We are considering the problem of observer design for a general class of state-affine systems in presence of output delay and sampling. Two novelties characterize the considered class of systems: (i) the state equation is subject to output injection involving future output values (not accessible to measurement); (ii) the injected output values come in the state equation not only through a driving term but also through the state matrix. These novel characteristics entail the loss of the system state-affine nature and lead to a new observer design problem never investigated so far. The solution we develop in this paper is a sampled-data based observer of Kalman-like type, augmented with inter-sample predictors and signal saturations. Using both Lyapunov and small-gain arguments, we show that the observer is exponentially convergent for sufficiently small sampling interval and delay, provided an observability condition is satisfied.