Speak for yourself.
I can think of numbers without materializing said number.
But you at least know what number it is for yourself. Otherwise you're not thinking of a specific number but about possible numbers or sets.
I bet none of them have been a randomly selected natural number from the "set of all natural numbers", otherwise we should be amazed at the probability of it not being over a googleplex (10^(10^100)).
Now if you happened to pick a random number over a googleplex, we should then be amazed at the probability that it was so easily compressible into a small finite representation such as (10^(11^(12^13))), for which "nearby" integers would typically not be so easily factored and require a massive number of digits to represent.
Yet no matter how large or complex a description for a number, we're effectively not even moving relative to infinity ... or are we?
Well, what's infinity in terms of natural numbers? It's larger than any natural number.
So if we counted to 3: 1,2,3 then infinity could be 4 ... except that once we created 4, then infinity, being larger would have to be 5 ... oops, I just created 5, so infinity would have to be 6, yikes, this appears to be circular.
Ok, let's try defining (infinity-1) to be the largest natural number and that makes sense because once we've gone past natural numbers (I'm just working on a natural number line), we're at infinity.
So if we try counting to 3 again, 1,2,3 then the next number should be infinity, so it would be 1,2,3,infinity or 1,2,3,... and we don't have a problem here because infinity isn't a natural number and infinity+1 isn't defined, so we can finally stop counting as we've reached infinity

Notice that we also no longer have a problem with trying to define infinity+1=infinity, which implies infinity=infinity-1 and leads to the vicious recursion infinity=infinity-1-1-1-1...=infinity-infinity.
If we begin with 1,2,3,infinity and if we created 2*3=6, would 6 now be larger than infinity? By definition, it couldn't be and if 6 was the largest natural number then infinity would be ~7.

The point of this is to show that infinity needs not preexist as some incredibly massive quantity that we can't even begin to approach but instead is dynamically created by processes and infinity effectively exists in the present.
In some mathematical texts there are various forms of infinities that may or may not be arranged into various magnitudes or relative rates of growth, and consider how such structures are created:
If we began to create a row of numbers by progressing with this sequence:
11 1
2 12 1 1
2 2 1
3 2 13 2 1 1
3 2 2 1
3 3 2 1
4 3 2 1...
We could "grow" this over time into this sequence
1
2 1
3 2 1
4 3 2 1
5 4 3 2 1
...
Now notice that every number at each position along this row never decreases and is periodically incremented. If we continued incrementing individual numbers according to the top sequence an "infinite number" of times, then we'd end up with a row of numbers and each number growing "infinitely large".
So we've converted a single infinite quantity into an infinite quantity of infinite numbers. (It's magic, right?)
Well, at moment, this structure is nothing more than a finite number of finite values and all that "infinity" means is that it's intended to be continually increased, but obviously we couldn't have it increased without being synchronized with us in time, otherwise we'd have no ability to measure specific values. For example, if even one of these numbers was truly infinite, in the context that if we were to somehow be able to measure it, we could never measure it as a specific number, then is there any usefulness to infinity in that respect. If there is, it's not very obvious to me.
Now image that we jumped forward to some unknown
time in this sequence (we took a vacation for 20 years and someone else kept repeating this pattern into larger and larger representations (in space ... he has to have room to record the results, otherwise it's some other sequence)), what properties could we assume to still hold true?
If we knew approximately how many updates per second were occurring over that period (we're assuming there's a synchronization in time between us and the generation of this structure) we could
divide this down in various ways to determine approximately what values should be at each location.
Also by knowing a single value at some location in this sequence (and these values must maintain a specific linear ordering in space (a spacial dimension) in order that "next" elements in the row can be determined), we could determine within a single unit what all the other values were in the row as we know they differ by 1, except for a possible single duplicate element.
Anyway, we could quickly "synchronize" ourselves with the state of this count by knowing just a few properties.
Now if we jump back to the vague description of having an infinite quantity of infinite numbers, we could try to create a table showing which of these are larger than other values and this should lead us to the question of which is the largest.
If we simply look at the numbers, no number in the row will be larger than the first number and in that case we might define infinity to be anything larger than the first number in the row, but there's still a quantity involved, though not immediately "tangible" that's larger than this ... I'll give you a second to think it over ... ok, 2 more seconds ... any guesses as to what's the largest quantity in this structure? It's the number of steps or increment operations that were performed.
So if we added up all the numbers, we could determine how many (vague) "units" of time this process had undergone, though it also takes a certain amount of time to locate elements in the row and add new spaces for additional digits or terms and these aren't ignorable in real life, but notice that if we added all the numbers together we get a simple count over time, 1,2,3,4,... or 1,2,3, infinity and "infinity" is always what's happening in the present.
Though there could, from one perspective, be an infinite number of infinite quantities, these infinities refer to the fact that there are no predetermined bounds on the time the process could execute for (though a random asteroid strike, boredom or a power outage could terminate infinity early) and there is truly only a single "infinite" value involved which is subdivided into ever increasing subsets that are smaller than it, just as we could create two infinite values by alternating operations between two numbers, but if we performed these operations on a single value, it would be (at least approximately) updated twice as fast. So there's always a largest infinity and it's always dependent upon what operations are available to be performed over time.
Imagining things beyond this leads to confusion over what properties infinity possesses.
In calculus we have functions that are reliant upon the statistical properties of convergence for an unbounded sequence such as 1/2+1/4+1/8+1/16+... or .999...
The properties of these sequences are based upon physical concepts of proximity or nearness or smallness etc.
For example, the "proof" that 1=1/2+1/4+1/8+1/16+... is based upon the idea of recursively cutting an object into halves, but there's a paradox in that if an object can always be cut in half, then something must always remain after than cut and the remainder cannot be 0 and we can prove that 1 is not equal to 1/2+1/4+1/8+1/16+... using similar logic because we begin with 1/2 and constantly multiply this value by 1/2, which generates the sequence, 1/2, 1/4, 1/8 and at each iteration the value is a positive, non-zero number by definition, if a>0 and b>0 then a*b>0 and so is (a*

*b>0 etc. and we can also verify this by recognizing that only a multiplication by 0 results in 0 and we never began with a zero available to generate a zero result from multiplication, so the value simply converges toward, but never reaches this boundary or limit.
Now the question could become one of, whether or not this difference can always be made smaller than any other number.
Notice that this requires, similar to concept of multiple infinities, that we never expect any other number to also be the smallest number.
So if we had the two series .5, .75, .875, ... and .9, .99, .999 we can see the differences between these and 1 as decreasing 1-.5=1/2, 1-.75=1/4, 1-.875=1/8, ... and similarly 1-.9=1/10, 1-.99=1/100, 1-.999=1/1000, ...
So we have two decreasing series 2^-n and 10^-m that converge toward 0, but can we select an element from each set that's smaller than the other? No, and for n.m>0 they can't be equal to each other either.
If we never used more than a single such limit in an equation, there would generally be no such problem and we could simply compute what a convergence appeared to tend toward - notice that the reason why this is possible is because there's only a single infinite quantity (time) and we would have no issues about how to multiplex it between various processes to determine a desired result because we'd only have a single choice - continue with the next value in a single series, but if it comes to generating multiple "infinite" sets, then the question over what the desired ordering of these computations is, comes into play.
If we have an equation like this:
y=lim(sqrt(pi*x)/x) as x->+0
Then we have irrational values embedded within a limit, but the irrational values themselves are (ill) defined as limits themselves.
So we can rewrite this abstractly to emphasize the multiple nested infinite structures like this:
y=lim(lim(sqrt(lim(pi)*x)/x)) as x->+0
In this case (if we ignore some potential problems with irrational numbers being defined in terms of inequalities only) we can still determine the convergence of the limit without major problems because the influence of irrational values is decreasing and we can, by increasing the number of iterations or precision of each inner approximation, find that y converges toward 0.
But we do not have the same ability to determine a specific convergence for a fourier transform:
fourier(y)=integral(f(x)*e^(-2*pi*i*y))dx (-infinity<x<infinity)
Here we once again have multiple infinite sequences involved, but they operate in parallel in determining the value of a non-monotonic function and are nested within an infinite structure that provides unlimited sensitivity to variations of these irrational numbers.
Notice that the typical response is to quickly rewrite this using Euler's Identity in terms of geometric operations on a circle (and I've shown the paradoxes that arise from this as well), but is this truly an identity? No.
Also consider that for physics, this process is rather backwards - circles are created from "e" processes being integrated into observations over time and so the description should of pi in terms of an integration of a frequency component of the "e" process.
Notice that if we create a complex unit of rotation:
e=lim((1+i/n)^n) as n->infinity
Consider that if this were to describe a circle (or similar object in some high dimensional space), then the radius should remain 1, but we find this is not the case:
|1+i/n|=1^2+(i^2)/n^2=1-1/(n^2)
And we can see that this spirals inward (though we could similarly have it spiral outward) instead of remaining on the perimeter of a circle. We can approximate for the magnitude of this change over iterations of the exponentiation by n by a simple rescaling to 1-1/n (n>>1).
Now as n->infinity, we have this magnitude approach 1 and less of a divergence occurs, but in the case of a fourier transform we have the influence of this integrated over an infinite number of cycles and sensitive to the irrational variances of pi as well as potentially unbounded values of y (frequency).
What we can step back and recognize that there is only one infinite value available - time and other infinite values must be created as smaller, but growing, structures of this.
So what's the correct ordering of these infinite values with respect to a physical process performing a fourier transform?
We can build these up from observations.
We first need at least 2 quantities changing over time.
I'll write it like this:
x(0)=1
y(0)=0
x(t)=x(t-1)+y(t-1)/n
y(t)=y(t-1)-x(t-1)/n
To give a quick example of this for small n, we have with n=1:
{1,0},{1,-1},{0,-2},{2,-2},{-4,0},{-4,4},{0,8},{8,8},{16,0},...
So we have a process that cycles every 8 phases and has a gain of 16:1 per cycle, or a mean increase in amplitude each cycle of 16^(1/8)=2^(4*1/8)=sqrt(2)~=1.4142
There are ways to make this better and generate some correlations with a fibonacci expansion instead of a binary splitting (the above algorithm is performing two divisions by n per unit of t and we could split this into a single division by n process which more closely resembles a single infinite process multiplexed between different operations).
But anyway, as we increase the value of n, which should be performed by increasing the values of x or increasing the times over which the process occurs, the growth of the magnitude of these spirals slows (in terms of ratios of changes of magnitude each cycle) and the time period over which a repetition occurs approaches 2*pi*n.
So for large n we can approximate the growth of these simple operations as:
x(t)~=cos(t/(2*pi*n))*(e^(a*t))
y(t)~=sin(t/(2*pi*n))*(e^(a*t))
Where a approaches 0 as n increases and I believe the quantity a*n should approach a non-zero constant, though I might have that wrong.
Anyway, the point is that the fourier transform is written in terms of a nested sequence of infinite operations that is likely not natural, though the approximate geometric identity of a rotation around a circle does not reveal the dynamic properties of the space and atoms in which this circle is dynamically constructed. So to begin with the constraint of a circle and then attempt to derive discrete units from this (as most continuous field theories attempt to do) is backwards. The circles arise from discrete operations over time and irrational numbers are not fundamental units as other processes would first be required to generate those before iterating those values in another infinite process, but the irrational process would never terminate and have time to allow its results to be interleaved with anything else.
So notice again, that a single simple discrete process of converting between two quantities generates a process that simultaineously creates irrational ratios for e, pi and trigonometric identities, and it we recognize that the irrational components are simply imagined to exist but intangible, then every present moment, t, can be associated with a precise value of pi, e and it's associated rotational vector all in a pair of numbers (which we could probably even find a way to encode as a single number or a growing string).
There's just one infinity and it's always finite when you look at it. Alephs are nothing more than studying the manner in which quantities are subdivided within that single process (or similarly the recurring elements and how the figurative DNA strand appears folded in space by these recurring properties).
If you still insist, you can have your non-materialized numbers, but if you can't even maintain a record of one of them in your thoughts, then you're still (irrationally) waiting for pi to end despite the fact that you've already imagined a circle intended to define it - what came first - a circle or pi? Don't tell me 3.1415... came first. Now, what came before the infinitely continuous circle or did we begin with circles? No, the center came first - and how did we determine a center - we had a collection - and what came first in the collection - an element and what came before an element? You can jump to conscious qualities and say you can't precisely describe it to me and I'm fine with that, though you can still maybe take it back further in an immaterial form, but you're on your own there, except if you tell me you were already finding numbers before that, we're just going to have to agree to disagree on that. You need a memory to construct numbers.