Profitability of competing flexibility options in renewable-dominated energy markets: Combining agent-based and machine learning approaches

German transmission system operators received around 400 GW of battery storage connection requests in 2024, an enormous increase from the current 2.3 GW installed capacity. However, uncertainty remains regarding the impact of large-scale storage deployment on electricity market dynamics and a potential cannibalization of profits. To investigate this, we use the open agent-based electricity market model AMIRIS. We extend it with machine learning-based electricity price forecasts during runtime. This enables modelling of competition among flexibility options. We evaluate forecast accuracy and its effects on the operation and profitability of competing flexibility options, with implications for flexibility option revenues. Operational strategies significantly affect revenue: dynamic programming approaches exploiting all profitable spreads yield higher revenue and more charge cycles than threshold-based strategies filtering for spreads exceeding safety margins around mean prices. We then parametrize the model to a 2030 scenario based on the Ariadne report that shows a pathway towards a decarbonized German energy system. At the system level, we identify a profitability plateau for homogeneous storage systems with installed power between 4 to 8 GW and 32 GWh total capacity. Depending on installation costs, annual return on investment can reach around 20% through day-ahead market arbitrage. For heterogeneous storage technologies, requested battery capacities far exceed economically profitable levels implied by simulated market-based returns. This suggests that current German grid connection requests may be economically unsustainable under current market structures. Future research can build on our open tool chain to integrate additional revenue streams, including cross-market participation and system services remuneration.