This paper presents an experimental study of dynamic indefinite horizon R&D races with uncertainty and multiple prizes. The theoretical predictions are highly sensitive: small parameter changes determine if we should expect technological competition, and if so whether it is sustained, or if the market converges into one with entrenched leadership and lower aggregate R&D. The subjects’ strategies are far less sensitive. In most of the treatments, the R&D races tend to converge to entrenched leadership. We propose and apply a quantal response extension of Markov perfection that is qualitatively and largely quantitatively consistent with the experimental observations.