Diversification used to be the entire pitch. Somebody has decided it was cluttering up the story.
Four names surfaced at once out of a suite of seven, so there are three more sitting in the queue like the second half of a boxed set. Nuclear, semiconductors and defence tech are the ones still waiting their turn. Each fund is built around roughly ten holdings. Not ten as an unfortunate consequence of a narrow investable universe, but ten on purpose, as the product.
The word being used for this is “high conviction," which is a lovely piece of language. It takes the thing a risk committee spends its life trying to prevent and turns it into a feature you can put on a fact sheet. Concentration risk becomes focus. Ten names become a portfolio.
To be fair to Amplify, the logic holds together better than the instinct to mock it suggests. Broad thematic funds have a genuine problem: dilute the theme with enough adjacent names and you end up owning the same mega-caps you already own, with an extra charge for the privilege. If you actually believe in quantum computing, the honest expression of that belief is a small number of companies. This suite simply refuses to pretend otherwise. It is intellectually consistent. It will also move like a shopping trolley on ice.
Robotics turned up twice in the same batch, from opposite directions, which is the sort of thing that makes reading listings data worth the trouble.
Defiance is billing this as the first US-listed ETF dedicated to China's humanoid robotics ecosystem, and the construction is more considered than the pitch implies. It tracks a Solactive index of twenty Chinese companies, selected by a natural-language-processing screen that reads news and company disclosure to score thematic exposure, spanning four segments: humanoid robots themselves, motion control and precision actuators, robotic perception and the automation infrastructure that puts machines to work. Weights blend free float with that relevance score, capped at seven percent per name, rebalanced quarterly. Most of the constituents trade in Hong Kong or on the mainland, which is the actual point of the wrapper.
So investors now have a choice between roughly ten globally selected robotics names and twenty specifically Chinese ones, on the reasonable assumption that the machines will be assembled somewhere other than where they are designed. There is something quietly telling about a US issuer deciding that the humanoid robot trade is a China trade.
The AI trade has been subdivided again, and this time the cut is a real one.
The fund tracks a BITA index of companies building the chips that run finished AI models rather than the chips that train them, across six segments: inference GPUs, custom ASICs, FPGA accelerators, AI-optimised CPUs, accelerator modules, and neuromorphic processors. To qualify, a company must derive at least half its revenue from one of those segments or demonstrate material involvement, which is a more honest screen than most thematic indices bother with.
The underlying logic is sound. Training a model happens once. Inference happens every time anyone uses it, and the economics of that second job are about latency and cost per query rather than raw throughput. This is also the moment “inference" completes its journey from engineering vocabulary to ticker symbol, and once a word makes that trip it never comes back.
Both Defiance funds share a detail I enjoyed more than I should have. At launch, each held precisely one position: a total return swap with a French bank, sitting at very nearly one hundred percent of net assets. The full stack of the inference chip universe, the entire Chinese humanoid supply chain, and on day one it is one line item and a rounding error of cash. Every fund starts somewhere.
A box spread is four options positions arranged so that they cancel each other out and leave behind a fixed payout, which is a laborious way of manufacturing something that behaves like a Treasury bill. The appeal is the tax treatment rather than the yield, which is why these products exist at all.
Two things stand out. The first is the name, which is near enough word-for-word the name of the incumbent that already dominates this niche and runs well over ten billion dollars. The second is the fee. This is now the fourth product in a corner of the market where every previous entrant arrived by undercutting the last one, and this one comes in above the incumbent's net expense ratio and comfortably above the cheapest rival. It has joined a fee war without bringing a fee.
I want to be clear that I put real effort into this one.
The name tells you nothing. There is no theme in it, no asset class, no hint of direction. “CECflex Institutional” could plausibly be a European insurance wrapper, a structured credit vehicle, or a brand of office chair. It is, in fact, a systematic crypto strategy: a rotation across bitcoin and large-cap altcoins driven by quantitative research and risk signals, held through crypto products in cold storage, issued off a Swiss structured-products platform and now available in an institutional line alongside the existing retail one.
This is the most aggressive product in the batch by a comfortable margin, and it has the most sedate name. Everywhere else, issuers stack buzzwords until the fund name reads like a conference agenda. Here, someone had discretionary altcoin exposure to sell and chose to describe it as “flex”. There is a strategy in that too.
Fuller and Thaler build funds on the research of behavioural economists, one of whom won a Nobel Prize for demonstrating that investors make predictable, systematic mistakes. The firm has spent decades running that idea in mutual funds and separate accounts, hunting for companies the market has misjudged because it is still anchored to an old opinion. The entire premise is that other people's cognitive errors are a tradeable resource. This one is a US mid-cap strategy measured against a mid-cap growth benchmark, which is about as unglamorous a mandate as exists.