SILICON NEXUS
Research NotesSouth KoreaUnited StatesTaiwanJapan· Aug 14, 2026· AMAT· 6 min read

The Fifth Tier Went Vertical — The Week AI Capex Reached the Drill Bits, Reclaimed Wafers, and Ceramic Capacitors

Applied Materials booked $9.1B and Advantest hit a record — but the more revealing read is Neosem +2,208%, Jeju Semiconductor +2,745%, and a Taiwanese PCB drill-bit maker doubling capacity to 2028. The capex isn't just booked; across four countries, it's landed five layers deep.

Q2 2026 Operating Profit YoY Growth — Top of Funnel vs Long TailKorea Semiconductor Exports — The Money Is on the Ledger

When the drill-bit maker doubles capacity, you look up

Two data points hit the tape on 2026-08-14. Applied Materials (AMAT) posted a record Q3 of $9.1B in revenue, +25% YoY, at a 34% operating margin — its best print ever, driven by orders from Samsung, SK Hynix, TSMC, and Micron for HBM and advanced-node capacity. Advantest (6857-JP) hit consecutive record highs after Nomura raised its target price to ¥42,400. These are the 'top of the funnel' reads that the tape rewards on any day the AI trade is on.

But those numbers weren't the interesting ones this week. The interesting numbers were much further downstream, and they were much bigger.

Neosem (Korea, SSD tester supplier) reported Q2 operating profit up 2,208% YoY as Advantest exits the SSD tester market and Neosem inherits the residual customers. Jeju Semiconductor (Korea, niche memory) posted Q2 operating profit up 2,745% YoY on DRAM and NAND price increases. Semitech (Korea, AI substrates) cleared its 70% utilization break-even and reported Q2 profit up 11x. Solbrain (Korea, etchants) posted +128% YoY as Samsung's Texas fab ramps. TLB (Korea, memory PCBs) delivered +86%. Qualitas Semiconductor (Korea, IP) posted revenue +589% as Samsung Foundry adoption widened. In Taiwan, Sharp Point Technology (8021-TW, PCB drill bits) posted H1 net income +261% YoY. Apex Technology (also PCB drills) announced a 100% capacity expansion by 2028. Alchip (3661-TW) hit a Q2 same-period record on 3nm AI accelerator mass production. Chiao Hsin (1563-TW), a luxury-car forged-wheel supplier, cleared US semiconductor-equipment customer certification and is pivoting its entire business to semi-grade aluminum, targeting NT$1B revenue by 2027. In Japan, DIC (materials) posted record H1 operating profit and raised guidance. Samyang NC Chem posted record H1 profit of KRW 121B on AI chip materials demand.

This is not a top-of-funnel story. This is a fifth-tier story — and the fifth tier only starts printing multi-hundred-percent YoY when the money booked at AMAT and Lam has already worked its way through the fab, through the substrate supplier, through the material distributor, and finally into the invoice ledger of the drill-bit maker or the reclaimed-wafer specialist. In cyclical semis, this is the last shoe to drop — and when it drops this loud, in four countries, in the same three days, it means the wave has reached shore.

Why the pattern is cross-country, not Korea-specific

The temptation is to read this as a Korea data cluster (the thelec Q2 earnings scan is heavy on Korean small-caps). It isn't. Trace the money across four geographies.

An AMAT tool sells for $30-100M and gets installed at a Samsung Pyeongtak P5 fab (CK Solution just won a KRW 308B construction contract for the P5 Phase 1 build). Once installed, that tool consumes Solbrain's etchants (Korea) at every wafer pass, sits on a base built partly from Chiao Hsin's semi-grade aluminum (Taiwan → US-certified), tests wafers using Neosem's SSD testers (Korea), packages dies onto Nan Ya PCB's ABF substrate (Taiwan, hit limit-up on MSCI inclusion and ABF demand), routes signals through Sharp Point / Apex drill-bit-drilled PCBs (Taiwan), and stacks memory using DIC and Samyang NC Chem's specialty chemicals (Japan, Korea). Every one of these companies just printed a record. The fact that they printed together, across four countries, in the same reporting window, is the pattern.

The macro overlay confirms it. Korea's July semiconductor exports were $32.7B, +166% YoY, following $33.6B (+174%) in June and $29.4B (+154%) in May. Japan's July ICT exports hit a record $53.36B, with semi exports up 178.8% YoY. These are not 'AI is going to happen' numbers. These are 'AI has already been paid for, at every layer' numbers. When even SMIC — a Chinese foundry under US sanctions — raises prices on strong AI demand and cites 'spillover effects' boosting peripheral chip prices, the sanctioned tail of the industry is tight too. Capacity is the binding constraint everywhere.

Two useful reads and one uncomfortable one

Read 1 (bullish, near-term): The capex is real, it's paid, and the working-capital cycle is behaving. When the drill-bit supplier is expanding capacity 100% out to 2028, that's a company betting its balance sheet that AI PCB volumes hold. When a Korean etchant supplier posts +128% at a moment when Samsung's Texas fab is still ramping (not yet at peak), the etchant supplier is telling you Texas volume is coming, not gone.

Read 2 (structural, medium-term): The distribution of the gains matters. The top of the funnel (AMAT, Lam, Advantest) is up 25-40% YoY. The middle tier (Alchip, Nan Ya PCB, DIC) is up 25-100% YoY. The long tail is up 100-2,700% YoY. That's an accelerating fan-out — the further from the die you are, the higher your growth rate this quarter. Some of that is base effect, but the substance is that small suppliers are the ones absorbing the marginal AI-driven bill of materials that didn't exist a year ago. iHBM12 needs one more etchant step, one more tester type, one more substrate layer. Those are exactly the line items that show up in Solbrain, Neosem, and TLB.

Read 3 (uncomfortable): The last supplier to see the money is also the first supplier to see the money leave. When AMAT decelerates two quarters from now (if it does), the long tail won't decelerate — it will collapse, because these are 60-80%-of-revenue-from-one-cycle companies. Neosem's 2,208% and Jeju's 2,745% are the kind of numbers that only look sustainable at the peak. The tell to watch is Q3: if the same suppliers print +100% again, the cycle has room. If any of them miss, that's the leading indicator that the giants will miss in Q4.

What to do with the read

The trade this week is not to chase Neosem or Sharp Point at these prints — the multiples are already discounting the number. The trade is to use this pattern to stress-test the next Q3 earnings season: identify the three or four long-tail names most exposed to a single customer (Samsung Pyeongtak, TSMC 3nm, Micron HBM) and treat their Q3 print as the AI-capex hall monitor. If they still print big, the giants have another quarter. If any of them wobbles, sell the giants before the giants report.

The White House chip-belt tariff report and SK Square's holdco pivot are noise this week. The signal is buried in the small-print earnings tables of the second-through-fifth-tier suppliers, in four countries, in the same three days. When the drill-bit maker doubles capacity, you look up.

Key Sources: - Applied Materials Beats Q3 Earnings with Strong Guidance (SiliconANGLE, 2026-08-14) - Neosem Monopolizes SSD Tester Market as Advantest Exits; Q2 Profit Surges 2,208% (thelec, 2026-08-14) - Alchip Q2 Profit Sets Same-Period Record as 3nm AI Accelerator Enters Mass Production (cnyes, 2026-08-14) - Advantest hits record highs as Nomura raises target price (Nikkei, 2026-08-14) - Chinese chipmaker SMIC increases prices on strong AI demand (Reuters, 2026-08-14) - plus 12 more

If this analysis was helpful · Support Us · ✈️ Telegram