Advantest Corporation (6857) Earnings Call Transcript & Summary

Good afternoon, everyone. To start, I'll walk through a summary of the financial results for this period, which you can see on Slide 3 of the presentation. We're happy to report that our Q3 results came in above the internal projections we set in October, setting a new record for quarterly sales. Because of demand pull-in for SoC and memory test systems serving AI-related applications, the temporary digestion we had expected didn't happen, and that gave us a stronger-than-expected top line. Q3 operating income and net income also beat projections, helped by product mix improvement and only moderate increases in SG&A. With the solid performance through the first 9 months period and solid visibility into Q4, we are raising our full year FY 2025 guidance as follows: sales of JPY 1.07 trillion, operating income of JPY 454 billion, net income of JPY 328.5 billion and EPS of JPY 452.34. Looking ahead, with AI-related demand remaining strong, we expect sustained growth in the tester market in calendar year 2026. We continue to push forward our capacity expansion initiatives to meet robust demand and are executing strategic priorities outlined in our midterm plan to evolve into an integrated test solutions provider.

By putting money into growth initiatives—especially beefing up our R&D capabilities—we're setting the company up to keep leading over the long term. Now I'll take you through the business environment around our company and what we see ahead for the tester market, which is laid out on Slide 12. Before I give the tester market size update, let me start with how we view the business environment. Uncertainty is still hanging around because of ongoing geopolitical risks and the chance of sharp swings in exchange rates, but AI applications remain the primary growth engine for the semiconductor market, within an AI market that is evolving quickly and dynamically. For calendar year 2026, our estimate for the SoC tester market size is $8.5 billion to $9.5 billion, which works out to roughly 30% year-on-year growth at the midpoint. We expect that growth to be driven mainly by continued momentum in AI applications, lifting unit volumes and adding greater device complexity—both of which push SoC tester demand higher. On memory testers, we put the calendar year 2026 market size at $2.2 billion to $2.7 billion, equal to approximately 20% year-on-year growth at the midpoint. Across a broad range of memory devices, we anticipate investment appetite for testers will stay solid.

Thanks.

Thanks a lot, Doug. Now I'd like to kick off the question-and-answer portion. We'll begin with Mr. Yoshida from CLSA Securities Japan.

Yoshida here from CLSA. Congrats on the extremely robust results and on the guidance as well. Would you mind giving us a little more granularity on how the SoC tester market TAM breaks down? For instance, what does the AI-related exposure look like when you compare 2025 with '26? Should we expect that order growth heading into '26 comes solely from AI? Also, can this AI demand be split between the volume contribution and the intensity — that is, the test hour contribution? And on top of that, I'd also like to ask about the market share assumptions you're making for 2025 and 2026.

Thanks for that question, Yoshida-san. Doug here. First off, looking at it from an SoC standpoint, the biggest subsegment we keep seeing is clearly the high-performance compute, or AI, space. In '25, that accounts for roughly 80% of our SoC business, in our view, and we expect that to carry on into 2026. So as I mentioned during my prepared remarks, that will remain the primary driver. When it comes to test content, or test intensity, we keep seeing complexity translating into additional test insertions and longer test times. That trend, we believe, will persist. Then lastly, regarding the market share portion of your question, we'll be putting out the final calendar year '25 market share data in April. Still, speaking generally, we do think we gained a fairly substantial amount of market share in SoC, and we expect that trend to carry on into '26 as well.

Interpreted Moving on, we have Mr. Yamamoto from Mizuho Securities.

You said that roughly 80% of the SoC business comes from HPC/AI-related business. I'm wondering whether you could break that down further between GPU and ASIC for 2025. And it would also be appreciated if you could share your outlook toward 2026.

All right, and yes — many thanks, Yamamoto-san. Doug here once more. Beyond simply naming the computing and communications categories, we don’t provide further detail. So essentially, that 80% does include high-performance compute, the AI accelerators, and, most commonly, the application processors. And it should be reasonably clear that, at present, the bulk of that sits on the HPC side, and we expect that to remain the case. Now, regarding the GPU ASIC portion of what you asked, I’ll actually turn to Sanjeev to address that.

Sanjeev Mohan here. Thanks for asking. Doug already noted that our market position is quite robust, and we see it still expanding. During FY '25, a clear move toward custom ASICs took place. Naturally, given our share and footprint, revenue from those clients increased as well. That said, GPUs remain extremely, extremely powerful and are still the top growth engine for the company. Regarding the outlook, we anticipate that pattern will persist. We'll see additional activity with custom ASIC suppliers, yet the GPU side ought to stay quite healthy. That is as much as I'd say right now.

After that, we’ll go to Mr. Wadaki at Morgan Stanley MUFG.

You've reported solid results each quarter, and every time it catches me off guard. My question is along the same lines. You said the TAM for '26 is going to increase roughly 30%. What I'd like to get clear on is how those drivers split out — say, the 3 factors: GPU test time, ASIC growth, the custom ASIC business, and volume growth. Given those 3 breakdowns, how would you order the contributions to the possible TAM expansion for '26?

Thanks, Wadaki-san. I think the main factor here is simply going to be the extra unit volumes for the custom ASIC. As we've mentioned previously, the test content across what you'd call a typical custom ASIC architecture is quite comparable to what you see in the GPU accelerator products. So those carry a substantial amount of test content as well. What really shapes the trajectory, though, is the unit volume combined with greater uptake of the custom silicon relative to traditional GPU.

Moving on, we have Mr. Nakamura from Goldman Sachs Japan.

Congrats again on the strong results. I also wanted to ask about your calendar '26 market outlook for both SoC and memory. What scenarios are you assuming for both the low end and the high end of your guidance for SoC and memory? And it looks like this strength will continue, particularly on the AI side, for a while. So, based on the conversations you're having with your customers, is there any visibility you can discuss on calendar '27 and beyond?

Alright, thank you, Nakamura-san. This is Doug speaking. Let's begin with the SoC market — on the demand side, things are really strong. So when we consider our assumptions at the low end and the high end, much of that comes down to pure supply-side economics. What I'd say is that everything from wafer starts to CoWoS capacity to memory availability will essentially determine that low and high range. As a result, it's somewhat beyond our company's control. And as long as sufficient wafers, advanced packaging, and memory are available, a scenario at the higher end of that range is possible. Then on the memory side, it's somewhat different. How much memory can be supplied will still be supply related. But here, as you know, new factories are being built, and really by all of the DRAM suppliers. Those facilities are coming online later in '27 and into '28. And until that point, there's a great deal of simply reshuffling wafer starts — reallocation toward the AI market. So depending on how effectively the memory suppliers can optimize those wafer starts toward the high-end areas, that will really determine where we wind up in this range.

Can you see anything at all looking out toward calendar '27 and past that point?