FormFactor Inc

NASDAQ:FORM

Earnings Call Transcript 2024-Q3

[Operator]

Thank you, and welcome, everyone, to FormFactor's Third Quarter 2024 Earnings Conference Call. On today's call are Chief Executive Officer, Mike Slessor; and Chief Financial Officer, Shai Shahar.

Before we begin, Stan Finkelstein, the company's VP of Investor Relations, will remind you of some important information.

[Stan Finkelstein, executive]

Thank you. Today, the company will be discussing GAAP P&L results and some important non-GAAP results intended to supplement your understanding of the company's financials. Reconciliations to GAAP to non-GAAP measures and other financial information are available in the press release issued today by the company and on the Investor Relations section of our website.

Today's discussion contains forward-looking statements within the meaning of the federal securities laws. Examples of such forward-looking statements include those with respect to the projections of financial and business performance, future macroeconomic and geopolitical conditions the benefits of acquisitions and investments in capacity and in new technologies, the impact of global, regional and national health crisis, including the COVID-19 pandemic, anticipated industry trends, potential disruptions in our supply chain, the impact of regulatory changes, including the recent U.S.-China trade restrictions, the anticipated demand for products, our ability to develop, produce and sell products and the assumptions upon which such statements are based.

These statements are subject to known and unknown risks and uncertainties that could cause actual results to differ materially from those expressed during this call. Information on risk factors and uncertainties is contained in our most recent filing on Form 10-K with the SEC for the fiscal year ended December 30, 2023, and in our other SEC filings, which are available on the SEC's website at www.sec.gov and in our press release issued today. Forward-looking statements are made as of today, October 30, 2024, and we assume no obligation to update them.

With that, we will now turn the call over to FormFactor's CEO, Mike Slessor.

[Mike Slessor, executive]

Thanks, everyone, for joining us today. Driven by growth in both of our segments, FormFactor's third quarter revenue exceeded the outlook range we provided in July and set an all-time record. In addition, DRAM probe-card revenue set a third consecutive quarterly record and non-GAAP earnings per share was at the top end of the outlook range.

In the current fourth quarter, we continue to experience record demand for DRAM probe cards with contributions from both DDR5 and high-bandwidth memory, or HBM. This, together with moderate expected systems segment growth is helping partially offset a forecasted reduction in foundry and logic probe card demand.

Entering this year, it would have been difficult to predict the magnitude of the strength in DRAM and FormFactor's ability to capitalize on this DRAM strength has been a major highlight throughout 2024. The results from our leadership position and significant market share in DRAM, which is one of the core tenets of our diversification strategy, providing us broad exposure to industry growth trends like generative AI.

FormFactor's diversification strategy features a broad lab-to-fab product portfolio across Foundry and Logic, DRAM and Flash probe cards, together with our system segment products. The strategy differentiates us from our direct competitors. It makes it possible for us to compete for business across diverse demand pools at all major customers, which in turn produced relatively stable results during last year's downturn and the overall top line growth we're delivering this year.

These record levels of DRAM revenue, however, do produce a challenging product mix, and it's this DRAM rich mix that is largely responsible for the third and fourth quarter gross margins below our target model gross margin of 47% and $850 million of annual revenue.

As we've demonstrated in the past, a more balanced product mix with stronger foundry and logic probe-card demand produces gross margins approaching and in some cases, exceeding the levels of our target model. We expect to deliver stronger gross margins when growth returns to the consumer-driven high unit volume end markets that drive foundry and logic probe card spending like mobile handsets and client PCs.

Turning now to market and segment level details. As I noted, the third quarter strength in DDR5, which we anticipated, produced another record-setting quarter of DRAM probe-card revenue. HBM revenue declined from the all-time high levels of the second quarter as we experienced the expected customer digestion following record shipments of probe cards for HBM3 and HBM3E production designs in the 2 previous quarters.

In the current fourth quarter, we expect further growth in DRAM probe-card revenue. We anticipate continued DDR5 strength along with sequential growth in HBM to roughly half of our DRAM shipments with several initial HBM4 designs all across multiple DRAM customers. The beginning of HBM4 activity is significant as it puts us in a strong position as our customers start pilot production of this next major road map innovation in the high-performance memory that enables generative AI.

As we've discussed in the past, the advanced packaging processes used to construct HBM chips are powering FormFactor's current results and future opportunity. Each HBM chip, whether it's HBM3, 3E or 4 is a stack of individual DRAM die assembled with advanced packaging processes.

The disaggregation of a chip into subcomponent chiplets or tiles increases both test intensity and test complexity compared to an equivalent monolithic chip. The increase in test intensity is driven by the need for our customers to probe and test each component chiplet prior to stacking and then to probe and test the multi chiplet stack at various points during the assembly process.

The increase in test complexity is driven by HBM performance requirements that are significantly more advanced than for standard unstacked DRAM products, involving higher test speeds and more challenging thermal scaling specifications. As the industry moves from HBM3 to HBM4, both layer counts and performance requirements, such as operating frequencies and power densities are increasing dramatically.

For example, our customers' road maps feature HBM4 products with stacks up to 16 high, double that of HBM3. Increase in layer count to 16 high stacks increases the number of probe cards required for good chip out since each component chip within the stack must be tested and verified prior to stacking.

Similarly, the increase in clock speeds and power densities increases the performance requirements for the probe card necessitating higher test frequencies and more challenging thermal and power specifications.

As we've seen with HBM3 and HBM3E in 2024, our differentiated ability to meet challenging performance requirements drives both market share and profitability gains, and we expect to benefit from the same dynamics as HBM4 reaches volume production next year.

Shifting to the foundry and logic probe-card market. Here, as expected, the third quarter was similar to the second quarter, where the ramp of new mobile application processor designs together with client PC and server microprocessor designs produced steady overall revenue.

In the current fourth quarter, we expect a reduction in foundry and logic demand from the levels of the second and third quarter as high unit volume end markets in which FormFactor has a strong competitive position, like mobile handsets and client PCs remain weak.

It's our strategic priority to expand and diversify FormFactor's market position in foundry and logic, both through new customer qualifications in client PC and server applications and through new products that will improve our competitive position in high-performance compute GPU applications.

This will reduce our dependence on mobile and client PC by growing high-performance compute to achieve a more balanced opportunity profile with broader exposure to hyperscale generative AI investment, similar to the strategy we're executing in DRAM for creating HBM led exposure to AI.

In the Systems segment, we delivered the expected sequential increase in third quarter revenue as customers continue to engage us to solve the most complex electrooptical test and measurement challenges in areas like quantum computing and silicon photonics.

Co-packaged optics using silicon photonics is an exciting opportunity for FormFactor as we collaborate with leading customers in data center applications that utilize our turnkey measurement systems built on our CM300 and Summit 200 engineering probers together with our proprietary ferrous optical probes.

We're also extending our unique electrooptical portfolio by partnering with other suppliers on the equipment needed for high-volume production. These partnerships are designed to ensure that as co-packaged optics matures and moves to high-volume production in the coming years, FormFactor's leadership positions in combined electrical and optical tests will provide a new growth vector for both our systems and probe card businesses.

Finally, as validation of FormFactor's technological and market leadership, I will be delivering 1 of 4 invited keynote addresses along with executive from Synopsys, Meta and AMD at next week's IEEE International Test Conference.

In the presentation titled "From the Shadows to the Spotlight" probe's role in enabling electronics industry innovation, I'll explore how wafer test and probing are a key enabler of economically viable semiconductor manufacturing at the leading edge, especially for advanced packaging architectures that rely on 2.5D and 3D stacking of chiplets.

As we've discussed, these teams are moving FormFactor's markets and business today and underlie our long-term opportunity.

In closing, we're proud of FormFactor's all-time revenue record set in the third quarter and are excited about the future as adoption of advanced packaging drives increased demand across FormFactor's entire lab to fab product portfolio.

We continue to focus on investing in differentiated products and capabilities to drive market share and profitability gains and remain confident that when the foundry and logic market recovers that we'll achieve and then surpass our target model that delivers $2 of non-GAAP earnings per share on $850 million of revenue.

Shai, over to you.

[Shai Shahar, executive]

Thank you, Mike, and good afternoon. As you saw in our press release and as Mike mentioned, Q3 revenues were $207.9 million, $2.9 million above the high end of our outlook range and an all-time company record. Non-GAAP gross margin of 42.2% was 0.8 percentage points below the midpoint of the range. These, together with operating expenses slightly lower than the midpoint of the outlook resulted in non-GAAP EPS at the top end of the range.

Third quarter revenues increased 5.3% from the second quarter and increased to 21.2% year-over-year. The upside versus the midpoint of our outlook range was due to higher revenues in both our Probe Card segment as well as the Systems segment.

Probe Card segment revenues were $172.2 million in the third quarter, an increase of $5.4 million or 3.2% from the second quarter. The increase was driven by both higher foundry and logic and DRAM revenue. The Systems segment revenues were $35.7 million in Q3, a $5.1 million increase from the second quarter and comprised 17.2% of total company revenues up from 15.5% in the second quarter.

Within the Probe Card segment, Q3 Foundry and Logic revenues were $107 million, a 3.7% increase from the second quarter. Foundry and Logic revenues decreased to 51.7% of total company revenues compared to 52.5% in the second quarter.

DRAM revenues were a record $60.2 million in Q3, $2.1 million or 3.7% higher than the previous quarter -- sorry, higher than the previous record set in the second quarter, and decreased to 28.9% of total quarterly revenues as compared to 29.4% in the second quarter. Within DRAM, HBM revenue declined as expected from the record $43 million in Q2 to $29 million in Q3.

Flash revenues of $4.5 million in Q3 were down $0.5 million from the second quarter and were 2.2% of total revenues in Q3, as compared to 2.6% in Q2. GAAP gross margin for the third quarter was 40.7% as compared to 44% in Q2. Cost of revenues included $3 million of GAAP to non-GAAP reconciling items, which we outlined in our press release issued today and in the reconciliation table available in the Investor Relations section of our website.

On a non-GAAP basis, gross margin for the third quarter was 42.2%, 3.1 percentage points lower than the 45.3% non-GAAP gross margin in Q2 and 0.8 percentage points below the midpoint of our outlook range. The decrease as compared to Q2 at higher revenue levels, was, as Mike indicated, mostly the result of a less favorable product mix in both segments. As compared to the midpoint of our outlook range, gross margin was lower mostly due to unexpected quality-related costs in our Systems segment.

Our Probe Card segment gross margin was 42.3% in the third quarter, a decrease of 2.8 percentage points compared to 45.1% in Q2. Our Q3 Systems segment gross margin was 41.5%, a decrease of 4.7 percentage points compared to 46.2% gross margin in the second quarter. Our GAAP operating expenses were $66.9 million for the third quarter as compared to $69.4 million in the second quarter.

Non-GAAP operating expenses for the third quarter were $59.3 million or 28.5% of revenues as compared with $60.9 million or 30.8% of revenues in Q2. The $1.5 million decrease relates mainly to lower performance-based compensation.

Company noncash expenses for the third quarter included $8.9 million for stock-based compensation, $1.3 million lower than in Q2 due to grant for features, $0.6 million for the amortization of acquisition-related intangibles, similar to Q2 and depreciation of $7.6 million, slightly higher than in the second quarter.

GAAP operating income was $17.9 million for Q3, similar to the GAAP operating income of $17.8 million in Q2. Non-GAAP operating income for the third quarter was $28.3 million compared with $28.5 million in the second quarter, a decrease of $0.2 million or 0.7%.

The increase in revenue and lower operating expenses were offset by the decrease in gross margins. GAAP net income for the third quarter was $18.7 million or $0.24 per fully diluted share compared with a GAAP net income of $19.4 million or $0.25 per fully diluted share in the previous quarter.

The non-GAAP effective tax rate for the third quarter was 13.4%, 2 percentage points lower than 15.4% in the second quarter. We continue to expect our annual non-GAAP effective tax rate to be between 14% and 18%.

Third quarter non-GAAP net income was $27.2 million or $0.35 per fully diluted share, similar to Q2 and at the top end of our outlook range. As Mike said, we remain committed to achieving our target model, which produces $2 of non-GAAP earnings per share at $850 million of revenue with 47% gross margin and 22% operating margin.

While we expect fluctuations quarter-over-quarter chiefly due to product mix changes, achieving the 47% gross margin requires a more favorable product mix, mainly a larger contribution from our higher-margin foundry and logic market.

Moving to the balance sheet and cash flow. At quarter end, total cash and investments were $360 million, a decrease of $5 million in Q2. We generated free cash flow of $20 million in the third quarter compared to $14.2 million in Q2. The main reasons for the increase in free cash flow was higher operating cash flows primarily driven by greater noncash expenses of $2.4 million and lower outflows for our working capital of $3.1 million, partially offset by higher capital expenditures of $0.5 million.

We invested $8.9 million in capital expenditures during the third quarter compared to $8.4 million in Q2. There is no change in our expected CapEx range for 2024 of $35 million to $45 million. And with 3 quarters behind us, we now estimate to be closer to the midpoint of this range. At the end of the third quarter, we had one term loan remaining with a balance totaling $14 million.

Regarding stock buyback, during the third quarter, we used $16.9 million to buy back shares under the $75 million 2-year buyback program that was approved in Q4 2023. At quarter end, $36.6 million remained available under that authorization. As a reminder, the main purpose of the share repurchase program is to offset dilution from stock-based compensation.

Turning to the fourth quarter non-GAAP outlook. We expect Q4 revenues of $190 million, plus or minus $5 million, with a slight increase in DRAM and Systems revenues more than offset by a decrease in foundry and logic.

The lower revenues as compared to Q3 are expected to result in a non-GAAP gross margin of 41%, plus or minus 150 basis points. At the midpoint of these outlook ranges, we expect Q4 operating expenses to decrease to $56 million, plus or minus $2 million, $3 million lower than Q3, mainly due to lower performance-based compensation.

Non-GAAP earnings per fully diluted share for Q4 is expected to be $0.29 plus or minus $0.04. A reconciliation of our GAAP to non-GAAP Q4 outlook is available on the Investor Relations section of our website and in our press release issued today.

With that, let's open the call for questions. Operator?

[Operator]

[Operator Instructions] Our first question comes from the line of Krish Sankar of TD Cowen.

[Sreekrishnan Sankarnarayanan, analyst]

Mike or Shai, I had two of them. First one, just curious, your large U.S. customer at about 17% of revenues in September is well telegraphed about their weakness. I'm kind of curious, have you started seeing weakness there in your December guide? Or do you think it's a calendar '25 event? And how to reconcile their weakness to what you would expect from them in calendar '25 in terms of revenues?

[Mike Slessor, executive]

Yes. I mean calendar '25, remember, our lead times are pretty short, well within a quarter and certainly for the high-volume foundry and logic businesses well within a quarter. So our visibility doesn't really extend into '25.

Having said that, the overall foundry and logic market, we are seeing weakness across various aspects of it. As I said, in the prepared remarks, client PC, mobile handset are all, when I compare our fourth quarter outlook to our third quarter actual results, we're seeing softness in those two end markets.

That's one of the reasons why it's so important for us to qualify in some of these other AI-driven applications like GPUs to make sure that we're diversifying our Foundry and Logic business, which is our largest business and is expected to grow significantly.

We got to make sure we got the right market footprint there. But right now, foundry and logic outlook for 25 don't have significant visibility there. But in the fourth quarter, seeing pockets of sequential reduction in both mobile and PC.

[Sreekrishnan Sankarnarayanan, analyst]

Got it. Got it. That's very helpful. And then I just wanted to touch base on the HBM side. It looks like HBM is probably going to grow again after one quarter digestion in September. A, is the true? And B, what caused the digestion? And obviously, there's a lot of optimism on HBM revenue growth for the industry into next year. Do you worry about, again, another pocket or another quarter where you might see a digestion. Just kind of curious what caused it in September and what is the likelihood it will not happen again next year?

[Mike Slessor, executive]

Yes. So if we track HBM through the quarters of 2024, Second quarter, very strong results. We did convey to you that we expected some digestion in the third quarter. That happened. That can often be the case. Remember, our HBM revenue and overall market share really concentrated with the leading provider of HBM to the industry.

As we move into the fourth quarter, the growth is driven by a couple of components, the sequential growth off of Q3. And yes, there is sequential growth expected in HBM off of Q3. Some of it is the resumption of spending by the leading HBM provider, but we're also beginning to see some diversification and activity at one of the other DRAM manufacturers, one of our other DRAM customers as they begin to build their DRAM market share.

[Operator]

Our next question comes from the line of Charles Shi of Needham & Company.

[Yu Shi, analyst]

Mike, I want to ask maybe a little bit higher level question about HBM, plus just a few days ago, Teradyne, it mentioned about -- well, they talked about the next year, they think the tester TAM is flat to down next year. But since you're not quite -- your products are complement to testers, but it's not the same capital goods as testers. Are you -- how do we reconcile, let's say, testers can does go to flat to down next year? And how do we think that -- why would we think that maybe probe card can still grow next year?

And maybe related to this, I think this is -- you spent some time, talk about the technical details about HBM3, 3E and 4 that transition, how that does to probe card. Can you kind of help us understand whether HBM -- transition to HBM 4, does that -- does that help with the overall probe card TAM growth because it looks like it may not really help testers because testers are quite agnostic. Hopefully, you can touch these points as well.

[Mike Slessor, executive]

Yes. So you alluded to it in the way you asked the question, Charles. Remember, probe cards are a consumable that's specific to each customer chip design or each customer mask set. So if we take the HBM example, as a customer moves not just from HBM3 to 3E to 4, that certainly drives demand for new probe cards. But even inside those generations of HBM, there are multiple flavors of HBM. And so you can think about 3 or 4 major HBM3 designs, all requiring different probe cards.

Now all of those chips are going to end up being tested by the same tester. The purpose of a probe card is to customize the tester, so it can test these different chips. And so that's how you reconcile the different growth cycles between probe cards and ATE or testers.

We've seen this before in the past, right? In foundry and logic, there's been large dugs of tester investment that then cool off and we see some strong probe card activity as customers release new chip designs to be able to utilize those testers. So often through history in both foundry and logic and memory, we've seen that decouple it.

As to the transition from HBM3 to 3E to 4, we do this -- view this as being helpful to FormFactor, not just because of an increase in the available market. Remember, almost certainly when we go from HBM3 to 4, the number of die increases in the stack. I mentioned in the prepared remarks, going from 8 to 16, that's all other things being equal, going to require more probe cards.

But I think the other thing that's interesting is -- and that's what we generally refer to as test intensity. I think the other thing is when you look at customer road maps, as they go from 3 to 4 there's some significant increases in things like speed and power density, which drive a more capable probe card, which typically we're able to get a higher value, a higher ASP for.

So I hope that addresses the 2 elements of the question. But remember, probe cards are a device-specific consumable, different bicycles than capital equipment like ATE.

[Yu Shi, analyst]

Okay. Maybe I just want to ask another question, a little bit of clarification. I look at your geographical breakdown for the quarter. It looks like the revenue from China ticked up a little in Q3, Mike, if you could provide some insights into what's behind that number? Because we do understand that your China revenue can split between multinational and the domestic China. What do you see there?

[Mike Slessor, executive]

Yes. So it's an interesting question. Obviously, China is on lots of people in the industry's minds right now. We've been pretty clear that we think that our domestic China revenue, which still represents a significant fraction is going to go to 0 over time, some time frame because of the geopolitical issues, because of our presence as a U.S. supplier governed by export controls from the Department of Commerce.

Having said that, we did see a tick up in domestic China revenue, as you noted, in the third quarter. It's in DRAM. And I think I'd characterize it probably is almost a last time buy with one of our leading DRAM customers in China. We don't view it as sustainable largely because of the geopolitical environment.

To remind you, that's one of the reasons why we divested our China operations earlier this year. We're still able to do business there, but doing it through the company that bought our China operations as a distributor. So I'd classify us as being opportunistic. Here, we had an opportunity in the third quarter to go win some business, but I wouldn't model that through 2025.

[Operator]

Our next question comes from the line of Craig Ellis of B. Riley Securities.

[Craig Ellis, analyst]

Guys, congratulations on some of the record set in the quarter. Mike, I wanted to start just by following up the point that I thought you made well in responding to Charles' question around really intensity and complexity that are going up and just see if I could get you to reflect a little bit and comment on calendar 2025 for high-bandwidth memory.

And the question is really this. As you look at the drivers that are fueling high-bandwidth memory strength, the increased number of chips that are required for every GPU that's shipping as we go to more stacks and higher stacks. Are you seeing any signs that customers are able to get yield dynamics to a level where probe card intensity would decrease? Are we just running too fast from 3 to E to 4, et cetera, to get that kind of yield improvement?

[Mike Slessor, executive]

Yes. So it's a great question, Craig, because obviously, one of the things that's driving the very high test intensity with HBM is fairly low yields. The chiplets die themselves are pretty standard DRAM die. So they're not the yield issue. It's the complexity of the advanced packaging and assembly and stacking of those die that's causing the issues. And no question that was driving up test intensity.

I think to your point, the industry is moving so quickly to both move from 3 to 3E to 4, but also to build capacity, that I don't think there's going to be a really significant yield improvement trajectory.

Yes, it will get better. But when you look at the complexity, especially as you increase the number of die in the stack, it's simple arithmetic, right? 16 die is going to have lower yields than 8 die, even if you've got something close to known good die going into that stack.

So I think yields will improve for sure, but I think the overpowering increase, overdriving increase in test intensity really is associated with the dynamic of the industry moving so quickly to try and build both capability through the 3 to 3E to 4 transitions, but also building capacity as fast as it can to support the GPU demand, which, of course, is all being pulled through by hyperscaler investments in generative AI.

[Craig Ellis, analyst]

Yes, that's helpful. The tough math of many chips and many tech transitions that's happening. Moving on and then touching on the foundry and logic item. So in a few of the recent calls, you've noted that there was particular customer strength from the U.S.-based customer, and you cautioned that there might be some digestion or a breather coming and it sounds like some of that may actually be at play in the fourth quarter? Or is that, in fact, not the case? And have your views changed there? And if so, what's caused that view to change?

[Mike Slessor, executive]

I think if I look more broadly, any given customer, right, is going to have quarter-to-quarter ups and downs, right? There investing and ramping in individual designs or sets of designs, we saw this in Q3 with HBM with our major customer there. So you're going to have quarter-to-quarter fluctuations with any customer. That's one of the reasons why we're so committed to this diversification strategy and doing things like making sure we're qualified at all leading customers.

In this case, right, the qualification of the large fabless CPU and GPU customer remains a very high priority for us. So regardless of customer, I would expect to have quarter-to-quarter fluctuations. It's really the nature of the business.

[Craig Ellis, analyst]

And if I could sneak in one last one. You did mention the opportunity with GPUs. Can you talk about new customer penetration opportunities and progress made in the third quarter and potential for things to close as you get into 2025?

[Mike Slessor, executive]

Yes. So a couple of key customer qualifications that we've kept you at least at a high level update on. One is making sure we're qualified at the large fabless CPU, GPU manufacturer. We've continued to make good progress there, have worked with that customer to focus our qualification on an area where we're showing a differentiated capability. A smaller piece of the business, a specific application, but gets us the toehold in, and we do expect to be qualified there here in the fourth quarter.

For the major GPU manufacturer, who's already a pretty significant customer for us, based on the networking business and the silicon photonics partnership with co-packaged optics. The focus there is really on getting qualified and penetrating their advanced packaging business as they start to package GPUs together with HBM on the CoWos advanced packaging technology. And that's something we continue to execute on largely through the world's largest foundry because they're doing all of the packaging and test. But it's something we'll continue to keep you updated on as we continue to make progress.

[Operator]

Our next question comes from the line of David Duley of Steelhead Securities.

[David Duley, analyst]

Mike, I just had a couple of questions. You've kind of answered part of them, but as far as your pause in the third quarter, do you think that had something to do with your largest customer migrating from 3 to 3E and then to 4 or from 8 -- from 8 stacks to 12 stacks? Or -- I'm just kind of curious if that had something to do with your pause?

[Mike Slessor, executive]

Right. Well, remember, I think you're talking about HBM specifically, but I want to remind people for FormFactor overall, there was no pause in the third quarter. We set an all-time company revenue record. So we do want to keep that in context. There was a reduction in HBM revenue that we expected, again, through digestion. And as I answered Craig's question, this happens all the time, right?

Customers will take large shipments for the designs they're going to ramp and then have these periods of digestion. So I don't know that you can tie it to any particular transition, but it's, again, one of the reasons why this diversification strategy that we've run for a decade plus is so important.

If you look at any customer's revenue, it's going to be lumpy quarter-to-quarter, as they take shipments and then use those shipments of probe cards to produce wafer outs on a given design. Those happen for a quarter or two quarters and then that customer resumes their buying as they transition to the next design.

So when you look at our revenue, there's a whole bunch of lumpiness with each customer. You can chalk it up to some of the node transitions to some of the pauses to some of the changes, but that's why we're committed to this broad-based diversification strategy across all the leading customers and all the important markets in the industry.

[David Duley, analyst]

Okay. And then clearly, your largest customer on their conference call is talking up significant DRAM bit growth, particularly in HBM. You have them moving from 8 high stacks to 12 high stacks, perhaps 16 high stacks. Yields are definitely going down with that. You have the Blackwell transition with a number of die going up per GPU is significant. Is -- can you -- with all of these things, wouldn't you think your HBM revenue next year would grow significantly because it's not just unit volume driven, it's a bunch of different designs also on top of it?

[Mike Slessor, executive]

Yes. No, those are good observations. One of the things in laying out those points, there's a lot of moving parts as well, right? Yield design changes, wafer outs and obviously, does the HBM -- or at what level does the HBM output get driven by continued hyperscaler AI spending because that's the direct driver of it.

I think with all those factors, our baseline assumption is that our HBM business is going to grow, partially because if you think about how our customers are approaching the market, they're essentially saying they're all sold out through 2025 and they're adding capacity at a pretty good clip. And doing things like move from 3 to 3E to 4, which are all different sets of probe cards. So I think it's a reasonable assumption that if yields don't dramatically improve, as long as we're driven by those market dynamics, it's a reasonable assumption that our HBM business can grow.

[Operator]

Our next question comes from the line of Denis Pyatchanin of Stifel.

[Denis Pyatchanin, analyst]

This is Denis on for Brian Chin. Given that you've alluded to the HBM spending reduction at the leading maker being over, directionally, how confident would you be that HBM revenue is higher in 2025 versus 2024? And on a related note, over what time frame could we see the other 2 HBM customers add more significant revenue to your top line?

[Mike Slessor, executive]

So I think I answered the first part of that question for David. It depends on a lot of the underlying assumptions. But as long as hyperscaler AI spending continues, that drives GPU growth, that drives continued attach and growth of HBM. So I think there's a reasonable chain there. And of course, if you think about the underlying transitions next year in going to HBM3 and then HBM4, which all require different probe cards, I think there's a reasonable case to be made for growth there.

If you think about the customer base and we pointed this out in the prepared remarks and I think in one of the answers to the questions, we are seeing the customer base broaden a little bit. Now if you look at HBM market share, not probe card market share, but who's producing the HBM bits in the industry, it is very heavily indexed to a single one of our customers. So I think it's appropriate that our revenue will also be indexed to that customer.

In the fourth quarter, because we're seeing -- well not because the fourth quarter, some of the growth -- sequential growth in HBM we're seeing is driven by a second DRAM customer coming online, and becoming a more significant part of the revenue. But I would expect as long as those market share dynamics for our customers supplying HBM continue, we're going to continue to see some pretty heavy customer concentration in it.

I think the only other point I'll make is you talked about the dip in spending as being over. I'll again remind people that any one customer moves up and down quarter-to-quarter. And so I would expect quarterly fluctuations in digestion, again, at the risk of being repetitive, that's why it's so important for us to broaden our customer and application base as part of executing our diversification strategy.

[Denis Pyatchanin, analyst]

That's a really great detail. And for my follow-up, could you provide some additional color related to your GM guide for Q4 being a little bit lower versus Q3, the mix, what segment markets are doing, et cetera?

[Mike Slessor, executive]

Yes, I'll let in Shai.

[Shai Shahar, executive]

Yes, I'll take that. So if you look at the revenue outlook, the midpoint of the outlook for Q4 is $190 million, which is below the record of $208 million, we did in Q3. And this is the main contributor to the decrease in gross margin, we see Q3 to Q4, it's a 1.3 percentage points lower at the midpoint of the outlook. It's really about volume.

If we look at the mix, there is no major difference on the impact of the mix on the gross margin, although we see foundry and logic down, and we expect DRAM to go up a little bit, but we don't see a major impact on the mix of the gross margin. It's really about volume.

[Operator]

Our next question comes from the line of Christian Schwab of Craig-Hallum.

[Christian Schwab, analyst]

Mike, I know you mentioned high bandwidth memory 4, but high-bandwidth 3E should still be about 80% of the bits shipped next year, right?

[Mike Slessor, executive]

Yes. So 3 is definitely going to dominate the bits shipped in '25, next year. But remember, we're going to lead the bit shipped because our business, the probe cards have to be in place, the testers have to be in place, for those chips to then be produced, ramp and be shipped. So we're going to lead output. And the transitions are happening so fast from 3 to 3E to 4, that late '25 into '26, we'd expect HBM4 to start to take over, and that's why we're seeing the activity now so that everything is ready on what's a pretty complex process.

[Christian Schwab, analyst]

Great. Because my work says about 80% of the bits will be 3E, I just want to confirm that. Will we ship to all 3 high-bandwidth memory guys in '25?

[Mike Slessor, executive]

Well, we have shipped, and I said this in the prepared remarks, we are shipping HBM cards to all 3 major DRAM manufacturers. Now it's, again, heavily indexed towards one because one customer is producing the lion's share of the HBM bits in the industry. But I think a bright spot in Q4 is we are starting to see a significant contribution in our revenue from a second DRAM customer for HBM.

[Christian Schwab, analyst]

Okay. And then my last question, with Blackwell needing probe cards and the other one is not, could that customer, whether it be location in Taiwan, however, you're going to label that customer. Is that somebody you would anticipate becoming a 10% plus customer sometime in '25?

[Mike Slessor, executive]

It's sort of -- it's an interesting case because it depends how it works, right? Fabless customers have different procurement models for probe cards. Sometimes, they outsource it to particularly the foundry. And so the foundry is actually the customer that will report in our revenue.

Sometimes it's the fabless customer themselves. That one, I suspect is going to be revenue attributed to the foundry, given their ownership of the overall assembly -- the CoWos assembly and test process -- and they haven't been a 10% customer for a while. They've been awfully close, sort of high single digits. I would expect some of these new advanced packaging opportunities like CoWos to drive them up into double digits and have them become a 10% customer at least sometime in 2025.

[Operator]

Our next question comes from the line of Tom Diffely of D.A. Davidson.

[Thomas Diffely, analyst]

Maybe just a follow-up, another high bandwidth memory question on the back of Christian's question. Given your unique capabilities across doing different types of forms and formats. Would you expect this share to be fairly strong at all 3 makers of HBM over time? Or are there certain factors that go into perhaps not making that sell?

[Mike Slessor, executive]

Well, I think a couple of things, right? Customers can always make choices on where they put their test dollars, right? And there's times we've seen -- and we've seen this in standard DRAM where some customers have chosen to emphasize other areas of performance, maybe where we're not as strong, and that's caused share to go to our competitor.

I think for HBM, we're pretty focused on some of the core enabling technologies, things like high speed. This is an area that over the years we've invested in. Part of that was the acquisition of Cascade Microtech in 2016 to get us some RF know-how.

And although that was really focused on the RF business, some of that know-how has come over to our DRAM road map as well and is driving some of the competitive advantage for high-frequency test and enabling HBM. I think realistically, our share with their current leading customer is pretty strong. And I would not expect to be able to drive that degree of share concentration.

But I would expect -- you've got 2 suppliers worldwide who can do this. Both of us understand we're capable and are going to behave rationally as we serve the overall growth of the DRAM market driven by HBM, but I'm not sure that I'd expect the degree of share at the other two that we have with the leading HBM customer right now.

[Thomas Diffely, analyst]

Okay. No, that's fair. So Mike, I'm curious, what are you seeing in the kind of traditional DRAM market today? Have you seen signs of recovery there?

[Mike Slessor, executive]

I don't know that I say recovery, I am -- we are surprised at the strength in DDR5, right? If you look at Q3, for example, we had HBM step down, but -- sequentially, but we still set a DRAM record above $60 million in quarterly revenue. And we expect DDR5 to be a significant part of the growth off of that record here in Q4. So whether you attribute that to a cyclical recovery in DRAM, whether it's customers just managing their mix, I don't know. But the strength in DDR5, given end market DRAM things like spot pricing and inventories is a little bit surprising.

[Thomas Diffely, analyst]

Okay. Great. And then finally, Shai, you talked about the systems business having some cost issues. If you could maybe just elaborate on that, that would be great.

[Shai Shahar, executive]

Sure. Yes. So I talked about unexpected quality-related costs in Q3. These are things like warranty and inventory scrap, inventory write-offs. So that was the main reason for the additional expenses we saw in Q3. But there's no change in our goal and target for the systems business to achieve high 40s, low 50s gross margin for that segment.

[Thomas Diffely, analyst]

Okay. Maybe just to clarify, that was just a one quarter issue? Or do you see it dragging out for a few more quarters?

[Shai Shahar, executive]

These specific issues are things we identified and are limited to that quarter. But we need to remember that there could be some mix shift within the system segment as well. And we did sell the FRT business that was a higher margin within the systems business, but there's no structural change there. And then we think we're going to get back to the high 40s, low, 50s.

[Operator]

Our next question comes from the line of Gus Richard of Northland Capital Markets.

[Auguste Richard, analyst]

Looking at the gross margin in the quarter is down 280 bps sequential the logic versus DRAM mix is about the same was flash. And what changed was HBM versus DDR5. And I'm just wondering, can you talk a little bit about the decline in gross margin? Is that just purely a function of mix change in DRAM? And is there a premium for probe cards for HBM?

[Shai Shahar, executive]

Indeed, it was from the change of mix on higher volume, that's obvious that it's because of mix. And the HBM probe card is a differentiated product. We invested a lot in R&D in it. It's differentiated, and that's why within that band of DRAM gross margin, HBM is on the high end of that range.

[Auguste Richard, analyst]

Okay. All right. That makes sense. And then just any predictions on when you expect co-packaged optics to sort of hit volume production or get out more out into the field?

[Mike Slessor, executive]

Yes. I don't think our timing view has changed on that much, right? It's probably late next year, early into 2026. But given all of the work that has to be done between now and then to make this production worthy, there's a lot of activity right now. And that's one of the reasons why we keep updating people on our progress.

We have more than 100 systems in labs and pilot lines around the world, but those are really low-volume applications. Where our focus is now partnering with both leading customers in the industry and other equipment suppliers who have pieces of the high-volume ready things like testers and handlers and probers, we're continuing -- that's a huge focus for our systems business right now.

And we've got plenty to do, because it's very complex and combines both electrical test and optical test, it's going to take us until the end of '25 and into '26 to really together make this production worthy.

[Operator]

[Operator Instructions] And, as there appear to be no further questions in queue, I would now like to turn the conference back to Mike Slessor for closing remarks. Sir?

[Mike Slessor, executive]

Great. Thanks, everybody, for joining us today. Obviously, an interesting time in the industry as we continue to execute on the great position we're in with HBM. Looking forward to seeing you at some of the end of the year conferences, if you're able to attend. And if not, as we move into 2025. Stay safe and healthy and talk to you soon. Bye-bye.

[Operator]

This concludes today's conference call. Thank you for participating. You may now disconnect.