Fast-charging hubs for electric vehicles will soon become part of the newly built infrastructure for transportation electrification across the world. These hubs are expected to host many DC fast-charging stations and will admit EVs only for charging. Like the gasoline refueling stations, fast-charging hubs in a neighborhood will dynamically vary their prices to compete for the same pool of EV owners. These hubs will interact with the electric power network by making purchase commitments for a significant part of their power needs in the day-ahead (DA) electricity market and meeting the difference from the real-time (RT) market. Hubs may have supplemental battery storage systems (BSS), which they will use for arbitrage. In this paper, we develop a two-step data-driven dynamic pricing methodology for hubs in price competition. We first obtain the DA commitment by solving a stochastic DA commitment model. Thereafter we obtain the hub pricing strategies by modeling the game as a competitive Markov decision process (CMDP) and solving it using a multi-agent deep reinforcement learning (MADRL) approach. We develop a numerical case study for a pricing game between two charging hubs. We solve the case study with our methodology by using combinations of two different DRL algorithms, DQN and SAC, and two different neural networks (NN) architectures, a feed-forward (FF) neural network, and a multi-head attention (MHA) neural network. We construct a measure of collusion (index) using the hub profits. A value of zero for this index indicates no collusion (perfect competition) and a value of one indicates full collusion (monopolistic behavior). Our results show that the collusion index varies approximately between 0.14 and 0.45 depending on the combinations of the algorithms and the architectures chosen by the hubs.