Ambarella Inc (AMBA) 2018 Q2 法說會逐字稿

Good day, ladies and gentlemen, and welcome to the Ambarella Second Quarter Fiscal Year 2018 Earnings Conference Call.

As a reminder, this call is being recorded.

I would now like to introduce your host for today's conference, Ms. Deborah Stapleton, Investor Relations.

Ms. Stapleton, you may begin.

Thank you, Jonathan.

And good afternoon, everyone, and welcome to Ambarella's Second Fiscal Quarter 2018 Financial Results Conference Call.

Our speakers will be: Dr. Fermi Wang, President and CEO; Les Kohn, CTO; and George Laplante; CFO.

The primary purpose of today's call is to provide you with information regarding our fiscal 2018 second quarter results.

The discussion today and the responses to your questions will contain forward-looking statements regarding our projected financial results, financial prospects, market growth and demand for our solutions, among other things.

These statements are subject to risks, uncertainties and assumptions.

Should any of these risks or uncertainties materialize or should our assumptions prove to be incorrect, our actual results could differ materially from these forward-looking statements.

We are under no obligation to update these statements.

These risks, uncertainties and assumptions as well as other information on potential risk factors that could affect our financial results are more fully described in the documents that we file with the SEC, including the annual report on Form 10-K that we filed on March 30, 2017, for the 2017 fiscal year and the Form 10-Q filed on June 8, 2017, for the first fiscal quarter of 2018.

Access to our second quarter results press release, historical results and SEC filings can be found on the Investor Relations portion of our website.

I will now turn the call over to Dr. Fermi Wang.

Thank you, and good afternoon, everyone.

I am pleased to report today that our Q2 2018 revenue was $71.6 million, representing an increase of 10% from the same period in the prior year.

Our non-GoPro revenue grew at 16.3%.

During the quarter, we experienced strong growth from IP security camera market both from professional and the home monitoring markets.

We also continued to grow our OEM automotive business with strong design win activity and revenue growth from OEM automotive video recorders.

And we are continuing to invest in the technologies required to deliver future generations of highly-intelligent HD and Ultra HD cameras with particular emphasis on high-performance computer vision.

We see computer vision as a key differentiator in camera markets, including automotive, IP security, drones and the robotics.

The IP security camera and OEM automotive markets are now the largest available market -- available camera markets and represent the biggest growth opportunity for Ambarella.

In automotive cameras, computer vision is enabling the next generation of ADAS systems and increasing level of driving autonomy, while in IP security, it is enabling features including smart detection, people recognition and object classification.

As we discussed in our last earnings call, we have successfully brought up our first computer vision chip CV1, and I will now hand it over to our CTO, Les Kohn, to update you on our computer vision technology and chip status.

Thank you, Fermi.

I'm excited to share some highlights in our first computer vision chip, which we call CV1.

When we started CV1 development 4 years ago, our goal was to develop a platform truly optimized for computer vision.

Our approach has always been to start with a deep understanding of the core algorithms and end design and optimized architecture for accelerating those algorithms.

In the past, we did this for image processing and video compression, and we achieved a significant advantage in die size and power efficiency compared to general-purpose application processors.

CV1 starts with our image processing pipeline, which has been optimized over many generations to provide superior image quality, especially with low light and High Dynamic Range scenes.

This enables CV algorithms to produce significantly better results under challenging conditions.

Extending our algorithm optimized approach to CV acceleration was a big challenge because of the broad range and rapid evolution of CV algorithms.

In addition, we wanted to enable customers to pull up their proprietary algorithms to allow them to differentiate their products.

The problem comes down to scaling performance to trillions of operations per second without overly limiting flexibility.

CPUs and GPUs achieve this performance level by using many processor cores running in parallel.

This brute force approach requires software tools to break the algorithms into small steps running on many cores.

This leads to large die size and high power consumption.

CV1 uses a fundamentally different approach with a flexible CV hardware engine programmed with a high-level algorithm description.

This description allows the hardware to exploit all available parallelism without software intervention.

We call this high-level programming description "CVflow" since it specifies data flow connections between a set of optimized CV operators.

The CV operators include high-level functions such as convolution and matrix multiply, which are specifically optimized for deep learning algorithms.

In fact, neural network development tools like Cafe and TensorFlow use similar types of flow-based description.

This allows us to efficiently map CNN networks trained in these frameworks to CV flow descriptions that run on CV1.

The CVflow processor supports many other CV algorithms, including the stereo obstacle detection and terrain-mapping technology developed by our VisLab team.

We believe that stereo depth information provides an important augmentation to mono CV processing, resulting in an extra margin of safety for autonomous driving and other applications.

Mono CV processing depends on training to detect obstacles and may not detect obstacles that are not represented in the training set.

Stereo detects obstacles without relying on training because the depth information is used to directly construct a 3D model of the road and obstacles.

CV1 incorporates a dedicated stereo disparity engine, which provides depth images beyond the resolution of radar and LIDAR.

This depth information is then processed by the CVflow processor to generate high-level 3D models in real time of the obstacles and terrain.

CV1 supports other stereo CV algorithms such as curb and barrier detection, drone obstacle detection and localization.

In fact, CV1 can implement full autonomous drone functionality, including sensor fusion and navigation.

Now let's talk about performance.

Rather than quote meaningless teraflop numbers, we aim at system-level performance and real applications.

CV1 can process a stereo pair of 4K UHD image sensors or 4 stereo pairs of 1080p image sensors running at 30 frames per second.

4K stereo is ideal for long-range automotive applications and provides an obstacle detection range of over 150 meters.

Quad 1080p stereo sensors are ideal for 360 degrees surround view object detection at shorter distances for applications such as autonomous drones.

We received our first CV1 silicon samples in May and have successfully completed initial software bring-up.

We will be sampling CV1 to leading customers in Q4 this year, including a software development kit for them to port their algorithms.

We look forward to demonstrating CV1 capabilities at CES in January.

CV1 is the first member of our CV family.

Additional chips under development will cover multiple markets and price performance points.

Now that the foundation is in place, we expect to introduce new CV chips at our traditional 2 to 3 chips per year.

The new chips will be software compatible with performance upgrades for customers starting with their CV1.

This new platform has given our engineers an opportunity to push the technology in some of the most exciting market opportunities available.

We look forward to working with our customers in a variety of markets on the next generation of CV-based businesses.

Okay.

Thank you, Les.

And a lot of progress we have made in computer vision is a result of our acquisition of VisLab in 2015.

In May, the IEEE awarded Prof. Alberto Broggi, the founder of VisLab and now General Manager of Ambarella Italy, with the IEEE Medal for Environmental and Safety Technologies.

The medal was awarded for leadership in vehicular environmental perception and for taking worldwide milestones in safe and reliable intelligent vehicles.

Professor Broggi's pioneering activity in the field of vehicular robotics are recognized worldwide as milestones that have shift the early stage of perception for intelligent vehicles and the leading effort in autonomous driving.

Now I'll provide us a quick business update.

In the professional IP security market, Sweden-based Axis introduced its M family of 360-degree network cameras targeting small offices and retail stores.

The family includes 6- and 12-megapixel models that provide a full 360-degree panoramic view and up to 30 frames per second video.

The Ambarella-based cameras use advanced networking to provide distortion previews even in the far corners of the image and to support the ability to zoom in to analyze live or recorded video footage.

Also during the quarter, Hikvision, the world's largest security camera company, introduced its mini PanoVu indoor panoramic surveillance cameras.

Based on Ambarella's S3L camera SoC and the B5 SERDES companion chip, the new cameras integrate 3 IP security cameras plus 1 pan-tilt- zoom camera into a single, small form factor enclosure to provide a full 360-degree field of view.

We are now seeing the major U.S. and European camera makers adopting the HEVC video standard for their next generation of cameras following the current step in the China market over the last couple of years.

As a result of this trend, we are seeing strong design win activity for our S3L, HEVC SoC family as well as the S5L family, which was introduced earlier this year.

In the home monitoring market, we continue to see significant design win activity from retail customers, including new battery-powered designs with advanced analytics.

During the quarter, Logitech introduced its Circle 2 HD home security camera.

The Circle 2 can operate both indoors and outdoors, is weatherproof and can operate as wired or 100% wirefree camera based on Ambarella's S2LM camera SoC, it offers a 1080p video, night vision and up to 180-degree field of view.

Its smart location feature automatically detects when you leave the house or come home allowing the user to receive alerts only when they are away.

Advanced analytics also enable it to differentiate between people and pets.

In the action camera market, China-based Xiaomi introduced its Mijia 4K action camera featuring a large 2.4-inch touch screen display and wireless connectivity.

The camera used Ambarella's A12 camera SoC to enable 2 hours of recording at 4K video resolution.

The Mijia also supports high-speed 200 frame per second 720p video capture and 8-megapixel photos, 6-axis electronic image stabilization.

In the drone market, U.S.-based Vantage began shipping its Snap consumer drone.

The easily portable drone can be snapped together using magnetic connectors.

It uses Ambarella's A9SE SoC to offer 4K video, 12 megapixel still photography with up to 10 photo burst per second while offering over 20 minutes of flight time.

It can also track the user in virtual wire, air tripod and free form following modes.

In the automotive market, we continue to see increased shipments into original equipment and dealer-fit dash cameras, especially in Chinese and Japanese markets.

During the quarter, Shanghai Automotive Industry Corporation, or SAIC, China's largest automaker, launched its Roewe ERX5 electric SUV.

The Roewe ERX5, which SAIC describes as the "world's first electric Internet connected SUV," has advanced Internet features and an Ambarella A12-based embedded dash camera with WiFi connectivity.

Also during the quarter, Chinese automaker Geely introduced its BoRui B class vehicle featuring an integrated camera using Ambarella's A7LA SoC.

The dash camera is fully integrated with the in-car infotainment system and features full HD video recording.

In summary, we are pleased with our Q2 results and our continuing success in IP security and automotive markets.

We are also excited about our progress in delivering advanced computer vision technology and believe that this technology will provide additional opportunities for us in our current and the future markets.

I will now hand it over to George for more details on the financials and our guidance for Q3.

Thank you, Fermi, and good afternoon, everyone.

Today, I will focus my review on the financial highlights for the second quarter of fiscal 2018 ended July 31, 2017, review the financial outlook for Q3 of fiscal year 2018 that ends on October 31, 2017, and then give a brief update on the full year revenues.

During the call, I'll discuss non-GAAP results and ask that you refer to today's press release for a detailed reconciliation of GAAP to non-GAAP results.

For non-GAAP reporting for Q2, we have eliminated stock-based compensation expense adjusted for the impact of taxes.

Our Q2 2018 revenue of $71.6 million represents an increase of 10% from the $65.1 million of revenue in the same period for the prior year.

Revenues in Q2 excluding sales to GoPro and their ODMs grew 16.3% from the previous year, finishing the quarter at $67.9 million.

In Q2, we saw strong year-over-year growth in IP security, followed by in-line growth in auto and non-GoPro wearable markets, partially offset by a decline in drone and GoPro sports camera revenues.

In IP security, year-over-year growth of approximately 29% from combined professional and consumer revenues was led by solid performance from the China and North America professional security markets and the home monitoring market in the U.S. Solid revenue increases from the OEM automotive video recorders plus in-line, after-market shipments resulted in good year-over-year growth for the auto market.

Drone revenues in the quarter were down from last year as new product launches in the prior year made comparisons difficult in addition to a significant weakness in -- from our Tier 2 drone customers.

Non-GAAP gross margins for Q2 was 63% compared to 64.3% in the preceding quarter and 67.1% in the second quarter of the prior year.

As previously discussed, the decline in gross margin from the previous quarter was primarily due to the expansion of the lower-margin professional security market in China, which began to ramp in the quarter.

Non-GAAP operating expenses for the second quarter were $25.4 million compared to $25.7 million for Q1 of this fiscal year and $23 million for Q2 of the prior year.

OpEx came in at the low end of our previous guidance due to slower-than-anticipated headcount increases as well as a change in the timing of the start of a new chip.

We expect the investment in headcount and chip development to accelerate in line with budget in our Q3.

Non-GAAP net income for Q2 was $16.5 million or $0.48 per diluted ordinary share compared with non-GAAP net income of $18.5 million or $0.54 per diluted ordinary share for the same period in the previous year.

The non-GAAP effective tax rate in the quarter was 17.3%.

In the second quarter, the non-GAAP earnings per diluted ordinary share are based on 34.6 million diluted shares as compared to 34.2 million diluted shares in Q2 of fiscal year 2017.

Total headcount at the end of Q2 was 680 compared to 672 at the end of the previous quarter, with about 82% of employees dedicated to engineering.

Approximately 72% of the total headcount is located in Asia, primarily in Taiwan and China.

We ended Q2 with cash and marketable securities of $400.8 million, adding $10.6 million of cash from operations in the quarter.

The reduction in cash in the quarter from the prior quarter was primarily the result of the company's repurchase of 595,770 shares at an average price of $50.25 for total consideration of approximately $29.9 million.

In total, 551,351 shares were purchased in the quarter under the previous $75 million program that ended on June 30, 2017, and the balance from the new 12-month $50 million program that commenced on July 1, 2017.

The total number of shares purchased under the expired $75 million program was 1,119,178 at an average price of $50.63 for a total consideration of approximately $56.7 million.

Total accounts receivable at the end of Q2 was $38.4 million or 49 days sales outstanding.

This compares to accounts receivable of $22.9 million or 32 days sales outstanding at the end of the prior quarter.

Accounts receivable increased due to the timing of shipments in the quarter.

Net inventory at the end of Q2 was $17.7 million or about 63 days of inventory compared to $19.1 million or 77 days of inventory at the end of Q1.

Accounts receivable and inventory remain in line with company targets.

WT Microelectronics, our Asia logistics supplier, represented approximately 67% of revenue in the quarter compared to 72% for the same period in the previous year.

WT was the company's only 10% customer in Q2.

I would now like to discuss the outlook for Q3 of fiscal year 2018.

We expect revenues for our third quarter ending October 31, 2017, to be between $87.5 million and $90.5 million, representing a decrease of between 12.9% and 10%, respectively, from Q3 of last year.

Non-GoPro revenues in Q3 of fiscal year '18 are expected to grow between 4.9% and 9.4%.

Non-GoPro revenues excluded estimate revenues from GoPro and their ODMs of $34.6 million and $18.4 million in Q3 of fiscal years 2017 and '18, respectively.

We expect solid year-over-year revenue increases in the IP security and auto OEM markets as well as continued growth in the non-GoPro wearable, offset by a substantial decline in GoPro revenues and a moderate decline in the drone market.

We now expect the drone market to be down for Q3 of this year due to the negative impact the recent launch of the lower-priced Spark drone from DJI appears to be having on the higher-end segment of the drone market as well as continued weakness of Tier 2 customers.

For fiscal year '18, we now believe our revenues will be down 3% to 7% from the prior year versus the previous forecast of plus or minus 3% change from last year.

Several factors are impacting the second half of the year.

First, we expect drone market to decline year-over-year as lower-priced drones from DJI negatively impact the higher-priced market segment plus a substantial decline in revenue from Tier 2 customers.

The second factor is the slower-than-expected growth in the new virtual reality market.

In addition, our customers are now beginning to see a shortage of certain memory components in the market that may impact their camera build schedules later in the year, negatively impacting their demand for Ambarella chips.

This slower growth will be partially offset by higher-than-expected revenues from GoPro in the second half of the year.

The combination of factors will lower our total non-GoPro year-over-year revenue growth to 9% to 12%.

We estimate our Q3 non-GAAP gross margin to be between 62% and 63.5% compared to 63% in Q2 of 2018 and 66.3% in Q3 of the prior year.

As expected, we are seeing the increase in product mix of security revenues having the largest impact to margins in the quarter.

We expect non-GAAP OpEx in the third quarter to be between $28 million and $29.5 million, with the increase from Q2 primarily coming from increased chip development cost as we commence development of 2 additional 10-nanometer CV chips, including our initial ASO-qualified chip for the audio -- auto OEM market.

We expect the non-GAAP effective tax rate to be approximately 16% in the quarter.

We estimate our diluted share count for Q3 to be approximate 34.3 million shares.

I would like to thank everyone for joining our call today, and now we'll turn it back to the operator to manage the Q&A session.

(Operator Instructions) Our first question comes from the line of Quinn Bolton from Needham.

First, I want to start off on the CV1 and see if you guys could give a little bit more detail.

First, you talked about the ability to support neural networks and wasn't sure whether the chip is designed to work with some of the frameworks out there like Cafe and TensorFlow, or is -- you've mentioned the stereo capability that would allow it to not necessarily depend on trained models.

So could you give us a little bit more detail how you envision those -- that chip working in the field?

Sure.

We view those 2 technologies as complementary.

So we do support the existing CNN frameworks like TensorFlow and Cafe, and we do use CNN models to perform mono object detection and so on.

But at the same time, we think augmenting that with additional stereo information provides extra redundancy and safety in the system.

Got it.

So it's a dual approach, not one or the other?

That's right.

Great.

And then the second question just on the memory shortage.

Is that primarily in the security camera side?

Or can you talk about which parts of the business that's impacting?

And how long do you think the shortage of memory chips could impact the business?

Well, I think we are looking at several customers not particularly in each market, but we do see the shortage of memory, including DRAM and the non-flash that impacts some of our customers, who are -- who have design, fixed and shipping products as very helpful than to change the different components for -- if one of the those memory become shortage.

So we only start seeing the impact a little bit, but we don't know how long and how -- what's the total potential impact, but we want to highlight so that investors are aware of that.

We will continue to watch the impact.

Hopefully the shortage will recover soon.

Our next question comes from the line of Suji Desilva from Roth Capital.

So on the drone market.

Can you talk about if the revenues here has further downside risk?

Or whether it's kind of not recovering as fast from the levels of that versus a year ago?

Well, we try to take into consideration most of the downside risk now.

Obviously, DJI remains the primary customer there.

We are starting to get forecasts in from all our customers.

So we hope -- we've taken into consideration most of the changes that we see coming.

But right now, it's a little early.

We don't understand the complete impact of the Spark.

We're only reacting based on forecast we're getting in from customers at the high end.

Okay, George.

That color helps.

And then a question on the computer vision chips and the product there.

Can you help us understand how you think about the TAM there?

Is there's an incremental metric that you get from the traditional chips in terms of ASP or content per car or as part of design wins?

Any way to think about how it increases your opportunity.

Right.

From a TAM point of view, I think first of all, computer vision applies to all our existing markets, including security camera and drones.

And also because of the -- like I said before, I think the computer vision will trigger a refresh on the security market.

And also we believe that ASP of the new chip will be higher than in the past, so that will increase the -- not only the unit of TAM but also total dollar of that TAM.

In addition to that, I think the biggest opportunity for us in the future from a TAM point of view is really automotive, and we believe that the automotive solution that we are providing, not only the recorders, but e-mirrors and the surround views, they all will require ADAS function in the future.

And we are talking to customer.

They are asking us to put ADAS as a part of the road map for that.

So from that point of view, I think the automotive OEM is probably providing the biggest TAM growth in the future.

Our next question comes from the line of Matt Ramsay from Canaccord.

I guess the -- we have some pretty well-established metrics, I think, for evaluating Ambarella solutions versus the guys at Hisilicon and other folks in terms of power and image quality, et cetera, compression efficiency in your traditional markets.

Les, you talked a little bit in your prepared script about comparing CV1 versus GPUs and some other things.

But maybe you could -- are there certain metrics that your customers are focused on in evaluating your CV1 chip versus competitors?

And how is that competitive landscape set up there versus what NVIDIA or Intel or whomever is doing in that space?

Anything you could expand on there would be helpful.

Sure.

Typically, customers -- the more sophisticated customers tend to have their own internal networks that they use to evaluate.

So what we would expect is that they would port some of their internal applications onto our chip and then benchmark that against some of the other people that are out there.

But what I would say is that I talked to a number of customers now in both CV1 and on our road map and we've gotten very positive interest in CV1 on the road map.

And customers in general have told us that general-purpose CPU, GPU and app processors are just not the right long-term solution for CV.

Got it.

Got it.

That's helpful.

And then, I guess, George, how do we think about visibility in some of these end markets?

I mean it seems to be from a forecasting standpoint, a bit of a challenge with, I guess, the consumer-focused parts of the business.

As the business becomes more focused on security and automotive and some of these more stable end markets, how do you see the visibility that you have in your revenue growth over the next, say, 4 to 6 quarters improving over time versus where it is today, focused on drone and some of the GoPro things as bigger parts of the business?

Well, I think in the near term, visibility will still be difficult in the consumer market.

So I think for the next 3 to 4 quarters, before we start really landing customers and having revenues generated from these other markets, I think it's still going to be a bit difficult.

I think looking out, let's say, 5 to 7 quarters, we should start seeing revenue generation from CV products across multiple markets, which I think will help.

There'll still be a consumer portion of the CV revenues but I think the larger portion of the revenues will come from what we hope to be more stable markets like automotive.

Got it.

And then just one more to sneak in here.

I know you've talked about developing some of the automotive-grade chips specifically and some cost that might be with that.

Is there any way that you could bound for us at all how should we think about OpEx growth over the next 3 or 4 quarters?

I know you guys don't guide that far out, but it seems that there is going to be a ramp in spending here to support some of these chip designs and just want to make sure we capture that in models.

Yes.

I think at least in the next 2 quarters, I think they're in line with my guidance today.

I'll be talking quite a bit more about next year on the November call, but we can expect an additional step-up, but I'll be more specific on what that looks like in November.

But I think for the rest of this year, it's pretty much in line on a quarterly basis with what I said today.

Our next question comes from the line of Joe Moore from Morgan Stanley.

In terms of the numbers.

When you talked about sort of down 5% in the year at the midpoint, I guess it implies a pretty big shortfall relative to where people were for specifically the fourth quarter.

I guess, are the factors that you're seeing just sort of building up over the year and it manifests itself in Q4?

Or should -- just how should we think about that Q4 run rate?

Is there a correction built into that?

And as we head into next year, if you're 15% or so lower than where we were for January, I mean, how should we think of that for the following year?

Well, I think this really hasn't been building up.

I think the major change is in the drone market.

I think we indicated on the last call, at that time, we did have concerns over the impact of the Spark on the higher end of the market.

In addition to that, the Tier 2 customers have really fell in -- fell off quite substantially in our forecast for the second half.

I would say, they're down over 50% from last year in the Tier 2 segment.

So those 2 areas are the major changes.

I did put in some risk factor for the memory.

I think growth next year in the non-GoPro area still can be in the 10% to 15% range.

We'll have a lot more to say about that in November, but I'm still comfortable at this point that the non-GoPro areas can still grow next year.

Okay.

That's helpful.

And then can you talk about the automotive OEM businesses now, how it's progressing?

And it seems like surround view versus the sort of rearview mirror types of application, where are you with -- in terms of generating revenue on the rearview mirror side?

Well, first of all, I think we continue to see a lot of other OE and the Tier 1 product (inaudible) around video recording and e-mirror and surround view like you said.

And in fact, I will say the majority of revenue growth in automotive market comes from OEM right now.

And although most of them came from recorders and like how you said, we believe that our revenue from e-mirror and surround view will come -- start coming in the second half next year.

Our next question comes from the line of Jagadish Iyer from Summit Redstone.

So 2 questions.

First question is what kind of deep learning is built in the CV chip?

And I was wondering, which vertical would be the first to adopt the CV1?

And what kind of ASP multiple should we be thinking about it?

And is there an appetite for consumers in the consumer market to pay up a higher price for some analytics?

And then I have a follow-up.

Okay.

So responding to the first question, the CV flow processor that I described has specific acceleration support for deep learning networks.

And we believe it's going to be very competitive in terms of performance and power efficiency against general-purpose hot processor chips, and we will be demonstrating that at CES.

And I think to answer the second part of the question is really related to the ASP in the application.

We believe that the first adopter for the computer vision technology is, well, current market on security camera, on drones.

But I wouldn't be surprised that there will be new application and, of course, we also talk about auto in the future.

So in terms of ASP, we believe it's going to be higher but we haven't really start negotiating price.

I think price negotiation is always based on the market condition and competitive situation.

But I'm pretty confident that the ASP will be higher than our current ASPs.

Okay.

Fair enough.

Second question, George, I was just wondering given that there is such rapid growth in the IP security cameras, I was wondering how should we think about the growth margin -- gross margin trajectory as your mix shift changes now that drones come down, IP security goes up and your sports camera starts to kind of stabilize?

So how should we think about the margin trajectory for the next 2 to 3 quarters?

Yes, it's still pretty much in line with my guidance over the first couple of quarters.

We should see the margins flow down towards the high end of my target margin of 6 -- 59% to 62% as the mix in China security increases.

And I think we're still comfortable with that guidance.

Our next question comes from the line of Ross Seymore from Deutsche Bank.

Just one clarification, first on the housekeeping side.

George, you mentioned what the GoPro revenues were a year ago and then what's implied in your guidance.

I think you said it was $34.6 million last year.

Can you tell us what it was implied this year?

I just couldn't get that one down quick enough.

Yes, I think -- let's see, $18.4 million is what I'm including in my guidance for this year.

For this year's quarter.

And forgive me if I can't do the math, but going back to Joe Moore's question on the fourth quarter, that it seems like in the implied guidance or implied drop sequentially of about 20%.

Is that -- I know you said drones are the incremental change for the year as a whole, but when we think about that sequential drop, is a lot of that $18 million just simply going away in the fourth quarter?

Or sequentially, are there more things that are dropping off?

The 2 biggest impacts later in the year are drones and the drop year-over-year in GoPro.

Those are the 2 biggest declines.

Drones is a significant decline at least at this point, we're forecasting for the fourth quarter.

And as we think about the IP security camera, that's been strong for a number of quarters in a row.

Can you talk a little bit about and just lend some context between the professional and the consumer side of things?

Relatively speaking, which one is stronger?

And at times in the past, it's gotten a little lumpy over in China.

It seems of late it's been lumpy good.

Are there any concerns about that lumpiness coming in on the back side as we look forward?

Well first of all, the lumpiness is going to be always there because China -- that's how China market operates.

It's hard for us predict.

Sometimes lumpy good, sometimes lumpy bad.

But in the last few quarters, we are in the positive side.

And from the -- comparing the professional and consumer security camera, I think consumer security camera definitely is continue to grow very strong.

I think on the retail side, we see a lot of design activity and also we see multiple customer taking our product into production.

We talk about factory power design and also ring bell -- the doorbell type of applications, and we continue to see momentum in that area.

At the same time in professional security camera, we talk about -- we have HEVC solution in production.

Our H3L and H5L family are doing well in professional and we've got some design activity -- not only design activity but also products in production with the Hikvision Dahua, but also we have multiple products production with outside China customers.

So overall, professional security camera continues growing well for us.

But I will say that the growth rate for the consumer of home monitoring market is faster than professional.

Great.

And then I guess one final one.

George, you had mentioned in answering a prior question that the CV side of things would likely be, I think you said, 5 to 7 quarters out before you get the revenue ramp.

Obviously, you get design wins before that.

Is that duration kind of puts us out into the fiscal year '20 time frame?

Is that because of the markets you're addressing, the complexity of the chip and the different designs that will be necessary?

I'm just a little bit unsure as to why it would take that long.

And so any color you can give us would be helpful.

Well, I really think it's the software side.

And in fact, chip -- well, like I said, folks are sampling chip this year -- second half this year.

And also -- but in the past, the first generation of any new application takes the longest because it requires brand-new software and going through brand-new application.

And that's where I think most of the time helping our customer write their software and optimize our software and get ready for that production.

And that and for any new product, we're talking about 12 months at least.

So that's where we expect the revenue start.

That's also pretty much in line with our history.

It's always been about 9 to 12 months revenue after we start sampling.

So it's reasonably consistent.

Our next question comes from the line of Charlie Anderson from Dougherty & Company.

George, I wonder, could you quantify the memory shortage impacts this year in any way?

And then I've got a follow-up.

There really hasn't been a significant one to date.

We're starting to see the impact in Q3, but customers are still working through delivery and where they can obtain the right memory.

So I'll probably be able to, in hindsight, at the end of Q3, actually, be able to quantify it based on what our customer order rate changes are.

But as a forward-looking quantity -- quantification, I cannot do it at this point.

Got it.

Okay.

And then on CV, I heard in the script that your architectural allows you to have lower power consumption and lower die size than some of the competing alternatives out there.

I wonder if you guys could just talk a little bit about what that means in terms of pricing and what that means in terms of some of the opportunities that maybe you can handle than maybe some of the competing approaches can't handle with the power consumption angle?

Right.

In fact, power consumption impact 2 things.

One isn't definitely surprising because that give us more competitive edge.

But I think based on the last 10 years of experience selling video processing solutions, we always have much lower power consumption than our competitors.

And what I give the best benefit is to secure design win easier because it's not only the battery power limitation but also the form factor design that our customers choose to use.

When the form factor becomes smaller and smaller, lower power consumption make the design easier and the form factor smaller.

And usually that is a very critical consideration for the design wins.

So I would say that the power definitely is impacting not only our getting better ASP but also a secure -- easier for us to secure design wins.

Our next question comes from the line of Brad Erickson from KeyBanc Capital Markets.

One for Les or Fermi.

Can you share anything you heard out of the Tier 1 auto suppliers or OEMs you've spoken to regarding the type of vision architecture you're pursuing?

You called out some of the technical advantages of stereo assuming over mono.

But just curious what you've heard out of the customer base or the Tier 1s you've had dialogue with that led you to pursue that route.

Well, let me say, in general, I think as I mentioned, the feedback we've gotten is that customers like our architectural direction and they do not believe that a general-purpose platform such as a CPU or a GPU or an app processor is a good solution because it's too inefficient.

I mean, it's okay to start with, to prototype with, but in terms of ramping up volume, they don't like that.

Specifically about stereo, I think the situation right now is there are many different type of sensors that people are looking at: radar, LIDAR, image sensor.

They all have their advantages and disadvantages.

But what's clear is that you need to have some good depth-sensing capability in the system.

And stereo is a particularly low-cost and high-resolution way of doing that.

It's not necessarily going to be the only depth-sensing technology, but it's a very good way to do it that can augment other types of depth-sensing technology.

Got it.

And then just on the Q3 OpEx, George.

I think you called out that there's both CV tape-out cost in there as well as ASO qualification spending or the beginning of it anyways.

Can you just unpack both of those?

Are they roughly the same contribution?

Is one more than the other?

And then just remind us, I think you've said this a couple of times, I just want to be clear that it's probably a 4-quarter type of spending ramp for both of those?

And just where we are particularly with CV.

Yes, the biggest component of the Q3 guidance increase is in tape-out fees.

Adding 2 10-nanometer chips to the production cycle is a big chunk of it.

ASO is a smaller portion of it, but will ramp in the next year.

So we'll have more cost associated with ASO next year as we get into later phases of the chip development.

Our next question comes from the line of Quinn Bolton from Needham.

George, just wanted to follow up on the GoPro.

If I heard you right, I think you said it's $18 million, or you anticipated $18 million in the fiscal third quarter.

And if that's the case, I think it sounds like you're tracking above prior guidance for GoPro to be 3% to 6%.

If that's correct, is that just GoPro burning through inventory of older cameras, coming back and reordering?

Or can you give us some sense for that greater-than-expected strength of GoPro?

Yes, it is simply them buying products for their older cameras.

There's actually no other change associated.

I just underforecasted their needs and the length of time that they would be ramping down their old models.

Understood.

And then just quickly on the auto business.

Can you give us some sense of the split now between the auto OEM revenue versus aftermarket revenue?

Is that becoming more of a 50-50 split?

Or is it still greater on the aftermarket side today?

The aftermarket's still greater, but I would expect next year for them to be very close.

So the growth rate is obviously much higher in the OEM marketplace, and I think it's contributing to most of the growth in the marketplace for us.

I would think they would be very close to equal next year.

Our next question comes from the line of Richard Shannon from Craig-Hallum.

Just 2 quick ones for me.

Maybe we'll talk about security and the home monitoring market.

You called out most -- I think you're in your prepared remarks you're talking about strength in the United States.

Are you seeing any sense of improvement there?

And then do you also get a sense of how far penetrated these services are in the U.S.?

Are we getting close to a fuller attach rate there?

Or just any sense you get from your customers of how helpful that market is.

You are talking about U.S. market, right?

Yes, U.S. for home monitoring.

Yes.

Yes.

So I really think that the U.S. home monitoring market is still at the early stage of the -- I shouldn't say early stage, but definitely we are not having seen the full potential of the market yet.

We have been in this market for more than 2 years now and we continue to see new design wins and new customer come in and I think the growth rate continue to be very healthy and we expect the healthy growth rate will continue next year.

So overall, we believe that if you look at penetration rate and as well as potential or different applications for the home monitoring market, we believe that even in the U.S. market, you will be -- still have a couple of years growth on this.

And at the same, we haven't seen other markets outside U.S. or China having this kind of growth yet.

So potentially, there are -- for other markets, there might be potential growth, too.

I think one other thing to add to that, Richard, is that what we're seeing is multiple companies being successful.

That's not -- we haven't seen that in either the sports camera or the drone market.

But in the home monitoring market, there are many companies that are selling high volumes of product, which is a real positive for us.

That's a helpful perspective.

My last question, on the other side of security in IP professional.

You talked about for a couple of quarters at least of HEVC being a driver of your business, I think, mostly in China.

How would you characterize the relative size of HEVC to your overall IP security or maybe just China business?

And in -- by implication, how is the -- how is your 9 HEVC market growing for you?

Well, I think HEVC definitely is a growth engine.

I -- If you talk about overall market, I still think that H.264 is larger, but I -- in our script we mentioned that outside China we start seeing people start adopting the HEVC worldwide.

So I wouldn't be surprised that the HEVC market share will grow in the next few years.

But overall, in the whole market, I still think H.264 is bigger than HEVC, but we like to see -- I think we're going to see a lot more growth on the HEVC in China and outside China next year.

Our next question comes from the line of Kevin Cassidy from Stifel.

This is John Donnelly on for Kevin.

Is there an opportunity for design wins in the future for low-cost DJI drones for 2018 shipments?

Or should we just consider that consumer drones as a long-term declining market?

Well, I think our position on the drone market is that this market is going through a transition.

Basically, we believe in the long term both consumer and even the commercial drones will require a CV function.

And that we believe the single chip which can do both CV function and as well as a very good video processing technology is going to be ideal chip -- ideal solution for low-end consumer drones.

And we hope that our road map will address that very soon.

And so that in the future, we will be able to address a low-end drone market where we sell CV function.

Great.

And then also can you maybe touch on what the percentage of home surveillance camera revenue was from battery-powered camera?

It's very hard because the -- some customers use our chip for both markets.

It's hard for us to differentiate one from the other, so we can only guess but we cannot really provide an accurate guidance.

And this does conclude the question-and-answer session of today's program.

I'd like to hand the program back to Dr. Fermi Wang for any further remarks.

And thank you.

I would like to thank all our colleagues, who worked hard to deliver the computer vision technology and other products and services for our customers.

And thank you for joining today.

I will talk to you next time.

Thank you.

Thank you, ladies and gentlemen, for your participation in today's conference.

This does conclude the program.

You may now disconnect.

Good day.