Memory Upcycle, AI Capex: Is AMAT Finally Set to Break Out?

In the first deep-dive on AMAT, Dolphin Research mapped the company’s core products and positioning across equipment sub-segments. In this piece, we tie AMAT’s outlook and valuation to wafer fab capex cycles.

As a front-end tool vendor at the top of the semi value chain, AMAT sells primarily to foundries and memory fabs, spanning logic (TSMC, Intel, etc.) and memory (Samsung, SK Hynix, etc.). Hence, forecasts hinge on wafer fab capex and AMAT’s equipment share across segments.

Within semi systems, logic is AMAT’s largest revenue pool, with logic:memory roughly 2:1 today. Memory capex has risen recently and will lift related sales, but the biggest swing factor remains logic.

With China consumer electronics subsidies fading and memory price hikes, OEMs are cautious on next year. Looking at fab capex guides, logic upside still skews to TSMC, while Samsung Foundry and NVIDIA’s outsourced capex remains conservative. TSMC’s 3nm remains tight for AI chips, so it will keep expanding capacity.

This implies a split outlook for AMAT next year: memory should deliver more certain growth, while logic looks softer. Logic is driven mainly by TSMC, with spend concentrated in 2nm ramp and High-NA EUV buys, making broad-based operating upside harder to achieve.

Based on company commentary and market checks, Dolphin Research expects 2026 logic capex to reach $88.4bn and memory capex to hit $75.8bn. Samsung, SK Hynix, and Micron should all step up spending, though the bulk is HBM-related where AMAT’s tool competitiveness is weaker.

After a cautious 2026 for most logic players, if end demand normalizes, 2027 capex should also rise. We therefore anchor our estimates on 2027.

Dolphin Research estimates AMAT’s core net OP at $9.5bn in FY2027 (fiscal year ending Oct-2027), assuming revenue CAGR of ~10% over two years, GPM at 48.4%, and a 13.1% tax rate.

As upstream tool vendors, sustained rapid growth is hard without a broad-based upcycle in downstream applications. This cyclicality also caps multiples for equipment names.

Helped by the memory upcycle, AMAT shares have run to $270. On PE and DCF, a reasonable range is ~$240–300, implying ~20x–25x PE on FY2027 core net OP; upside from chasing here looks limited.

Even under an upside case where TSMC further raises 2027 capex and both Intel and Samsung rebound, lifting combined capex by $26.5bn, and AMAT’s logic share recovers to ~18%, the current market cap still implies ~20x PE on FY2027. That sits around the mid of its historical band, while at earnings peaks, the multiple typically skews below mid due to cyclicality and rarely tops 20x.

Overall, the recent rerating already prices in higher memory spend and TSMC’s expansion. Risk-reward at current levels is not attractive.

For Dolphin Research’s detailed read-through of AMAT’s results, see below. More specifics follow.

I. Wafer fab capex trends

Semi Systems (front-end equipment) accounts for roughly three-quarters of AMAT’s revenue, and is the primary driver. It dominates the earnings profile.

AMAT’s semi revenue growth broadly tracks semi capex growth. Fab capex trends therefore feed directly into AMAT’s top line.

To frame AMAT’s revenue outlook, we first assess fab capex. Our core sample aggregates nine global leaders: TSMC, Samsung, Intel, SK Hynix, Micron, SMIC, and others in the prior chart.

AMAT splits semi revenue into logic and memory (DRAM and NAND). Given different spend cycles, we also group the nine fabs into logic and memory cohorts for analysis.

1.1 Logic

Logic capex covers TSMC, Intel, SMIC, UMC, GlobalFoundries, UMC (duplicate in source), and Samsung Foundry. The key names are included.

YTD, only TSMC has been lifting capex, while others remain tight. The industry is bifurcated: demand is concentrated in AI compute, with NVIDIA, AMD, and Broadcom orders largely at TSMC, while traditional segments have not truly recovered.

TSMC spent $29.8bn last year, and is guiding $40–42bn this year on AI and advanced-node ramps, roughly +35% YoY. That is a sizable step-up.

However, a notable chunk at TSMC appears tied to CoWoS packaging, so pure logic WFE is not up as much as 30%+. In aggregate, other core logic players lift capex by only ~2%, keeping overall logic spend constrained.

Intel’s weak fundamentals and Samsung Foundry’s lost orders drove cuts this year. SMIC, UMC, Hua Hong, and others show no clear expansion, so TSMC alone cannot pull up logic market-wide.

Looking to next year, logic capex should recover modestly. TSMC is likely to maintain double-digit growth; we model ~$51bn. Intel, backed by U.S. Gov. support and new partners, looks to be off the floor, with full-year capex likely to rise from the trough.

As legacy semi demand normalizes, we see logic capex reaching ~$120bn by 2030, a ~10% CAGR. That frames the longer view.

1.2 Memory

Memory capex covers SK Hynix, Micron, and Samsung’s memory division. These are the key beneficiaries of AI-driven memory demand.

Driven by AI/HBM, SK Hynix and Micron have already ramped spending this year. Samsung stayed conservative as its HBM struggled to pass NVIDIA’s qualification. Combined memory capex reaches $58.3bn this year (+45% YoY), led by Micron and SK Hynix.

Next year should see continued strong growth. Micron and SK Hynix plan further increases, while Samsung likely provides the biggest delta as its HBM3E reportedly passed NVIDIA’s qual and HBM4 samples have shipped.

Coupled with AI capex pulling up legacy memory (DDR, NAND), the three are set to raise memory capex again next year to about $75bn (+30% YoY). That is a robust trajectory.

But memory is highly cyclical. After capacity adds, supply-demand resets dynamically. If pricing rolls over, capex will tighten again until the next upcycle.

1.3 AMAT’s share in semi equipment

AMAT’s semi equipment revenue splits into logic and memory, which can be mapped to fab capex. AMAT captures close to ~20% against logic capex, while its share against memory capex has slipped to ~12%.

AMAT’s share in memory capex has trended down for three reasons:

a) Value reallocation from tech shifts: In 3D NAND, etch value-share rises from ~25% to ~50%. In HBM, TSV tools are key at roughly 30% of value.

b) Incremental markets are not AMAT’s strengths: Lam Research leads memory etch. Its competitive position is entrenched.

Lam’s Vantex is now standard for 3D NAND, with clear leadership in high aspect ratio etch, while AMAT is still catching up. HBM is also Lam-led, with Syndion for deep silicon etch addressing TSV needs.

b) Strategic focus shift: AMAT has prioritized AI-driven logic, where TSMC dominates, and AI capacity can reuse smartphone-era process capacity rather than memory lines. This indirectly hurt AMAT’s memory competitiveness.

In advanced memory, Lam has become a core vendor to SK Hynix and Samsung, exceeding 60% share in their etch tools. That positioning is hard to dislodge.

Thus, despite memory leaders stepping up spend for 3D NAND/HBM transitions, AMAT’s memory portfolio did not fully benefit, and the stock has underperformed Lam. The divergence has been notable.

While memory capex grew nearly 45% in 2025, AMAT’s memory revenue rose only ~10%. As HBM-led spend mixed up, AMAT’s wallet share fell, keeping the stock range-bound until Sep.

In 2H, the setup improved. Data center storage demand broadened beyond HBM into DDR and NAND, sending pricing higher across the board. DDR’s rally since mid-year illustrates the broad memory recovery.

With legacy memory rebounding, the market lifted capex expectations again, and this time not just for HBM/3D NAND. As DDR prices climbed, AMAT’s stock finally broke out in 2H as well.

From Samsung, SK Hynix, and Micron’s latest guides, all three are upbeat, with capex skewed to advanced nodes. Thus AMAT should grow in memory next year, but its share may still slip. Share gains likely require tech breakthroughs in HBM/3D NAND or a sustained upturn in legacy memory.

Logic matters more for AMAT, with revenue about 2x memory. On capex, next year’s logic growth will be led by TSMC’s 2nm ramp.

Note that capex is related to, but not equal to, equipment purchases. Part of TSMC’s capex targets node enhancements and process optimization, not just new tools. Also, High-NA EUV ASPs have doubled (~€200mn to ~€400mn), lifting EUV’s share in WFE as 2nm/High-NA scales, which dilutes AMAT’s value share within logic tools.

So even as TSMC raises capex, AMAT’s logic share next fiscal year likely dips. Only a broader logic recovery will drive larger absolute upside; AMAT is ultimately levered to the semi cycle.

II. AMAT’s outlook and valuation

2.1 Outlook

Based on fab capex trajectories and share, AMAT’s core Semi Systems revenue should keep rising. We model Semi Systems approaching ~$30bn by 2030.

Beyond Semi Systems, AMAT has AGS (Applied Global Services) and Display & Adjacent. AGS, which includes services such as maintenance and remote diagnostics, is steady and has historically grown.

Display & Adjacent, a cyclical business, is recovering from its trough. The mix should turn more balanced as that cycle improves.

Overall, we model total revenue rising from $28.3bn to ~$40bn by 2030, a ~7% CAGR. The slope is driven primarily by Semi Systems.

For core earnings, margin and opex matter in addition to revenue. We outline key assumptions below.

① GPM: Supported by memory-related demand, we expect margins to stay elevated in FY2026–FY2027, then drift lower as memory tailwinds fade.

② Opex ratio: AMAT plans to cut headcount by ~4%. With the action implemented, we expect opex to decline in FY2026, with more discipline thereafter even if spend rises.

Putting it together, we model core net OP reaching ~$12.7bn by FY2030, a ~9% CAGR. The margin profile remains relatively stable, so the revenue path is the key driver.

Despite solid memory demand, the main lever remains logic, especially legacy logic. That is where breadth matters most.

With China subsidies fading and memory price hikes weighing on phones/PCs, end-demand expectations are soft for next year. That dampens legacy semi capex and AMAT’s next-fiscal growth, until legacy demand truly recovers.

2.2 Valuation

① PE lens: AMAT’s historical range is ~15x–25x. That is the standard band through cycles.

Our FY2027 core net OP estimate is ~$9.5bn (revenue CAGR ~10% over two years, GPM 48.4%, tax 13.1%). Assuming 20x–25x, implied EV/market cap is ~$190–237.5bn, or $240–297/sh.

Before the current memory upcycle and TSMC capex step-up, the stock oscillated in the $150–200 range. Stronger memory and higher TSMC capex helped AMAT hold above $200.

Given soft end-demand for phones/PCs under fading subsidies and higher memory prices, the legacy semi outlook remains muted. Sustained improvement across legacy would be needed for another leg up in earnings and the stock.

② DCF lens: With WACC at 8.69% and a 3% terminal growth rate, DCF yields ~$248/sh (~$198.1bn market cap), below today’s level. This equates to roughly 21x FY2027 core net OP.

Under an upside scenario, if TSMC lifts 2027 capex further and Intel/Samsung rebound strongly, adding $26.5bn combined, and AMAT’s logic share returns to ~18%, the current market cap still implies ~20x FY2027. That is around the mid of its 15x–25x band, and at cycle peaks the multiple tends to sit below mid, not above.

Net-net, the recent rerating already embeds higher memory spend and TSMC’s expansion. The entry point is not compelling.

Dolphin Research on AMAT

Nov 14, 2025 Trans: AMAT: Flat 1H Next Year, Growth Skews to 2H

Nov 14, 2025 First Take: Memory Is Red-Hot — Can AMAT Ride the Wave?

Sep 18, 2025 Deep Dive: AMAT: AI Spending Is Rising — When Will AI Capex Go Full-Stack?

Risk disclosure and statement: Dolphin Research Disclaimer and General Disclosure