server over the network fabric. The entire data center becomes one giant compute engine. Where compute starts and networking starts and ends, it's going to be harder to see, and it's going to be one continuous compute plane. The second is the alignment of the vision of the future of the industry.
Thirdly, it has to make all the economic sense that the strategy would suggest. This deal is going to be immediately accretive. You have two companies that are growing. Our strategic tenets are still fundamentally sound. We believe that the future is going to be about acceleration. Mellanox's business is growing, and they're growing fabulously. For all the dynamics, the reasons that I've already outlined is the reason why they're growing. We have two growth companies coming together, two best-in-class companies coming together, two cultures that were built from the ground up coming together. We see a common vision of the future.
Thank you.
Your next question is from the line of Mark Lipacis from Jefferies.
Hi. Thanks for taking my question. Congrats on the announcement. I had two questions, one software and one interconnect. On the software side, Jensen, you've discussed transforming NVIDIA from a chip company to a platform company for a while, and I'm wondering if the Mellanox acquisition necessarily mean that you're now going to be extending what you've done with CUDA on GPUs to a higher level software capability than you have before with what will become the combination of Mellanox and NVIDIA. That's the software question. On the interconnect question, you talked about system to system. Does the Mellanox interconnect capability, does that extend down to the chip-to-chip level? Does it mean that you consider multi-chip packaging architectures in the future? Is that on the roadmap? I'll stop there. That's all I have. Thank you.
Yeah, I'll go backwards. NVLink is unquestionably the best processor-to-processor interconnect the world has today. It is just incredibly high speed, and it's designed so that CUDA programs could view all of the processors that are connected together as one large computing engine. NVLink has a very special purpose, and it's really about connecting processors to processors within one system. Mellanox is about connecting multiple systems, also with the same sensibility of very low latency, very high throughput, and just incredibly well-optimized software stacks that are integrated with the gigantic body of software that's running on top of a data center. If you look at their body of work, whether it's in high-performance computing, supercomputing, enterprise computing, or in hyperscale computing their software stack is integrated deeply into the world's body of work, if you will.
I think their interconnect and our interconnect, their software and our software, has very specific places to exist, and they're going to evolve from there. At the higher level, it is very clear that compute and networking will be one large connected compute plane, if you will. The networking will be an extension of the compute fabric. Another way to say it is that in the future, computing will be done not just within the server, but computing of an application will be done at the data center scale. One application will run across the entire data center.
It's either all of the servers coming together, all of the nodes coming together, working together to solve one problem, or the application that's been broken apart into a whole bunch of little containers with AI programs running within it, and the results of one is being essentially emailed to, if you will, sent to the input of the other. The traffic inside data centers has exploded because of this phenomenon. At the highest level, the easiest way to think about that is, in the future, we're looking at data center scale compute engines. The servers will be connected by networks that will create a compute plane that is continuous. The best architecture is going to have to contemplate this holistic nature.
That's helpful. Thank you.
This is the last question? Okay, you guys, this is the last question. Thanks for joining us. This is a really exciting day for our company, I'm super delighted that we were able to acquire Mellanox today and announce the acquisition of Mellanox today. This is going to be the beginning of a new NVIDIA that I look forward to sharing with you as time goes. Thank you.
This now concludes today's conference call. You may now disconnect.