Good morning, everyone, and welcome to the Arteris IP third quarter 2021 earnings call. For opening remarks and introductions, I will now turn the call over to Erica Mannion at Sapphire Investor Relations. Please go ahead.
Thank you and good morning. With me today from Arteris IP are Charlie Janac, Chief Executive Officer, and Nick Hawkins, Chief Financial Officer. Charlie will begin with a brief review of the business results for the third quarter ended September 30, 2021. Nick will then review the financial results for the third quarter, followed by the company's outlook for the fourth quarter and full year of 2021. We will then open the call for questions. Please note that this call will include forward-looking statements that involve risks and uncertainties that could cause actual results to differ materially from management's current expectations. We encourage you to review the safe harbor statements contained in our press release for a more complete description. All material contained in the webcast is the sole property and copyright of Arteris IP, with all rights reserved.
Please note, this presentation describes certain non-GAAP measures, including non-GAAP net loss and non-GAAP net loss per share, which are not measures prepared in accordance with U.S. GAAP. The non-GAAP measures are presented in this presentation as we believe that they provide investors with a means of evaluating and understanding how the company's management evaluates the company's operating performance. These non-GAAP measures should not be considered in isolation from, as substitutes for, or superior to financial measures prepared in accordance with U.S. GAAP. Listeners who do not have a copy of the quarter ended September 30, 2021 press release may obtain a copy by visiting the investor relations section of the company's website. Now, I'd like to turn the call over to Charlie.
Thank you, Erica, and thanks, everyone, for joining us on the call this afternoon. We're excited to report our first quarter as a public company. I want to extend special thanks to the Arteris IP team for their hard work and support during the IPO process. We are excited about this next step in our company's journey, notably since Arteris IP is the first pure play semiconductor IP company to go public on a major U.S. stock exchange in 19 years. I'll start this afternoon with a few highlights from the quarter, and then, because it is our first earnings call, I want to take some time to discuss our business and market opportunity, as many of you may be new to the Arteris IP story. We reported a strong third quarter with GAAP revenue of $9 million, up 39% year-over-year.
Demonstrating our continuing momentum through the end of third quarter 2021, we have added 28 new System IP customers, marking the highest number of new customer additions in a nine month period in company history. In the third quarter alone, the total number of active customers increased from 166 to 179, with new customer applications ranging from automotive to machine learning to consumer 5G wireless and data center SoCs, with the highest number of customers creating new machine learning SoCs. This included two major automotive design wins for leading-edge SoC projects, including one with BMW. We believe that our products are utilized in 70%-80% of all automotive ADAS SoC projects. Geographically, the largest licensing regions for the quarter were EMEA, followed by China, and then closely followed by U.S.
There were 22 customer design starts in the third quarter, continuing the strong trend and taking a total of nine months ended September 30 to 65, our highest ever number in a nine month period. We saw strong performance in the third quarter across all of our System IP application market segments, with notable new license contracts in machine learning and automotive. In addition, we were especially encouraged by the acceleration of automotive royalties as car electrification and automated driving deployment continues to expand. We have made good progress on the integration of Magillem following the acquisition November 2020, and are encouraged by the high renewal rates from our IP deployment software customers. Looking forward, we believe our market opportunity remains very attractive on a sustained basis. For those of you who are new to the story, let me tell you a little bit more about Arteris IP.
We are a leading provider of semiconductor system IP that enables our customers to deliver increasingly complex SoCs that not only process data, but are also able to make decisions. The growth in the number of connected on-chip IP blocks and subsystems places an increasing premium on the interconnect IP's capability to move data inside complex SoCs, as well as our IP deployment technology's ability to automate important aspects of the SoC creation design flow. We believe this increase in SoC complexity is creating a significant opportunity for growth of our sophisticated system IP solutions. Founded in 2003, we have pioneered and emerged as a global leader in the development of interconnect IP technology for on-chip communications that addresses the complexity, performance, and cost requirements of advanced SoC semiconductors.
Traditional on-chip communication methods, such as hybrid buses, are generally inadequate in handling communications within advanced semiconductors for sophisticated applications. We leveraged our extensive technological expertise to develop a new method for on-chip communications to address these critical semiconductor development challenges. This method is called Network-on-Chip or NoC Interconnect. By pioneering the use of proprietary networking techniques to remove the inherent architectural limitations of traditional logic communications to improve ease of integration, performance, silicon area, and power consumption, we help our customers to achieve their design goals faster, easier, and at lower R&D and SoC unit costs. We're also a leading provider of IP deployment software, which enables our customers to more efficiently deploy their internally developed and commercially licensed semiconductor IP block libraries to accelerate creation of SoC type semiconductors.
Our IP deployment solutions complement our interconnect IP solutions by helping to automate not only the customer configuration of interconnect IPs, but also the process of integrating and assembling all of the customer's IP blocks into an SoC. Products incorporating our IP are used to carry critical data inside complex SoCs for sophisticated applications, including automated driving, artificial intelligence, machine learning, 5G and wireless communications, data centers, and consumer electronics, among other applications. IP deployment software, together with our NoC interconnect IP and several NoC interface IPs that improve SoC data flow performance, make up the System IP market segment. We work directly with our customers throughout the SoC development process and seek to develop long-term sustainable relationships with them as our technology becomes embedded in their products and their processes.
We also leverage our long history in interconnect IP design and are able to serve a broad range of applications and deliver customer-specific features that are useful to other customers. For example, in the third quarter, we announced Aeva Technologies, an innovator in lidar, as our 200th customer in Arteris IP history, and Uhnder, a leader in innovative automotive radar systems. The 200th customer milestone quickly follows the company's announcement of its 150th customer in November 2020 and its 100th customer in 2018, reflecting Arteris' rapid growth as semiconductor design teams license the company's technologies to speed development and increase performance of SoCs. Specific to Aeva, they chose Arteris IP on-chip interconnect technology as part of their proprietary digital processing chip because our technology complements Aeva's high standards for functional safety.
In addition, the Arteris IP interconnect technology increases digital processing chip performance by optimizing on-chip communications for high bandwidth and low latency data flows, including state-of-the-art quality of service or QoS techniques. Within the artificial intelligence machine learning market in the third quarter, we also announced that the imaging SoC leader, Eyenix, licensed FlexNoC interconnect IP for its next-generation image processing SoC. Eyenix chose Arteris IP on-chip interconnect technology as part of their proprietary image processing chip because it enables them to design and integrate a complete and superior imaging solution without dependency on external IP blocks for the machine learning function. In addition, the Arteris IP interconnect increases image processing chip performance by optimizing on-chip communication for high bandwidth and low latency data flows using state-of-the-art quality of service techniques. Our technology also provides opportunities for our customers to reduce power consumption in battery-operated applications.
To ensure we stay at the forefront of technology innovation for these and other opportunities, we intend to continue ramping up our investments in the next-generation product offering. As an example of this, we recently announced the launch of Arteris Harmony Trace design data intelligence solution to ease compliance with the semiconductor industry functional safety and quality standards. For design teams with functional safety requirements or creating complex SoCs or systems, Arteris Harmony Trace increases system quality and the ability to achieve functional safety certification by creating and maintaining traceability between disparate systems for requirements, specification, hardware design, software code, test, and documentation, so that engineers will know immediately when a change occurs and the effect of that change on other design artifacts and part of the system.
The development of Harmony Trace was driven by our customers' needs to establish an automated traceability flow and implement change management best practices between their existing requirements, specification, their electronic design automation tool suite, their software code repository, and documentation tools. Harmony Trace allows our customers to use their existing tools and automatically link data between them due to Harmony Trace's unique semiconductor industry-specific semantic computing technology. As we look at the multitude of both captive and greenfield system IP opportunities in the market, we estimate that our TAM will expand to an estimated $3.2 billion in 2026, but $1.1 billion in 2020, driven by an increasing number of SoC design wins and growing complexity, increasing average selling prices of interconnect IP and IP deployment software, and our move deeper into the NoC interface IP market.
More specifically, we believe our growth will be driven by technology trends requiring more sophisticated on-chip processing in automotive, machine learning, 5G and wireless communications, data center, and consumer electronic markets. Also, the need for sophisticated System IP products is growing rapidly in order to address the requirements of smaller die size, lower power consumption, and higher operation frequencies, as well as management of critical net latencies in a timely and cost-effective manner. There are also structural growth factors, such as regionalization of the semiconductor industry, more system houses building their own custom or partner-customized silicon, and the lengthening of the semiconductor supply chain, which favors our IP deployment software. As a result, we believe these trends have led to an increased economic benefit of in-licensing of System IP, and in fact, all commercial semiconductor IP. This is an exciting time to be at Arteris IP.
Looking ahead, we recognize that our IPO is just one step on the journey of our company. While we're excited to have achieved this milestone, we have much more to do. We look forward to continuing our momentum in the years ahead and updating all of you in the quarters to come. With that, I'll turn it over to Nick to discuss our financial results in more detail.
Thank you, Charlie, and good afternoon, everyone. As I review our third quarter results today, please note that I'll be referring to non-GAAP metrics. A reconciliation of GAAP to non-GAAP metrics is included in today's earnings release, which is available on our website. Given this is our first call as a public company, I want to start by sharing some perspectives about our business model and financial profile. Then I'll walk through financial highlights for the third quarter, and finally, I'll close with guidance for the fourth quarter and the full year before we open up the call for questions. Our revenue is derived from licensing intellectual property, licensing software, support and maintenance service, professional services, training services, and royalties. Our interconnect IP solutions include software licenses, services, software updates, and technical support.
We enter into licensing agreements with customers that typically range from two to three years and generally consist of delivery of a design license that grants the customer the right to use the IP license to design a contractually defined number of products and stand-ready support services that provides the customer with our application engineer support services. While revenue derived from our interconnect IP license agreements is typically recognized ratably over the duration of the license agreement, customers typically pay fees either shortly after signing their agreements with us or at the start of each year following the design license term, allowing our cash flow to run ahead of revenue recognition. In conjunction with our interconnect IP licensing agreements, when products incorporating our IP are sold, we generate royalty revenue.
Royalty revenues are recognized during the quarter which customer shipments occur and are calculated either as a percentage of customer chip ASP or on a per unit basis as specified in the license agreement. Our deployment solutions include software licenses, software updates, and technical support. The software licenses are time-based licenses with terms generally ranging from one to three years. Revenue allocated to the software licenses are generally recognized at a point in time upon the later of the delivery date or the beginning of the license period or on a token basis. Revenue allocated to support services is recognized ratably over the support term. Given the variation in revenue recognition methodologies between our product offerings, as a management team, we focus on annual contract value, or ACV, as a leading indicator of financial performance.
We define ACV for an individual customer agreement as the total fixed fees under the agreement, also referred to as the total contract value or TCV, divided by the number of years in the agreement term. As this calculation does not include the contributions from royalty payments, we also refer to ACV plus trailing 12 month royalties as a metric which provides a more complete picture of our total revenue proxy. We monitor this metric to measure our success and believe that the historical increase shows our progress in expanding our customers' adoption of our platform. Moving on to our quarterly results. ACV plus trailing twelve-month royalties and other revenue as of September 30 was $45.6 million, up 21% year-over-year on a pro forma basis, including Magillem, driven in particular by growth in automotive and machine learning applications and up 4% quarter-over-quarter.
Total revenue for the third quarter was $9.0 million, up 39% year-over-year, while remaining performance obligations, or RPO, were at $50.6 million, up 48% year-over-year. Gross profit in the quarter was $8.1 million, representing a gross margin of 90% compared to $6.2 million in the year-ago period. The high gross profit for the quarter was primarily attributed to the increase in licensing and support revenue, partially offset by lower royalty revenues from the significant reduction of shipments from our customer HiSilicon.
R&D expense for the third quarter was $7.6 million, or 85% of revenue, compared to 62% in the year ago period. The increase was driven by additional development costs and an additional headcount and payroll expense as we continue to invest in new and improved product offerings. Sales and marketing expense for the third quarter was $3.2 million or 36% of revenue, roughly in line with the year ago period. We intend to continue to invest in sales and marketing as we continue to drive awareness in the market and expand our sales and application engineering forces and marketing efforts to capture the significant opportunity in front of us. G&A expense for the third quarter was $1.7 million or 19% of revenue, compared to 40% in the year ago period.
G&A reflects an increase in people and infrastructure related expenses associated with our public company-ready efforts, offset by a reduction in professional fees associated with our IPO. GAAP operating loss for the third quarter was $4.5 million or 50% of revenue, increasing from a loss of $2.6 million or 41% of revenue in the year ago period. Non-GAAP operating loss was $4.0 million or 44% of revenue, increasing from a loss of $2.4 million or 37% of revenue in the year ago period. GAAP net loss in the quarter was $5.0 million or net loss per share of $0.24.
Non-GAAP net loss for the quarter was $4.4 million or net loss per share of $0.21 based on approximately 20.6 million weighted average diluted shares outstanding. Turning to the balance sheet and the cash flow. We ended the quarter with $11.2 million in cash. Cash flow from operations was a usage of $2.5 million in the quarter, while free cash flow includes capital expenditure, was a usage of $2.6 million. Subsequent to the quarter end, we closed our initial public offering, resulting in net proceeds of $771.1 million. I would now like to turn to our outlook for the fourth quarter and the full year 2021.
For the fourth quarter, we expect ACV plus trailing 12 month royalties of $48.5 million-$50.0 million and revenue of $10.0 million-$11.1 million, with non-GAAP operating loss margin of 32%-54% and non-GAAP free cash flow margin of -37% to -52%. For the full year, we expect revenue of $36.3 million-$37.5 million, with non-GAAP operating loss margin of 40%-50% and non-GAAP free cash flow margin of -20% to -35%. With that, I would like to open up the call to questions. Operator?
At this time, we will be conducting a question- and- answer session. If you'd like to ask a question, please press star one on your telephone keypad. A confirmation tone will indicate your line is in the question queue. You may press star two if you'd like to remove your question from the queue. For participants using speaker equipment, it may be necessary to pick up your handset before pressing the star keys. One moment please while we poll for questions. Our first question is from Mark Lipacis with Jefferies. Please proceed with your question.
Hi. Thanks for taking my question and congrats on the first quarter out of the blocks here. I had a couple of questions. Charlie, maybe for you, could you talk about the process of you undergo for adding new customers? What do you have to do to convince them? How long does it take to convert somebody who's never used your IP before? Has becoming a public company helped that process?
I think it has definitely helped. One of the reasons for going public was to reassure the market that we are gonna be a long-term sustainable company, and that we were not interested in being bought, right? The market was really afraid that we were gonna go away, the way some of the other competitors that we had due to acquisition. That helped a lot. In terms of the process, it's actually shortening, I would say. Usually you go through the usual demonstration, evaluation, and then basically license negotiation and then closure. You know, that typically takes, I would say, you know, there's a range on that, right?
You know, I would say four to five months is a typical cycle, with sort of the outliers being, you know, as short as a couple of weeks. In one of the few cases we just laugh about it's taken 10 years. I would say four to five months is what that process takes, and the IPO has definitely helped.
Gotcha. That's helpful. Once you win a customer, once you break in, and you have a successful program there, would you expect the customer to embrace your technology broadly for other projects? I guess maybe another way to ask the question is, you know, you've articulated. You know, the benefits of using your IP as, you know, time to market, lower cost, lower risk. Are there other benefits for the company to reuse your IP, to reuse it once they've used it once, for, I don't know, maybe continuity of their own product portfolio?
Yeah.
Thank you.
I mean, yes. I mean, Mark, this is a sticky technology, right? Once you break in and once you're successful, the customer has a track record of reusing the technology on subsequent projects. You know, we have some restrictions about, you know, what we can disclose, but one of the customers that allow us to disclose this is essentially Mobileye, where we've been essentially in five generations of projects spanning 11 years, right? Those are, you know, we aim to foster those kinds of relationships. For us, the initial license deal is really pre-sales for the next deals. What we typically see is that we get used on the complex projects where the customer has some traditional difficulties with alternative technologies.
As they realize how productive our products are and our methodology is, they expand it to their less complex products in order to get the acceleration of essentially SoC generation across the entire portfolio. Yes, you know, we have to be very, very careful not just to have great high-quality products, but also to have very strong customer support that makes it easy for the customer to use us on subsequent projects.
Gotcha. Very helpful. I'll go back into the queue. Thanks.
Our next question is from Matt Ramsay with Cowen. Please proceed with your question.
Yes. Thank you very much. Good afternoon, everybody. Guys, congratulations on the first quarter out of the gate, and congrats, Charlie, for getting through that script without flubbing your lines at all. That was quite an accomplishment. My first question is sort of a bigger picture one. The way that my team has sort of characterized the opportunity in front of Arteris in the automotive market and in the AI machine learning market is a comparison back to, I don't know, 2009 through 2011 when a whole bunch of companies that were designing chips for mobile phones were faced with this smartphone inflection, and needed to then design SoCs and needed on-chip interconnect to do that.
That was sort of the first phase of Arteris, the company, and the big inflection. Where we're at now is, it seems like in the automotive industry, a bunch of companies that are designing microcontrollers need to design SoCs, and some of the auto companies themselves are designing chips for the first time. In the AI market, you either have big cloud companies that are starting up chip design projects internally, or you have new startups in AI that all of those companies don't have the interconnect teams. Am I articulating the inflections that are driving opportunities for your company correctly? Maybe Charlie, if you comment on that a little bit, I'd appreciate it. Thanks.
I mean, you know, the smartphone SoC inflection was that you had architectures, there are many to many communication, right? Instead of the simpler ones where you had many initiators to a single target, which was the memory. The inflection point we're dealing with now is that these chips are allowed not to just process data, but they're actually allowed to make decisions. That requires a much greater data traffic complexity. So those kinds of chips use much larger number of System IP products. They are more complex to design, so you essentially have to use productivity and automation to fight the growing complexity.
So the current inflection is that these SoCs are making decisions, and that's driving our business both today and in the future. Basically, you know, the car is, you know, one way to characterize the car of today and the near future is basically a power-constrained server on wheels, right? So the mechanics are becoming less important and the software and electronics are becoming more important, and that's essentially driving the industry to put more and more SoCs inside the car, including SoCs that are able to make, you know, some decisions.
Got it. No, that makes a lot of sense. As my follow-up question, I'd be remiss if I didn't ask about BMW since you brought it up in the prepared script. If you could give us some sort of characterization of that customer now that you're talking about the design wins that you have. Is this for one chip or one program that they're doing? Was it or is it broader than that?
Something that was led by the NoC IP or maybe the design services parts of your business. If you could characterize the scope of that engagement, that would be helpful. Thank you.
We actually, you know, to follow up on sort of Mark's question, we typically do break in on one project, right? Initially. This definitely is one project. We're not allowed to say what it's for, except to say that it's for a car. The bigger picture here is that the car companies have lost control of their architectures, right? They were relying on tier ones to give them subsystems which they assembled. Technologically, those subsystems really haven't changed a whole lot, in decades at least, you know. They've been improved, but structurally they really haven't changed. Now, the architecture of the entire car is being revolutionized, you know, partially forced by Tesla.
The car companies are now in a fairly aggressive mode in getting back the control of the architecture of the car. I don't think BMW is unique in that, you know, many of the ridesharing companies and many of the car companies are going to get involved either in building IPs, at least defining the architectures together with semiconductor partners so they can get customized chips for their software architectures, or in some cases they will build SoCs themselves. This again is something new and something of an inflection point. BMW is just, you know, one example that we're, you know, where BMW was kind enough to allow us to say who they are, but there are others.
Got it. Very clear. Thank you very much. Congrats on the early success. I'll jump back. Thank you.
Our next question is from Hans Mosesmann with Rosenblatt Securities. Please proceed with your question.
Thanks. I also echo Matt's commentary about congratulations and a good job at the intro. Just a clarification, Charlie, the BMW engagement, this is your first engagement with BMW?
Yes.
Okay. Congrats there. Just a high level question regarding renewals, and in terms of the outlook. Is that still at a high level? What part of the outlook for the quarter and the year's re-engagements with your licensing customers?
I think, you know, it varies a bit by quarter by quarter, but generally I would say, that about 70% of our business, certainly of the licensing revenue is renewals and about 30%, maybe, low 30% comes from new customers. 70% is reorders.
Okay. Lastly, and I'll bow out. The opportunity here in terms of ASP expansion per license, what are the trends there all in? How much can you get from a customer if they license everything? Thanks.
Our ASP has been going up, you know, throughout the history of the company. It doesn't come from raising prices of each individual product. What it comes from is the fact that the SoCs need more System IP functionality, right? You wind up, you know, current SoCs needing cache coherency as well as a non-coherent interconnect. They have machine learning sections, so machine learning traffic is its own class, so it needs a machine learning interconnect. There is resilience for functional safety for mission-critical SoCs. There's physical awareness because now you need to take into account physical constraints upfront.
You have to decide which data traffic stays on chip and which goes off chip. You have a last level cache called CodaCache. You have all of the IPD software, which allows you to package all of your other IPs in the SoC. You know, if you're gonna buy all that from us today, it's gonna be per project, it's gonna be somewhere around $1 million, whereas on the average, we're probably somewhere around low $400,000 per SoC. It'll go up more in the future, because you're gonna have chiplets, and you know, other types of functionalities that will make the data traffic patterns even more complex.
Great. That rundown, Charlie. Again, congratulations to the team. Thanks.
Thanks, Hans.
Our next question is from Gus Richard with Northland. Please proceed with your question.
Yes, thanks for taking the questions and congratulations, guys. Great job. Charlie, can you talk a little bit about, you know, where your incremental customers are coming from? Are they semiconductor companies or is it expanding into, you know, non-traditional system houses, cloud houses, et cetera?
Yes. Obviously, there are a lot of system houses now starting to do SoCs. Some of that is done with partners, so they're partnering with the traditional semiconductor companies to give them not custom chips but customized chips based on existing platforms, but modified with a few additional IPs for their own particular requirements. You also have a tremendous growth in startups in China. Those are the two things. The system houses are starting to build chips, and there's a lot of semiconductor startups in China. Also some of the major semiconductor companies are now starting to, you know, have issues with their own internally developed System IP.
We think that also provides an additional growth potential for our products.
All right. That leads me to the
Three things.
That leads me to the second question. You know, basically, some companies have issues with the design for reuse and you know a shortage of integration SoC integration engineers. Can you talk about you know sort of how you help alleviate those problems and you know incremental increases in productivity, and then how does that map into you know IP-XACT and how will that help accelerate designs to market?
There are two things here. One is the interconnect, and one is the IP deployment software even though ultimately they're extremely synergistic. Before Arteris, it took tens of days to build an interconnect. Maybe, you know, if you weren't particularly good at it could take even months. It took tens of days even for the good guys. With Arteris, the really effective teams can build interconnect in days, right? That's a major productivity increase, which is one of the reasons why Arteris has grown so well. But as you say, there is a tremendous shortage of interconnect engineers and even semi-experienced semiconductor engineers. What we need to do is cut down the generation of interconnect to hours, right?
That's something that we need to be working on. The other thing is what I call the disintermediation of the semiconductor supply chain. Some of these new entrants only do portions of the design cycle. They're not IDMs, right? So they build a couple IPs, and they buy the rest. They have a fairly long chain. They may use layout contractors, packaging houses and so on and so forth. They have a long chain of contractors, a long supply chain. The issue for them is how do you get it all to work when it comes back, all the stuff that you've bought in your license comes back, and you need to turn it into an SoC?
You need a standard that is essentially allows you to package all of that stuff into a form that is easily integratable. IP-XACT is a very good way of doing that, right? It allows you to build an integration model around the IP without worrying necessarily about what's inside the IP. It allows you to configure the registers, the exit boards. It allows you to manage high-level connectivity. You have a much more, much better chance of taking all that stuff that's coming back to you from all these suppliers and turning that into a working SoC quickly, efficiently, and at lower cost and lower risk. We're involved in both those areas.
One of the things that we find is that the IP-XACT tools are actually making our interconnect IP subsystems easier to integrate in their SoCs. We're actually in some sense eating our own dog food, and it turns out that the two things are quite synergistic.
Got it. Very, very helpful. If I can sneak in a third, I apologize. You know, as we're going to system and package, heterogeneous integration, you know, as SoCs disintegrate a little bit into separate die, how does that impact your business?
It makes it more valuable. It makes it more complex. We thrive on essentially allowing customers to manage the complexity. You know, think about it just as an example of the Ncore cache coherent interconnect, right? The reason you need cache coherency is that you need to have smaller multiple cores that aren't so power hungry and generate so much heat and those kinds of things. The software programmers are not trained in schools to worry about hardware. I know this from my oldest son who's a computer science guy, and he was aghast at what it takes to manage hardware.
We need to handle that for the software teams, and so we came up with cache coherency, which essentially allows multiple processors to look like a single programming space. Now what people want is to make multiple dies look like a single programming space. This is obviously a more complex and more expensive and more profitable problem, right? The chiplets is a technology that is here to stay. It's still not deployed by a huge numbers of people except on the leading edge, but it makes the interconnect much more complicated and much more sophisticated yet again.
Got it. Thanks. Thanks so much. I'll jump back in the queue.
Our next question is from Ambrish Srivastava from BMO. Please proceed with your question.
Hi. Thank you. I don't want to make Nick feel too left out. Right out of the gate, Nick, pretty solid keeping the numbers tight. Maybe, as we learn the business model and the quarterly ebbs and flow, can you just please frame for us visibility with respect to the KPIs? If you look at ACV plus the LTM, RPO, deferred revenue, that would be very helpful. I did have a question for Charlie, while you catch your breath.
Well, I'm delighted that you decided to give me a chance to shine, Ambrish, so thank you for that. So yeah, in terms of visibility, we have you know, I think from our previous conversations, we have a fairly strong and visible pipeline from whether you look at ACV and TTM royalties or you look at revenue. The one element of revenue that is not so hundred percent predictable is our IPD revenue, because most of that is point in time upfront. So if you look at it from quarter to quarter it can fluctuate. But in long term, because we have a very high proportion of repeat customers, so the...
What Charlie was talking about, just to make it clear, was that of our revenue, 70% of that, or our bookings, 70% of that comes from previous customers, existing customers, and the other 30% comes from new, roughly. The 70% that's pre-existing customers run on a 98% retention rate. In other words, 98% of our customers come back. Even if there's a point in time rev rec for an IPD customer, it might appear sort of in Q1 or it might appear in Q2, depending on when the customer decides to place that repeat order.
The benefit of having ratable revenue recognition on pretty much all of our SIP or interconnect revenue and or contracts and some of our IPD contracts is that once that goes into the pipe, into the waterfall, essentially, we can see for the next two, three, four years how that's going to play out into revenue. The same's true, you mentioned ACV. The same's true with ACV. ACV is again a waterfall of all current contracts. So if we have a contract that's say a three-year contract, which is not unusual, then that gets basically divided up into 36 months, and it goes on at one thirty-sixth of that, or sorry, one, because annual, one one-twelfth of that. So a year's worth.
A three-year contract would of $300,000 would be $100,000 of ACV, and that would go on the day the contract's signed, and it would come off three years later. We can predict with a fair degree of clarity, which is why you don't see a great deal of variation on these numbers from where we originally talked to you, because we do have this incredibly visible pipeline. We do also on operating expense of course, because we can control that quite carefully. Does that kind of answer your question?
It does. It helps to solidify what we've been talking about in the past. Thank you for that. Charlie, turning to you on the design starts, what are the areas that these are in? You gave us some good comparisons from a year ago for some of the metrics, but how are the starts changing vis-à-vis end markets from a year ago, couple years ago, and then where do you see that going? I'm assuming it's gonna be more automotive SoC, AI machine learning, but it would be helpful to see what the comparisons were from a couple of years ago. Thank you.
I mean, the hot market, well, you know, I know a couple of years ago, but a bunch of years ago was smartphone, right? There was, you know, the majority of the designs came there. As that market consolidated, the volume stayed the same, but the number of designs and the number of companies have shrunk, right? Now it's expanding a little bit with 5G because 5G is gonna be used for machine-to-machine communication. That part of it is expanding yet again. Kind of the first, you know, IoT market that actually works from a business point of view, and this is just my opinion, is the automotive market.
The car is really an IoT device. It's this whole huge multi-billion-dollar investment that's essentially turning the car from a mechanical device to a device that's full-time attached to the cloud, that's attached to the electronic road infrastructure, which is going to emerge, and the cars are also gonna communicate to each other, right? The car is really a very complex and actually the initial IoT device. It's a very, very nice business, and consequently, everybody's piling into automotive, right? The number of customers that are potential automotive customers is just increasing because there's such a growth in the number of chips that are gonna be inside the car. The biggest market is machine learning, right?
Some of the car chips are machine learning chips. The ADAS chips are machine learning chips. Machine learning is going to be everywhere, and it's gonna be used for making this relatively complex system functionality to be more user-friendly, right? That the machine is gonna figure out, you know, how to operate itself in certain cases. The largest number of designs in the future are gonna be in machine learning. However, from our experience so far, I think Nick would say that we probably have, ignoring the automotive guys, one non-automotive machine learning customer paying royalties, right? The volumes of that market haven't really started yet. Yet that is what is the future market.
There's a lot of other companies, there's servers, there's consumer devices. We'll take all of that, but the focus for us right now is really on automotive, machine learning, and 5G from a product development point of view and feature development point of view.
Got it. Thank you and good luck.
Our next question is from Mark Lipacis with Jefferies. Please proceed with your question.
Hi. Thanks for working me back into the queue. Quick question here. It's a follow-up on the ASP question before. You know, as you look at, you know, your projection for your own internal projections for ASP improvement over time, to what extent is that ASP expansion predicated on new IP that you haven't developed yet versus IP that you have already developed, and you're just waiting for, you know, the customers to reach for it? Thank you. That's all I had.
Well, that's a good question and not completely easy to answer quantitatively. Clearly the products that we have, the customer spend on those products is increasing, right? There's more cache coherent chips, there's more machine learning chips, there's more automotive chips. Those require more of the products that we already have, right? But there's also other things that need to be developed. You know, our strategy has been to not shortchange R&D spending, and so we are putting out a new product every year. That's the goal, is one new major product every year and as well, three releases, two to three releases of every major product that we have that are sort of customer enhancement releases.
We're trying to keep pace with this market, because we're basically doing R&D for 10% of the chip, right? The interconnect is about 10% of the silicon area, roughly, maybe a little bit more depending on what functionality you put on. We need to keep pace with this, and we are gonna be delivering new products continuously. I don't know, if I was to take a guess, maybe I would say 40% of the growth will come from the products that we have, and maybe 50%-60% will come from products that haven't been developed yet.
One of the advantages Arteris has is because we see so many designs by so many customers, we have a fast rate of learning than almost anyone because we can synthesize the customer requirements of what they're going to need from a much bigger sample base than, for example, any internal group, right? We are gonna be continuously delivering at least one product a year for the foreseeable future because that's what the market can absorb and that's what the market requires.
Got it. Very helpful. Thank you, Charlie.
We have reached the end of the question and answer session, and I will now turn the call over to Charlie Janac for closing remarks.
Yes. I would very much like to thank everyone on the call today. It's been a pleasure working with you. Also thank you for your interest in Arteris. Before we close, I would like to acknowledge my thanks to our employees for their continued dedication and diligent work and to our shareholders for their support. We look forward to updating you in the future, and we'd like to have you have a pleasant afternoon and evening, depending where you are. Thank you for your support of Arteris.
This concludes today's conference, and you may disconnect your lines at this time. Thank you for your participation.