Artificial intelligence (AI) now permeates the entire investment workflow, from data acquisition and prediction to research, portfolio construction, execution, and tooling. Technical capability alone does not imply investment profitability.
This review synthesizes publicly available research up to August 31, 2026, covering listed equities, ETFs, centralized crypto spot, perpetual futures, and on‑chain markets. We organize the evidence using an “alpha‑translation chain”: point‑in‑time information → stable signal → feasible position → executable order → risk‑adjusted return after costs. Across machine learning, time‑series foundation models, financial language models, reinforcement learning, and agents, progress is evident in upstream tasks such as prediction, text processing, portfolio design, and workflow integration, yet evidence for durable net performance remains thin.
Temporal contamination, repeated selection, survivorship bias, weak benchmarks, implementation costs, venue mechanics, and capacity constraints can erode the translation to net alpha. Strong historical backtests coexist with predictor decay, corrected look‑ahead failures, mixed prospective evidence, and very few audited live‑capital records. Crypto adds informative state but requires separate handling of spot, perpetual, and decentralized cash flows and execution. No general AI architecture in the public literature demonstrates persistent, cross‑regime, capacity‑aware net alpha. More credible claims demand point‑in‑time data and models, decision‑aligned objectives, joint portfolio‑execution evaluation, controlled adaptation, prospective testing, and governance matched to regulatory authority. These conditions can improve evidence quality and implementation, but they do not guarantee profit.
Review