AI Stock-Pickers Just Crossed the Live-Capital Threshold — the Real Test Starts at the Fee Stack

On February 5, 2026, FINQ — a Tel Aviv-based fintech — listed the FINQ FIRST U.S. Large Cap AI-Managed Equity ETF and the FINQ Dollar Neutral U.S. Large Cap AI-Managed Equity ETF on NYSE Arca. Reuters reported the launch on February 10, calling them the first U.S. ETFs in which a proprietary model selects, weights, and rebalances the portfolio with humans limited to \"oversight and governance.\" The funds run the AI's stock-ranking engine across all 500 S&P 500 names daily, using market, financial, and textual inputs to drive positioning.

The structure matters more than the returns. AIEQ, the 2017 IBM-Watson equity ETF, was an experiment — FINQ is the first SEC-registered shop to let the model alone hold the trading authority. SEC Division of Investment Management Director Brian Daly told the ICI Winter Board on Feb 3, 2026, that advisers using AI for portfolio and proxy decisions is \"a near-term reality\" and signaled an open regulatory posture.

The live test is not whether AI can beat the S&P — AIUP has answered that. The live test is whether AI-managed products can scale AUM faster than humans can cut fees. The early data is mixed and meaningful.

Pictet's PQUS is the proof of concept that the fee structure works: launched Feb 25, 2026, it cleared $92.4M by mid-May — approaching the $100M mark in three months on a 0.22% expense ratio that sits on par with plain-vanilla passive. HIAI (Ai Funds) charged 0.87% —109% above its active-ETF category average of 0.42% — and ended with just $19M in its first two months. FINQ's two funds sit between, with the long/short AINT charging 1.25% — roughly five times the category average.

T. Rowe Price, the closest public-market proxy for what AI funds are eating, reported an annualized effective fee rate of 38.1 bps in Q2 2026 and net outflows of $6.5B for the quarter, $8.2B in July, and $12.2B in September. Average firmwide AUM of $1.78T in Q2 2026 was up only 0.4% sequentially — the firm is holding its dollar volume by market appreciation, not by inflows.

The structural worry is not whether AI funds perform — they do. It is whether the dozens of newly launched AI managers are independently generating signals or learning from the same data and converging on the same trades. AIUP's October 2026 top three holdings — Micron Technology at ~14.5%, Sandisk at ~14.1%, and Meta Platforms at ~14.6% — span 43% of the fund in three names. AI infrastructure and mega-cap tech dominate the rest: NVIDIA, Arista Networks, Palo Alto Networks, Amazon.

Goldman Sachs' March 2026 quant commentary attributed an earlier drawdown to three converging factors: \"losses in crowded long positions, adverse moves in high-beta short positions, and crowded factor exposure.\" That same convergence risk compounds when the underlying inputs — earnings transcripts, macro feeds, sentiment — are also AI-curated. If HIAI's BAILA model, FINQ's daily ranker, and Pictet's enhancer all train on overlapping signal universes, the 2026 AI-fund cohort could become a single crowded trade dressed in five different wrappers.

The legal architecture is permissive on purpose. The SEC has no AI-specific rule for investment advisers — the 2023 proposal was withdrawn — and Form ADV Part 2A already obliges advisers to disclose methods of analysis and investment strategies. Director Daly's Feb 3, 2026, speech told advisers to \"come talk to us about AI\" and framed agentic AI in proxy and portfolio decisions as a \"near-term reality\" rather than a future concern.

The practical effect: incumbents face a higher compliance bar to credibly claim AI use, while dedicated AI managers can publish prospectuses that already name the model. BlackRock is leaning into this asymmetry — its Q3 2026 earnings call confirmed systematic equity now accounts for ~$500B of AUM and30–40% of annual iShares volume.

The transmission into listed equities runs in two directions. Active managers are the incumbents being priced; mega-cap AI infrastructure names are the convergence trade that AI funds are increasingly forced to hold.

The second-order beneficiaries are not the AI managers themselves — they are private (FINQ, Pictet, Ai Funds). The first-order capital beneficiaries are the stocks the AI managers must own as the cohort converges. Micron Technology sits as AIUP's largest position because memory bandwidth is the bottleneck in AI training; Meta Platforms and NVIDIA sit because hyperscaler capex is the through-line. If active ETF AUM triples to $4.2T by 2030 and even5% of that lands in AI-managed wrappers, that is roughly $210B of structural flow into the same twenty mega-cap AI-infrastructure names — a tailwind that flows directly to issuer fundamentals.

Short-term (days to quarters): watch AI fund AUM. Pictet's $92.4M in three months proves the fee structure scales. The signal will be when AIUP — currently $5.5M — clears $50M and $100M. That crossover is the moment retail and RIA platforms begin to add the wrappers to model portfolios. Each subsequent launch (HIAI, AINT-style long-shorts, Pictet's world ex-US variant) compresses the time to that crossover.

Long-term (one to three years): the displacement of human-led active equity is incremental, not catastrophic. T. Rowe Price's $12.2B September outflow is 0.65% of AUM — meaningful, not existential. BlackRock's projected $2.8T of active-ETF additions by 2030 can absorb AI-managed growth many times over before it starves incumbent active managers. The real winners are issuers that own the AI-fund convergence trade: NVIDIA, Microsoft, Meta Platforms, Amazon, and the memory supply chain anchored by Micron Technology. The real losers are mid-tier active equity shops that cannot automate at 22 bps.