Google, IBM and Rigetti use transmon qubits; these are basically fancy LC circuits where there is a josephson junction coupled to a superconducting island. Because of this, they are also often referred to as superconducting qubits. The qubit states are the various charge levels that can exist on the circuit; since the lowest two levels are separated in energy with respect to the higher levels, a two-level system arises.
Intel also used superconducting qubits, but lately has also been interested in quantum dot qubits. A quantum dot is like a 0-dimensional island on which a single electron can be placed; since the electron is a fermion it has only two natural states (and therefore makes a good qubit). The encoding can also be different, by encoding the qubit into two rather than one electron in the quantum dot. These quantum dots are built on semiconductors (like silicium, known as the go-to material in classical computing). Therefore they are also known as semiconducting qubits.
Microsoft is trying a different route: they are trying to built a topological quantum computer. This is a different type of quantum computer where the qubits are encoded in topological states of matter, using quasi-particles known as (non-Abelian) anyons. A likely candidate for a physical implementation is the Majorana fermion, which can act as an anyon. You can think of these quasi-particles as a delocalized pair of electrons on a super-conducting bridge. It is worth noting that this is a considerably harder design than your 'run of the mill' transmons etc, but these topological states are intrinsically protect to many types of noise, thereby reducing or even omitting the need for quantum error correction.
D-Wave's systems is based on a yet more different method of quantum computing: the adiabatic quantum computer. The way computations are performed on these computers are not alike the circuit model (which is the most used model, exploited by transmons, super-conducting and semi-conducting qubits and the like). Moreover, the qubits themselves act very differently, and the comparison of 'adiabatic-syle' qubits and 'circuit-type' qubits is not a good or well-defined comparison. An adiabatic quantum computer needs many more qubits to have the same computational power as a circuit-based quantum computer, but they are (at least on paper) equally powerful (in terms of complexity classes).
There are also other types of qubits (that are not used by any of the companies you listed). Two to look out for are:
Trapped-ion qubits. Qubits are encoded into states of ions; these ions are trapped by optical tweezers (light) and therefore localized and isolated.
Photonic quantum computation. Qubits are encoded into degrees of freedom of photons (=light), most often the polarization. These photonic machines normally use the computation model of measurement based or one-way quantum computation, which is comparable to the circuit model but creates all entanglement in the beginning of the computation.
There is no clear best implementation (yet). Transmon qubits are the most mature by most standards, but they are relatively big which will give big implications and problems when these devices will be scaled to include millions of qubits. Semiconducting qubits are a very interesting candidate because they are much smaller and implemented on (the very well developed technology of) semiconductors, but not much has been developed. Trapped ions are promising as well, but they can only be manufactured in a line (as a one-dimensional array of qubits). I'm interested to see what will happen with photonic quantum computers; they can be very promising but not many large companies are working on them; the measurement based model of QC is less popular.
A topological quantum computer is the dream of many, but for now it seems out of reach in the near future, due to the exceedingly exotic nature of its design.