HBMX ETF Review 2026: Tuttle Capital Concentrated Memory Stack ETF

HBMX at a glance

What the memory stack is — and why it matters for AI

Every AI training run requires massive amounts of high-bandwidth memory (HBM) chips stacked directly next to the GPU. An NVIDIA H100 GPU uses 80GB of HBM3, a B200 uses 192GB of HBM3e — and hyperscalers are ordering tens of thousands of these accelerators at a time.

The memory semiconductor ecosystem HBMX targets includes:

• DRAM manufacturers (Micron, SK Hynix, Samsung) — the raw memory chips
• NAND flash suppliers — storage that feeds inference pipelines
• Advanced packaging companies (Amkor, ASE) — stacking HBM dies requires cutting-edge 2.5D/3D packaging
• Semiconductor test equipment (Advantest, Onto Innovation, Camtek) — every HBM chip is tested before shipping
• Wafer fab equipment (Applied Materials, Lam Research, ASML) — manufacturing the chips requires the most advanced lithography and etch tools

The concentration is the point. Most broad semiconductor ETFs (SOXX, SMH) dilute the memory theme across logic chips, CPUs, and networking. HBMX concentrates specifically on the stack that powers AI memory.

Top holdings (as of June 16, 2026)

The portfolio spans the entire stack from equipment (ASML, AMAT, LRCX) → manufacturing (MU) → packaging (AMKR) → test (ONTO, ADNT, CAMT). No single tier dominates — the fund spreads risk across the value chain.

The AI memory demand story

HBM is the defining bottleneck of the current AI compute cycle. GPU manufacturers cannot just add more CUDA cores — training large language models is fundamentally memory-bandwidth-constrained. The industry response is HBM: DRAM chips stacked 8–12 layers high using through-silicon vias, delivering 10x the bandwidth of standard DRAM per watt.

• SK Hynix has a 2+ year HBM lead — but Micron is ramping HBM3e aggressively and gaining share
• Samsung's HBM qualification delays for NVIDIA created a windfall for SK Hynix and Micron in 2024–2025
• HBM pricing has held firm at significant premium to commodity DRAM — $25–30/GB vs $3–4/GB for DDR5 — sustaining memory maker margins
• NAND: AI inference requires fast storage for KV-cache retrieval; enterprise SSD demand tracking AI workload growth
• Test intensity: HBM chips require significantly more test coverage per die than commodity DRAM, driving Advantest and Onto Innovation's outperformance

Bull case vs bear case for HBMX

Bull case:

• AI capital expenditure by hyperscalers (Microsoft, Google, Meta, Amazon) remains on an upward trajectory through 2027–2028 — each new data center needs more HBM
• HBM3e → HBM4 transition drives an upgrade cycle in packaging and test equipment in 2026–2027
• Micron catching SK Hynix in HBM market share creates a duopoly with pricing discipline — high margins sustained
• Concentrated portfolio means outperformance can be significant when the memory cycle is up

Bear case:

• Semiconductor cycles are severe — 2022–2023 saw HBM precursors (standard DRAM) fall 50%+; a repeat would hurt the whole portfolio
• 0.95% expense ratio is expensive; over a decade it compounds to meaningful performance drag vs SOXX or SMH at 0.35–0.44%
• Concentration risk: 8 positions = 60% of the fund. One bad earnings report from Micron or Onto can hit the fund hard
• New ETF (June 2026) — no track record; the fund must prove it can survive a downcycle
• China export controls on semiconductor equipment (ASML, AMAT, LRCX) remain a policy risk

Who should buy HBMX?

• Investors who want focused memory semiconductor exposure beyond owning MU alone
• Those who believe AI infrastructure buildout continues for 3–5 years and want the picks-and-shovels play
• Tactical traders looking to express a HBM or AI capex thesis in a single ticker
• NOT right for passive, low-cost, long-horizon investors — the 0.95% fee and concentration are a mismatch
• Best used as a satellite position (5–15% of a portfolio) rather than a core holding

HBMX vs alternatives

Analyze memory semiconductor stocks