Gravitational waves from similar to 90 black hole binary systems have been detected and their progenitors' properties inferred(1) so far by the Laser Interferometer Gravitational-Wave Observatory(2) and Virgo(3) experiments. This has allowed the scientific community to draw conclusions on the formation channels of black holes in binaries, informing population models and at times defying our understanding of black hole astrophysics. The most challenging event detected so far is the short- duration gravitational-wave transient GW190521 (refs. (4,5)). We analyse this signal under the hypothesis that it was generated by the merger of two nonspinning black holes on hyperbolic orbits. The configuration best matching the data corresponds to two black holes of source-frame masses of 81(-25)(+62) M-circle dot and 52 (+32)(-32) M-circle dot undergoing two encounters and then merging into an intermediate-mass black hole. We find that the hyperbolic merger hypothesis is favoured with respect to a quasi-circular merger with precessing spins with Bayes' factors larger than 4,300 to 1, although this number will be reduced by the currently uncertain prior odds. Our results suggest that GW190521 might be the first gravitational wave detection from the dynamical capture of two stellar-mass nonspinning black holes.