x may be wealth OR increment to present wealth, but it is important to keep in mind that the appropriate risk tolerance may increase as wealth increases. Modern portfolio theory is based on the implicit assumption that risk tolerance is PROPORTIONAL TO WEALTH, though empirical validation or invalidation of this assumption remains to be done. The idea of wealth dependent risk tolerance has been built into modern decision analysis programs such as DPL which allow the applicable risk tolerance to be set to different levels in different parts of a decision tree, depending upon what has transpired along the path up to any given chance node.
Note: Since addition of a constant and/or multiplication by a positive constant leads to another strategically equivalent utility function, we could have written the exponential utility function in the form A - B*EXP(-x/RiskTol). The given normalized form has the advantage that U(0) = 0 and the limit of U(x) as x -> infinity is 1. The finite upper bound on utility as x goes to infinity can be regarded as a reflection that man (and woman too!) is a finite being, and the utility placed on any monetary amount is finite as well. We can imagine the monetary amount going to infinity, but the utility assigned to the money remains finite.
Hence, in fact, the Exponential Utility function is a ONE-PARAMETER family of Utility functions, the one parameter being tau or RiskTol, for risk tolerance. Thus for this model, risk attitude is represented by a one-dimensional scalar quantity. Extremely risk averse attitudes are modeled by small risk tolerance values approaching zero (when decisions are made based on "worst case" outcomes), and extremely risk tolerance risk averse attitudes (called risk neutral) are modeled by large risk tolerance values approaching infinity (when decisions are made based on "expected monetary values" or EMV). Hence exponential utility may be viewed as a way of filling in the "in between" region where decision makers are neither extremely risk averse nor risk neutral, that is, somewhat risk averse without being extremely risk averse or risk neutral. Stated another way, one can say that exponential utility fills the gap between the MAXIMIN and the MAXEMV decision rules that apply at the extremes.