FTXL Is a Memory-Shortage Bet, Not the Semiconductor Allocation

FTXL Is a Memory-Shortage Bet, Not the Semiconductor Allocation

I've been watching a false narrative take shape this year, and it has started to leak into real portfolios. The version I keep hearing goes roughly like this: FTXL — the First Trust Nasdaq Semiconductor ETF — is "the semiconductor trade," so the only question worth asking is how big a position to carve out. The fund has earned the attention. It has more than doubled over the past twelve months and sat up roughly three-quarters year to date through mid-August, at or near the top of the major chip funds. That performance is real. What is not real is the implication that a fund's trailing momentum tells you how to size an allocation. FTXL is not diversified semiconductor exposure. It is a concentrated, factor-tilted bet on a narrow slice of the chip market — the memory cycle and the US turnaround names — and it belongs in a satellite slot, a position sized for a specific bet, not in the core.

What FTXL actually is

Start with construction, because everything else follows from it. FTXL tracks the Nasdaq US Smart Semiconductor Index, which screens the US chip universe on volatility, value, and growth before applying a modified factor weighting. The practical effect: the biggest names get capped, and the money tilts toward whatever scores well on value and growth at that moment. That produces a 35-stock fund in which the top ten holdings carry roughly two-thirds of assets — against a category average of about half — and where Micron runs about 12% and Intel around 11%, effectively a quarter of the whole fund in two names.

Sit with that before you file the fund under "semis." Micron is the memory seller's market running at full throttle; Intel is a foundry turnaround still generating negative free cash flow. The supporting cast — Broadcom near 7%, AMD near 7%, Marvell — is AI networking and custom silicon. This is a bet on the memory and AI-infrastructure supply chain, in a wrapper that holds Nvidia's weight well below the share it carries in the market-cap funds. It is also entirely US-listed, which separates it from the default vehicles most people compare it to: TSMC is about 9.5% of SMH and roughly 2.5% of SOXX, and FTXL holds none of it. In a geopolitical crisis you buy zero Taiwan chokepoint risk; when the foundry oligopoly is printing money, you collect none of that either.

Memory is the chokepoint

The reason this fund doubled is that the bottleneck of this AI cycle has moved from the GPU to the memory that feeds it. DRAM — the commodity memory inside nearly every computer — and HBM, the high-bandwidth variant stacked directly onto AI accelerators, ran effectively sold out through the year, with memory prices expected to rise more than 50% in a single quarter at the start of 2026. AI data centers are diverting a growing share of memory supply away from PCs and phones, precisely because they are the ones paying the premium, and supply is expected to stay tight well into 2028.

The numbers on FTXL's two heaviest bets say the demand is structural, not cyclical. As of mid-August market data, Micron is running year-over-year revenue growth north of 160%, gross margins near 73%, roughly $26 billion of trailing free cash flow, and a net-cash balance sheet with debt at about 6% of equity. Intel, by contrast, is the price you pay for the tilt: revenue up in the high single digits, gross margin in the mid-30s, operating margin still negative. Around those two, Broadcom is compounding revenue at a 30%-plus clip on a 43% free cash flow margin with nine consecutive years of dividend growth, and Nvidia's revenue is up more than 70% on a 74% gross margin. When every hyperscaler is expanding capital spending at the same time and much of that spending lands on the same memory, networking, and custom-silicon names, the AI demand is reinforcing itself — the cross-pollination effect — rather than topping out.

The July sell-off was the wrong headline

That is why the market's July mood looks, in my opinion, like a false narrative in motion. A single early-July session knocked double digits off Micron on valuation anxiety, and the unwinding kept running: the Philadelphia Semiconductor Index fell more than 3% in a single late-July session as investors kept exiting the hottest trade of the year, with Micron down close to 40% from its high while memory prices were being hiked by double digits and the shortage shows no early end. Fundamentals strengthened while the price fell — the same footage every cycle, just louder.

The FTXL-specific crowd behaved exactly as you would predict. The fund absorbed roughly $1.5 billion of net outflows in a single month, against a current asset base of about $1.3 billion — a dominant fraction of the shareholder base turning over in weeks. Call it momentum in, momentum out. That is fine for a satellite and a poor premise for a core. And note what the outflows are not telling you: investors stamped out of an up-76%-year fund over a single-digit pullback even as the memory earnings underneath accelerated. That is an irrational overreaction, the reason the sector thesis survives, and the reason new money should enter through deliberate sizing rather than through whatever is left of the run.

The sizing answer

Before deciding how much of it you want, price the vehicle itself. FTXL's 0.60% expense ratio is nearly double the 0.35% charged by the big market-cap chip funds, and its daily volume is a small fraction of SOXX's, so in a fund this thin your own buying and rebalancing is a meaningful part of the tape. Add a beta near 1.7 — the fund tends to amplify the market's moves by roughly two-thirds in either direction — and annualized volatility in the 55%-to-65% range over the past year, and you have a wrapper engineered to swing. None of that is fatal to an aggressive satellite allocation. It compounds against you for twenty years if you mistake it for a core holding.

The honest answer to the allocation question splits in two, and the split is the whole thing — the sector decision and the vehicle decision are separate. The semiconductor sector earns an overweight in a diversified equity portfolio for investors who can tolerate the volatility: the demand is real, accelerating, and funded by capital spending that keeps reinforcing itself. For that core overweight, I favor the low-cost, broad, liquid chip funds, and I would let the market's existing technology weight do the compounding work underneath.

FTXL is the satellite you add when you specifically want the memory and US-turnaround axis — Micron's pricing power, Intel's foundry optionality, the networking and custom-silicon names — in one concentrated sleeve. Sized as a satellite, in my judgment that means a single-digit position: no more than roughly 5% for most investors, and meaningfully less for anyone who does not explicitly want the memory-and-turnaround tilt, because a fund that is a quarter Micron-Intel is not diversified chip exposure by any construction. For investors who can tolerate short-term volatility, the semiconductor overweight still makes sense as part of a diversified core built around low-cost broad-market exposure; FTXL sharpens that sleeve where you deliberately want it sharpened, rather than defining it.

That being the case, the answer to the sizing question is: not as much as the momentum says. The memory shortage is, in my opinion, the strongest structural data in the entire AI trade, and FTXL expresses it directly. But it expresses it through 35 names, two of which carry a quarter of its assets, at nearly double the fee of the benchmark funds, in a wrapper that churns the minute the narrative wobbles. Own it as a satellite sized for a specific bet, not as the sector allocation — and expect the same headlines that sold it in July to reappear the next time memory pricing has a bad minute.