AIFR: The Foundry ETF Blends the Market's Two Halves Into One Trade

Foundries are the part of the chip industry that owns the factories, so an ETF built around them sounds like a clean one-click claim on the physical backbone of AI. That is exactly how Defiance is selling AIFR, the first U.S.-listed foundry ETF, launched September 1 at a 0.71% expense ratio. The problem is that the pitch and the portfolio are not the same trade. The foundry business stopped being one market years ago, and the index inside AIFR averages its two halves together instead of letting an investor choose which one they are buying.

The foundry market has already split in two

Reading the industry as a single "foundry theme" is the mistake the product quietly institutionalizes. By node, the business has bifurcated into a leading-edge market and a mature/specialty market, and they do not share economics.

One the one side sits leading-edge fabrication, effectively owned by three names. Pure-foundry market share data shows TSMC at 73% in the second quarter, with Samsung Foundry at 7% and everyone else in single digits. This is the AI capacity trade: TSMC's N2 slots are sold out through the second quarter of 2027, and Samsung's foundry is approaching full utilization while raising prices on new 4nm orders. Pricing power here is structural, because demand is running into a technology-and-capacity constraint only a handful of fabs can clear.

On the other side sits the mature and specialty node — UMC, GlobalFoundries, Tower, Hua Hong, VIS, PSMC — the commodity end that builds the power-management, analog, high-voltage and embedded chips around the AI processors. This half is historically the squeezed one: structurally oversupplied, competing on price. It is only now, as AI demand spills into the supporting silicon, that capacity is tightening enough for prices to start climbing across those nodes. The interesting fact is that even in that upcycle the economics stay far apart, because the two halves capture value very differently.

TSMC's gross margin is 64%, its operating margin 56%, and its return on invested capital roughly 30%, on trailing-twelve-month capex of about $46 billion. GlobalFoundries is at a 26% gross margin and about 5% ROIC, with revenue growth of roughly one percent year over year; UMC's ROIC is around 9% with essentially flat revenue. The two halves are not the same asset class. One has pricing power, the other has been living on the edge of overcapacity.

What AIFR actually holds

The index inside AIFR is a ten-name market-cap blend, rebalanced quarterly with each name capped near 20%. Its own published composition shows the three leading-edge names — TSMC at 21.8%, Samsung at 20.9%, and Intel at 19.8% — together making up about 62% of the weight. That leaves roughly 38% of the fund in the mature and specialty half: UMC at 14.6%, Tower near 6%, GlobalFoundries near 6%, Hua Hong near 5%, plus PSMC, VIS and Visera below 3% each.

That construction is the product in miniature. An investor who buys AIFR because they want the AI manufacturing bottleneck gets the bottleneck capped at under a quarter of the fund and is simultaneously handed a large slug of the commodity half they may not have wanted. If the thesis is the leading-edge constraint, that pricing power resides overwhelmingly in TSMC, and AIFR dilutes it down to 21.8% rather than letting the strongest link of the chain carry the position. If the thesis is a broad bet that the whole chain — winners and losers alike — keeps rising as AI pulls every node along, then the mix makes sense, but that is a wider and weaker claim than "foundry bottleneck."

The blend also forces three very different operating stories into one return. Intel Foundry is a turnaround with a long runway of losses: $4.5 billion in fourth-quarter revenue against a $2.5 billion operating loss, roughly eight years and well over $100 billion of committed expansion behind it, and now a narrative of fixed 18A yields riding a 265% one-year stock move. TSMC has already earned its 81% climb on 64% margins. Samsung carries an enormous memory business alongside its foundry. These are not interchangeable units of "foundry exposure."

Who benefits from the packaging

There is a sharper way to see what is being sold. Defiance filed a 2x daily bearish fund against TSMC on August 26 and listed this long foundry fund five days later. The same sponsor is packaging both tails of the same trade and collecting a fee in both directions. That is the tell of a product engineered around whatever narrative is trading, not around a durable investment characteristic. The 0.71% fee also sits on top of a portfolio that is mostly three large, directly held and heavily covered equities; the middleman cost buys structure, not access to private capacity data.

None of this makes AIFR useless. But it means the fund is best understood as a weighted-average bet across a bifurcated industry, not as a targeted capture of the AI manufacturing shortfall. The name sells the bottleneck; the mechanics sell the whole curve. An investor who actually wants the leading-edge constraint would get closer by holding TSM on its own than by paying a 0.71% fee for a 21.8% position in it, and anyone mistaking the mature-node half for AI pricing power is buying the sector's weakest economics on the strength of its strongest narrative.

The condition worth watching is not which foundry "wins" AI — TSMC has already won that argument. It is whether the mature and specialty node upcycle turns out to be real pricing power or just a reprieve in an oversupplied corner of the market. If pricing holds and utilization stays tight there, AIFR's less glamorous 38% earns its keep. If it is a cyclical lull that reverts to overcapacity, then the fund is mostly a diluted TSMC with a commodity drag attached. The two-market split is not a reason to dismiss AIFR. It is the reason to decide which half of the market you are actually paying to own.