--- title: "Why does an AI trading strategy lose its edge as it gets bigger?" slug: markets-ai-trading-strategy-capacity-ceiling date: 2026-06-11 excerpt: "AI surveys keep crediting systematic funds with a three-to-five-point edge, yet the best funds are closing to new money. The square-root law of market impact explains why a strategy's edge cannot scale with the capital poured into it." featured_image: "https://bbtxujdxvidaghmhxkqs.supabase.co/storage/v1/object/public/generated-images/blog-1781168418041-markets-ai-trading-strategy-capacity-ceiling.webp" featured_image_alt: A rising curve of fund assets under management plotted against a flattening curve of net return, with the gap between them widening as size grows, illustrating a capacity ceiling. canonical_url: https://cerevisor.com/blog/markets-ai-trading-strategy-capacity-ceiling updated_at: 2026-06-11T09:00:19.176818+00:00 --- # Why does an AI trading strategy lose its edge as it gets bigger? TLDR The AI-edge surveys keep crediting systematic funds with a three-to-five-point annual advantage, yet the strongest funds are quietly closing to new money. The reason is the square-root law of market impact: the cost of trading grows with the square root of how much of a day's volume you push, so a strategy's edge cannot scale one-for-one with the capital behind it. Useful for edge sizing and for knowing when an edge has stopped being real. Last week Goldman Sachs’ prime brokerage desk reported that hedge funds bought US equities at the fastest pace in six months, with net leverage sitting near the top of its one-year range and the S&P 500 riding a nine-week winning streak. Underneath that risk appetite sits a claim repeated all year: that AI hands systematic funds a durable, repeatable edge, with some industry surveys putting the gap between AI adopters and everyone else at three to five percentage points a year. The claim is plausible. At scale it is also, in part, an accounting illusion, and the reason why is one of the most reliable pieces of arithmetic in markets. --- ## The claim, and the money chasing it The pitch has two halves. First, that AI compresses research and surfaces signals faster, which is true. Second, that this advantage compounds as a fund grows, which is where the trouble starts. Coverage this month described the six-point edge systematic funds once commanded as fading rather than compounding, as the same foundation models and the same datasets spread across the industry and the trades converge. Now look at what the funds actually do with their own money. If an edge genuinely scaled with capital, the rational move would be to take every dollar on offer. Instead, after the industry pulled in roughly $116 billion of net inflows in 2025, its strongest single-year haul since 2007, a growing list of managers is refusing new cash. Funds capping assets rather than diluting the edge (2026) FundCapped near Helikon Investments**$8bn** Kintbury Capital~$3bn ADAPT Investment Managers~$2bn Greenvale Capital~$1.8bn These are not weak funds turning away money out of modesty. ADAPT closed after returning close to 16% in 2025. The contradiction is the whole story: a fund that believed more capital meant more edge would never cap itself. The act of closing is a confession about a ceiling. > "net leverage at the 89th percentile of its one-year range and the fundamental long/short ratio at the 99th" FXEmpire, on Goldman Sachs prime-brokerage positioning, June 2026 That positioning matters here because record leverage into a crowded book is precisely the condition under which the ceiling stops being theoretical. --- ## The square-root law of market impact Here is the mechanism, in plain English. When we buy, we push the price up against ourselves before the order is done. That cost, called market impact, does not grow in a straight line with order size. It grows roughly with the square root of how much of a name’s average daily trading volume you need to move. The CFA Institute put the point cleanly this spring: doubling the size of an order does not double its cost, and a manager buying even 2% of a day’s volume no longer trades at the quoted price. The square root looks like good news, and for a single trade it is. The problem appears when capital scales. To put twice the money to work in the same strategy, a fund must trade a larger slice of each name’s daily volume, every day, across hundreds of positions. The cost per dollar of edge captured climbs even though the cost per share rises sublinearly. Research on AQR’s own live trading found these costs ran an order of magnitude larger than the quoted bid-ask spread. Somewhere on that curve, the next dollar of assets funds a trade whose impact cost exceeds the edge that dollar was supposed to harvest. That point is the capacity ceiling. Key Insight AI moves the numerator (the speed and breadth of finding signals). It does nothing to the denominator (the liquidity of the names a fund must trade). The capacity ceiling is set by the denominator, so a smarter model raises the edge per trade but not the amount of money that edge can carry. > A strategy does not run out of edge because the model got worse. It runs out of edge because the next dollar has to trade a fatter slice of the same shallow pond. There is a second twist when the models are shared. If many funds run similar systems on similar data, they crowd into the same liquidity-limited names, which lowers the ceiling for everyone at once and synchronizes the exit when the trade unwinds. --- ## The honest version of an AI edge A fair statement of the edge is not “AI adds three to five points.” It is “AI adds three to five points at a few hundred million dollars in these names, and that figure decays toward the market as the assets grow into the billions.” A liquid strategy at $500 million, as the trade press has noted, may simply not be liquid at $10 billion. The edge is real and the decay is real, and both are true at the same time. This is why the fund that publishes its capacity and closes on schedule is behaving more honestly than the one that keeps the doors open and lets the average return drift down quietly. The first is pricing the square-root law. The second is hoping we do not. --- ## What we do with this For sizing an allocation, treat a fund’s decision to close to new money as a positive signal about its candor, not a problem to route around. For reading a track record, discount returns earned at small size when the same strategy is now running many times the capital in the same instruments. And for our own orders, the law applies in miniature: buying a thin mid-cap in one click at the open quietly costs basis points that splitting the order over a day or two would not. We are all somewhere on the same curve, just at different sizes. The question worth sitting with is not whether AI funds can beat the market. Plenty can, at the right size. It is what happens to a whole cohort of strategies that found the same edge with the same tools, raised into it together, and now sit on the same crowded names while the leverage that funds them reads near the top of its range. This is editorial analysis, not investment advice. Cerevisor does not hold or recommend the named positions, and information here can become stale within hours of publication. #### Sources - [Goldman's Six-Month Hedge Fund High Hides a Fragility Signal](https://www.fxempire.com/forecasts/article/premium-goldmans-six-month-hedge-fund-high-hides-a-fragility-signal-1603326) - FXEmpire, 2026-06-09 - [Goldman Says Hedge Funds Buy Stocks at Fastest Pace in 6 Months](https://www.bloomberg.com/news/articles/2026-06-01/goldman-says-hedge-funds-buy-stocks-at-fastest-pace-in-6-months) - Bloomberg, 2026-06-01 - [The New Demands of Optimal Execution](https://rpc.cfainstitute.org/blogs/enterprising-investor/2026/new-demands-optimal-execution-machine-learning) - CFA Institute, Enterprising Investor, 2026-04-27 - [Quant Funds Face a Capacity Squeeze: When Too Much Capital Threatens Alpha](https://www.hedgeco.net/news/04/2026/quant-funds-face-a-capacity-squeeze-when-too-much-capital-threatens-alpha.html) - HedgeCo Insights, 2026-04-15 - [Hedge funds close to new capital after year of strong performance](https://www.hedgeweek.com/hedge-funds-close-to-new-capital-after-year-of-strong-performance/) - Hedgeweek, 2026-02-06 - [AI Trading Crowding Erases Quant Edge, 2026](https://pomegra.io/news/ai-trading-crowding-erases-quant-edge-2026) - Pomegra News, 2026-06-04