It is an insufficiently remarked feature of Kant's discussion of God in the Critique of Pure Reason that the idea of God follows directly from our ability to make disjunctive judgments and reason using disjunctive syllogisms (or eliminative arguments). Deleuze talks about it somewhere (Logic of Sense, I'm thinking), and people will mention it when summarizing the Critique closely, but that's about it. I find this curious, because it is the foundation of Kant's entire discussion of the arguments for the existence of God, which is perhaps the most widely discussed portion of the Critique; and it's also a pity, because I think it's easier to understand what Kant's saying if you emphasize this aspect of Kant's discussion rather than passing over it quickly.
Some basic background: Kant holds that there are three basic kinds of judgment, each indicating a different kind of relation. There are categorical judgments, hypothetical or conditional judgments, and disjunctive judgments. These play a major role in CPR. For instance, Kant's discussion of substance is actually about predication (the relation in categorical judgments, which say that an attribute is in something); his discussion of causation is actually about ground and consequence (the relation in hypothetical or conditional judgments, which say that one thing follows from another), and his discussion of coexistence or reciprocity is actually about composition (the relation in disjunctive judgments, which say that a whole is divided into things that are opposed to but united with each other). One can easily misinterpret Kant if one ignores this.
Each of these judgments is associated with a particular kind of synthesis or unification that reason can exercise. In categorical judgment, reason thinks of things as belonging to a subject; in hypothetical judgment, reason thinks of things as belonging to a series; and in disjunctive judgment, reason thinks of things as belonging to a system. Likewise, each kind of judgment can be used as the major premise of a syllogism, giving three kinds of syllogism: categorical, hypothetical, and disjunctive. Disjunctive syllogisms are what we usually think of as eliminative arguments: We take a 'sphere of cognition', to use Kant's term, divide it up into options, and then limit or restrict the options in order to get a conclusion. The two functions, synthesis and syllogism, are necessarily related; it is our ability to form judgments that make possible both the interpretation of experience (as we might call it) and reasoning about experience, and by linking the two, Kant argues that he has established that our interpretation or synthesis of experience has a necessary foundation in human reason itself.
From our ability to make each of these three kinds of judgment come what Kant calls the transcendental ideas -- transcendental ideas are concepts of reason that have no sensible object (indeed, cannot possibly be an object of sense) but which provide a sort of ultimate or unconditional unity to our experience. We could only form such ideas by one of the three judgments, and so there are three key transcendental ideas. The first such idea is that of a subject that is never a predicate; the second is that of a ground or presupposition that requires no further ground or presupposition; and the third is that of a totality of distinct possibilities that requires nothing further in order to include all distinct possibilities. Each of these follows directly from our ability to make one of the kinds of judgment and reason using one of the kinds of syllogism. But when we look more closely at these three transcendental ideas, it becomes clear that they are quite familiar, and Kant's entire argument is that these familiar non-empirical concepts of reason are inevitable given our ability to form judgments and to reason. The transcendental idea that follows from our ability to form categorical judgments and reason with categorical syllogisms is, when represented as being something particular, that of thinking substance, the I or Ego, the undivided individual, the unifier of all experience. The transcendental idea that follows from our ability to form hypothetical judgments is (when represented as being something particular) that of freedom, the kind of thing that determines other things but requires nothing to determine it, the unconditional condition: a unifier of series that depends on no prior conditions. And the transcendental idea that follows from our ability to form disjunctive judgments is (when represented as being something particular) that of God, a being of beings, an undivided individual unifier of the entire system of possibilities.
Think of a disjunctive syllogism:
A or B
The major premise here is the disjunction, A or B. A or B could be anything, and it can be as long as we please. What happens, however, when we push it to its limit, so that A, B, etc. include all possibilities? The ultimate major premise would be the whole system of possibility. If we posit a principle capable of unifying such a system we get a most-real being, an ens realissimum, whose possibility makes possible (so to speak) all other possibility, and whose possibility includes such a richness that other possibilities can be said only to approximate it. This is God considered transcendentally. He is the ideal (the hypostasis or ground) of the idea of a system of all possible being, which comes directly from the disjunctive judgment taken to its limit.
Kant being Kant, of course, he insists that this is only the idea and the ideal, not anything that can be asserted to be real. Disjunctive judgments according to Kant are only problematic; they cannot of themselves be assertoric. That is to say, disjunctive judgment only gives you possibility. What is more, in achieving the transcendental idea we are not even establishing the possibility of God, properly speaking: we are establishing that reason can formulate the idea of the possibility of an unconditional unifier of possibility, and can represent it as an ideal for reason's task of synthesizing experience, which reason increasingly approximates but which reason can never actually reach; this ideal is God. All we've reached is a principle organizing our concepts, not an object of a concept. And Kant is very insistent that we can only take things as real or actual when they are given to our sensibility, i.e., when we are actually affected by them. Because of this, Kant calls the representation of the idea in this particular principle, God, a 'dialectical illusion'; which is to say, that it depends entirely on the subjective tendency of our mind rather than any objective foundation. ('Illusion' here does not indicate that it is 'illusory' in our usual sense, but only that it is based not on our experience but on reason's natural progress in connecting things. That is, it's due not to any experience of God's actuality, but solely to the fact that if reason keeps doing what it does, God is the ideal limit of that.) We could put Kant's point another way. The concept of God is a need of reason, not a necessity of experience. But as a need of reason, it is inevitable, if reason is given its due and not arbitrarily or accidentally prevented from following out the implications of its task.
This is all presupposed in Kant's discussion of the arguments for the existence of God. Were it just a matter of heuristic supposition, human beings would not (necessarily) put much weight on it, but there are other relevant factors. We have what might be called a restlessness to get to this end-of-the-road for reason's task, and therefore, on the basis of actual experience, we tend to try to get to find something in experience that will be both necessary and a ground for all these disjunctive possibilities. This leads to something like the cosmological argument, which attempts to argue on the basis of experience of the contingent for some necessary being. Once this necessary something-or-other is established, reason has to find some concept that fits it, and it does so by elimination, ruling out all concepts that are inappropriate for a necessary being. This is why Kant insists that we have to show that the necessary being is the ens realissimum; every argument for the existence of God requires a two-step process: an argument to something and an identification of that something with ens realissimum, the being of beings. Thus the argument is guided by this prior ideal of reason. The ontological argument is the argument that attempts to argue that because we have this ideal of reason, it must have some corresponding object; however, this argument won't work for the reasons given in the previous paragraph. The reason Kant argues that the cosmological argument (from contingency) and physico-theological argument (from design) presuppose the ontological argument is that their improvements over the ontological argument consist entirely in the first step of the argument, to something's being necessary. Whether this is God, however, depends entirely on whether it is also the object of the idea of ens realissimum; but we have only reached necessary being conceptually as the possible completion of the task of reason, which means that we haven't actually reached an object that is the necessary being; so in order to come to an object that is necessary being and thus ens realissimum, we must get to the object by way of concluding that the idea of the ens realissimum requires that its object be the necessary being -- which is just the ontological argument. Neither the cosmological nor the physico-theological argument, on this account, can get around the fact that it cannot prove that the ideal of reason is anything more than just a supposed possible completion of reason's task, except by the ontological argument (which concludes directly from the ideal to positing its object).
The idea of God is implicit in reason itself, namely, in its ability to form disjunctive judgments, thereby making it possible for us to think of a being of beings as grounding the limit case of these judgments. The very fact that we are capable of reasoning with disjunctive syllogisms or eliminative arguments guarantees our ability to have this idea, and, indeed, the necessity of developing it if we are taking reason seriously. But we can never reach an object of this idea. In a sense, it's the very absoluteness of the idea, its position as the goal and end of the capacity to make disjunctive judgments, that guarantees the failure of all speculative arguments for the existence of God. The idea is the posited expectation, so to speak, of the end of the road of looking at the world in terms of systems, the one that would ground the system of all possibilities; but this is the limit of an ongoing task of reason that can never be completed by experience, which supplies objects for our concepts; and therefore the only way of getting an object for this idea of God, and thus saying that God exists, is by assuming somewhere that you are at some point allowed to get the object immediately from the ideal.
The moral argument, on the other hand, is able to escape this problem because it does not rely entirely on the same task of reason as the ontological argument, but depends on an entirely different tendency of reason. This is the only kind of argument Kant allows. This does not mean -- and it is important to understand this -- that every other argument for God's existence is useless; it does mean that the most it could ever serve is a heuristic function. It also does not mean that the atheist is better off than the theist, even if we ignore moral arguments; given Kant's argument, the atheist can no more prove that God does not exist with speculative arguments than the theist can prove that God does, but the theist at least has the excuse that reason's overeagerness to conclude that God exists is based on a natural illusion. This illusion arises directly from the natural ability of reason to judge and reason with disjunctions, combined with the natural tendency of reason to try to achieve the greatest unity it can in understanding the world. It is, if you will allow the expression, a reasonable illusion; the ideal of God is itself something that must be posited by anyone who genuinely takes reason seriously. It is rationally faultless, and has to be, because it is just the positing as an ideal of the limit case of reason's own task. We just can't find anything that can genuinely guarantee that it has an object, at least if we are dealing with pure reason or proofs that lead to knowledge.
Such is Kant, anyway. There are obviously a number of questions that this account raises, but it's an interesting one that hangs well together. (My own view, in case you are wondering, is that it's sound in terms of its structural principles; any problems lie with Kant's assumptions about the nature of experience, which I think are often problematic and sometimes egregiously wrong. But that's a complicated issue.)