A next step device which can demonstrate long burn with a high probability of success is possible, without speculating on the accumulation of favorable effects and ignoring the other ones. From the data base available today this appears possible only with large devices such as NET, ITER or JIT. If such a device is to provide the needed database to go to DEMO it must be able to explore a sufficiently wide range of parameters. From our present knowledge this can be obtained by choosing a conservative value of the reference magnetic field away from technological limts and compensate with a larger size. There remains for all devices very serious physics problems to solve, the most acute being the exhaust which must be compatible with a good quality enhanced confinement regime and a sufficiently high wall load to make a reactor.