It depends. Although we have witnessed the exponential growth of computer performance over the past four decades, its very likely that we will reach the theoretical limit in the near future. For example, the size of transistors is very close to the limit where electrons can easily tunnel through the gate insulator and switching off the transistor becomes difficult. Although using novel designs like high-K insulators and 3D transistors can make devices even smaller, the fundamental limit still exist. Another limit is power consumption. Our state-of-the-art devices consume 6~7 orders of magnitude more power than the Landauer limit, a theoretical limit predicted by thermodynamics. Besides we can never reach the Landauer limit because at this point the error rate due to thermal fluctuations becomes unacceptable. So there is not much room for improvement in the energy efficiency either.
That being said, the Landauer limit does not prohibit the computer to solve the problems using more efficient algorithms. For example, the quantum algorithms running on quantum computers can bring exponential improvements to certain problems like factorizing large numbers, quantum system simulation, and solving gigantic systems of linear equations. If we can solve the decoherence and scalability problems, even a microchip containing only 1000 qubits can quickly solve problems that would take billions of years for even the fastest supercomputers to solve (so called “quantum supremacy”).
Another way to bypass the Landauer’s limit is using reversible algorithms. Landauer limit doesn’t prohibit zero-power computation as long as no data is erased in the process. If we can design a special algorithm that can cancel out all intermediate results instead of simply erasing them while leaving only the final result, we may be able to do complex computations while consuming little power. And like quantum algorithms, reversible algorithms cannot run on conventional CMOS devices (the building blocks of modern computers), but more exotic ones like superconductors.
However, even if these novel computers are invented, they can only solve some specific problems with very high efficiency, otherwise they are not much different from conventional computers.