Open in interactive viewer → charts, metric popovers & call review

Advantest's Third Pillar: CPU Demand Meets a Capacity Race

The tester leader raised guidance on faster capacity, not a bigger market—and admitted DDR inflation is now a visible margin tax.
6857.T · Earnings Call · 2026-10-28

The upgrade is composition, not just demand

Advantest's 2026-10-28 call had the headline everyone expected: a raised outlook, another strong quarter, and a confident tone on AI compute. The more interesting detail is what sat underneath it. Management lifted its SoC tester revenue outlook while leaving its calendar 2026 SoC tester market forecast unchanged. That distinction matters. The raise is not a mark-up of the demand pool; it is Advantest shipping more systems into an existing pool because its capacity expansion is moving faster than planned. CEO Doug Lefever was explicit about the demand mix: “the biggest factor is the CPU business... agentic AI” — Doug Lefever, CEO · 2026-10-28. That is a new third pillar. In prior calls, the demand story was GPU plus custom ASIC. In April 2025, Lefever framed it as “traditional GPU players as well as the upcoming custom ASIC and CSP players” — Douglas Lefever, Representative Director, Senior Executive Officer and Group CEO · 2025-04-26. In January 2025 he was “very bullish” — Douglas Lefever, Representative Director, Senior Executive Officer and Group CEO · 2025-01-29 on custom ASIC. CPU was not the lead. Now agentic AI has made CPU volumes and inference ratios a third layer on top of GPU and ASIC. Sanjeev Mohan explained why that matters for test: “test times for CPU are not as large as GPU or custom ASICs... but they're also not small... volumes of server devices is quite large” — Sanjeev Mohan, Executive/Management · 2026-10-28. This is not just a share story in the CPU market; it is a test-intensity story.

Share gains, and the 70% question

Advantest also sharpened its market-share narrative. Lefever said the company went from 56% SoC share in 2024 to 66% in 2025, and repeated that a path above 70% is credible: “we have growing confidence that we will be able to meet that 70% threshold and maybe even beyond in the near future” — Doug Lefever, CEO · 2026-10-28. That is a bigger claim than the usual boilerplate. It implies gains in 2026 and 2027, not a flat year followed by a hope. The SoC tester franchise is no longer just riding a customer's GPU ramp; it is picking up accounts where it was not the incumbent. The x86-versus-Arm debate came up, and the answer was refreshingly indifferent: the same 93,000 platform sits at both sets of customers, so Advantest does not need to pick the CPU architecture winner. That is a useful contrast with the prior call cycle, when analysts worried about Arm based CPU share shifting the tester landscape. The concern has not disappeared, but the current answer is that share of architecture is less important than share of test sockets.

investors are afraid that this 'special demand' may level off from July onward

Tetsuya Wadaki, Analyst · 2025-04-26
That was the fear in April 2025: that AI server yield problems had pulled forward tester demand and would fade. The current call does not dismiss the risk, but it reframes the demand as structural—GPU, custom ASIC, and now CPU—rather than a one-time yield anomaly.

Capacity is the new bottleneck

The operating story is now a production story. Advantest originally targeted 10,000 SoC tester systems by the end of fiscal 2028; on this call Lefever said that timeline has to be pulled forward and that the next step is a doubling: “I believe the next step will be another doubling from 10,000 to 20,000” — Doug Lefever, CEO · 2026-10-28. That is a capital commitment, not just an outlook. It also explains why the guidance raise was weighted toward the third and fourth quarters: more shipment capacity, not a sudden re-rating of the market. The prior-cycle comparison is instructive. In January 2025, management was “stress testing our production” — Douglas Lefever, Representative Director, Senior Executive Officer and Group CEO · 2025-01-29 to see how far supply could go. Now the constraint has moved from demand visibility to physical throughput—and to component allocation. Lefever was candid that memory scarcity is a real input risk: “Memory is scarce in the world right now... we're not immune to that at all” — Doug Lefever, CEO · 2026-10-28. The saving grace is that Advantest's customers need testers badly enough to help prioritize memory supply, a dynamic that echoes the 2021-22 shortage playbook.

The memory tax

The most important margin admission was about DDR pricing. Analysts noticed gross margin looked strong, but Lefever corrected the framing:

We felt the pain of DDR pricing. We consume a lot of DDR in both our SoC and memory testers. A good part of the first half or second half operating margin decrease has to do with that step up in pricing.

Doug Lefever, CEO · 2026-10-28
He also noted that first-half margins benefited from an inventory reversal that will not repeat, and that the business is settling into the low 50s operating margin range. That is the tension in the quarter: volumes and share are going up, but memory price inflation is a direct cost headwind. This is where the cross-company contrast gets interesting. For a memory maker like SK Hynix, memory pricing is a tailwind; for Advantest, it is a tax. The recent reporter set shows SK Hynix leading with AI infrastructure and demand growth, while Advantest is talking about absorbing higher DRAM costs. Same supply chain, opposite margin sign.

What the tape and the optionality say

The global tape is not confirming the AI infrastructure story with much enthusiasm. High bandwidth memory is down 7.16% over the last 90 days with 12 decliners and no advancers in the tape cluster. The broader AI data center complex has also been weak. Advantest's own price tape was not provided in this context, so the stock's reaction cannot be scored directly. But the setup is clear: a strong fundamental print against a tape that has turned skeptical on AI infrastructure. That is a different story from a company riding a hot theme into a hot tape. It could be a differentiated positive—or it could be a cyclical peak signal. Management is betting on the former, and putting capacity behind it. The optionality comes from CPO development and customer fab projects. Lefever said CPO test times are lengthy and could double or triple into a billion-dollar ATE market, still small versus an $11 billion-plus SoC tester market. On Terafab, he was careful but direct: “it's very real, and I think it will lead to meaningful business in the future” — Doug Lefever, CEO · 2026-10-28. Those are call options, not the base case.

The bottom line

Advantest is no longer just a GPU tester story. It is a compute test-intensity story with three legs—GPU, custom ASIC, and CPU—plus a share-gain story and a capacity race. The quarter's raise was real, but it came from execution and mix, not from a bigger market. The risk is equally clear: memory inflation is taxing margins, and the tape has already begun to doubt the AI infrastructure complex. If the tape is right, this is late-cycle strength. If Advantest is right about CPU and share, it is still early. The company is acting like it believes the latter.