Earnings Highlights: Advantest (6857.T) | FY2026 Q1 — GPU, ASIC or CPU, Every AI Chip Has to Pass Through Its Toll Booth First

2 days ago

7 min read

The key to Advantest's quarter isn't how pretty the numbers are (though they really are pretty) — it's that management said one thing plainly: who wins or loses the AI chip arms race barely matters to the company, because GPUs, custom ASICs and CPUs all end up on the same V93000 tester. Revenue for the quarter was ¥367.5B, operating margin surged to a rare 51.7%, full-year guidance was raised in one go with net income projected 42% higher than the prior forecast, and the midpoint of the SoC tester market size estimate was raised by about 20% mid-year. The only noise comes from memory: rising DRAM prices are a cost headwind, and memory testers will lose a bit of share this year. Below we break down the five signals that really matter this quarter.

This article is for technology and industry trend analysis only and does not constitute investment advice.

1. Three Core Conclusions (Executive Takeaway)

First, this was an "all cylinders firing" quarter: revenue grew 39.3% YoY, but profit grew far faster than revenue — net income up 93.8% YoY, EPS up 96.1% YoY, and operating margin reached 51.7%, a quarterly record. Management itself cautioned that this level of margin "won't be the norm".

Second, SoC tester demand has been structurally revised up, and the driver has shifted from "a single application" to "every application". This TAM raise came mainly from inference and compute demand, with custom ASICs, CPUs and networking chips ramping together. The key is that Advantest's V93000 is the de facto standard across all high-performance computing applications, so "which chip wins" is a false question for it.

Third, memory is the only headwind, but it splits into two parts: one is cost (the testers use a lot of DRAM, and rising memory prices erode gross margin), the other is share (memory testers will dip slightly this year and recover next year with new products). SoC is big enough to drown out the memory noise.

2. Revenue and Financials

Core numbers for the quarter (ended June 30, 2026, IFRS consolidated):

• Net Sales: ¥367.5B, up 39.3% YoY

• Gross Margin: 69.5% (gross profit ¥255.6B)

• Operating Income: ¥190.0B, up 53.3% YoY, operating margin 51.7% (quarterly record)

• Income Before Tax: ¥234.1B, up 92.9% YoY (including non-operating gains such as FX)

• Net Income: ¥174.8B, up 93.8% YoY

• EPS (basic, IFRS): ¥241.27, up 96.1% YoY

By segment, Test System was ¥333.6B, up 38.7% YoY, by far the bulk; Services and Others were ¥33.9B, up 46.0% YoY. Overseas revenue accounted for as much as 99.1% — this is a company almost entirely bet on the global AI supply chain.

The point here isn't that "the numbers are big" but that profit growth was more than double revenue growth. Management broke down the reasons clearly: favorable product mix, FX tailwinds, economies of scale kicking in as shipments grow, plus a one-off boost from an inventory write-down reversal. In other words, part of the 51.7% operating margin is "just this quarter" and shouldn't be extrapolated linearly.

3. Technology and Business Highlights: One Platform Wins All Three Wars

What's most worth noting from this call is how management answered the "AI chip civil war" question. Analysts kept pressing: is the TAM raise coming from GPUs, ASICs or CPUs? CEO Doug Lefever and Chief Customer Officer Sanjeev Mohan answered almost identically — "We don't really care, because they all use the same V93000."

There are three layers to the reasoning. First, the surge in inference applications is lifting unit volumes of CPUs and compute chips; second, nearly every hyperscaler is building its own silicon, so the number of custom ASIC programs is far larger than before, and these chips are "incremental beyond GPUs"; third, a segment that was never singled out before — networking chips — where Advantest actually has high share too but hadn't specifically mentioned, is also ramping.

Sanjeev even directly referenced industry leaders' views: the custom ASIC and compute market will become significantly larger than the GPU market; GPUs are still the largest today, but the gap will close. We broke down this thread of "custom silicon defining the shape of the market itself" in Earnings Highlights: Marvell (MRVL) | FY27 Q1 — From "Following CapEx" to "Defining the Shape of Scale-up Itself" — and Advantest sits right at the toll booth of this customization wave.

Test content itself is rising too. Accelerators are moving to 3D packaging and complexity increases generation by generation, so final test time and insertion counts are higher than for CPUs; Advantest therefore can't just expand capacity by "unit volume" but by "unit volume × per-generation test content growth". This is the technical root of its longer demand visibility.

4. Management Outlook: Guidance Raised, Capacity Pulled In, Visibility Extended to 18 Months

Full-year guidance (raised this quarter):

• Revenue ¥1,714B (previously ¥1,420B)

• Operating income ¥846B (previously ¥627.5B)

• Income before tax ¥891B

• Net income ¥660B (previously ¥465.5B, raised about 42% from the prior forecast)

• EPS ¥911.64

This is a very large mid-year raise; Nikkei and broker flash reports framed it as "record profit, raised again".

Three points in the wording are worth parsing for tone:

Capacity — the word is "certain", not "expected". The original target of 10,000 V93000 units/year by end-2028/early-2029 is being "pulled in", though no new date has been given yet; the 5,000-unit milestone will be "blown right through". Management even hinted that there's another tier beyond 10K (15K or higher). Compare that to annual output of about 1,000 units in 2020 (in smaller configurations): normalized for size, 10K is roughly equivalent to 20,000 units back then — 20x in seven or eight years. Its contract manufacturer has already added a second factory in Southeast Asia for support.

Visibility — upgraded from "expected" to "documented consensus". Customers used to give only about six months of forecasts; now, because advanced packaging and wafer capacity are the bottleneck, Advantest gets about 18 months of visibility. That visibility is the precondition for daring to expand capacity early. This signal of "AI CapEx directly stretching upstream visibility" is the same main theme as Earnings Highlights: Broadcom (AVGO) | Q2 FY26 — $30 Billion of AI Orders in One Quarter Pushes Optical Visibility Out to 2028.

Share — the wording mixes "target" and "confident". SoC tester share was 66% in 2025; management said "the upward trajectory will continue" and sees a path to 70%+, with HPC far higher still. On memory, it admitted share may dip slightly this year (NAND isn't a traditional strength, and device speeds and bandwidth are transitioning), but stressed that new products have already won customer commitments and that over the next few years, as new fabs come online, it will "more than recover".

Incidentally, when asked about AI trends, Doug framed the current moment as a structural shift driven by agentic AI — the real efficiency explosion is happening in corporate R&D and back offices, while physical AI will take a while to go mainstream. This view that "the agent era rewrites the demand structure" echoes the framework in Jensen Huang Took Computing Apart: In the Agent Era, Optical Connectivity Is the Critical Path to Revenue.

5. Supply Chain and Customer Clues

Shipment geography: Taiwan is the largest shipment destination, not because of local Taiwanese demand but because most US fabless SoCs are tested in Taiwan; in other words, the true source of demand is US fabless design houses. This "designed in the US, made in Taiwan" chain is the easiest thing to misread in Advantest's regional numbers.

China: about 19–20% of total revenue, of which roughly 80–85% is SoC and 15% memory, close to the overall mix. China is building a batch of new fabs expected to come online around 2028, and only then will China's share have a chance to rise; until then it stays at 20% or below.

Memory's two-sided signal: on one hand, rising DRAM prices are a cost headwind (the testers consume a lot of DRAM), and management said it will try to pass on part of it "case by case, application by application, customer by customer" — very conservative, very honest wording, meaning pass-through will be neither across the board nor immediate. On the other hand, the memory market itself will grow significantly as new fabs open, and Advantest is betting on its new products' roadmap alignment with HBM and next-generation memory, positioning this year's share dip as "short-term pain during a product transition".

Zooming out, Advantest's "whoever wins AI, I collect the toll" position is essentially the same story as the whole optical communications/AI infrastructure supply chain turning from growth stocks into index heavyweights — we went deeper on this in NVIDIA Buys Into Optics: When Optical Communications Turns from Growth Stock to Index Heavyweight.

6. Conclusion

Advantest's message to the market this quarter is clean: the messier the AI arms race, the more money it makes. When GPUs, custom ASICs, CPUs and networking chips all crowd onto the same V93000 platform, it doesn't have to bet on which horse wins — it is the track every horse has to run on. A TAM raise, capacity pulled in, visibility stretched to 18 months, and share advancing toward 70% — four things happening at once is itself evidence of a qualitative change in the demand structure.

What really needs watching is the headwind side: the 51.7% operating margin includes a one-off reversal and FX tailwinds, so a pullback is all but certain; this year's memory share dip and the progress of passing on DRAM costs will determine "how high gross margin can hold".

Three metrics to track next quarter (FY2026 Q2): (1) whether a "new target date" for 10K units/year of capacity is formally given — the most direct thermometer of demand strength; (2) how far operating margin falls from this quarter's peak, to back out the "normalized profit level"; (3) memory tester orders and new-product shipment progress, to confirm whether "short-term pain this year, recovery next year" materializes.