diff --git a/.buildkite/pipeline.yml b/.buildkite/pipeline.yml
index 5392dd1f42..c033ed5efb 100755
--- a/.buildkite/pipeline.yml
+++ b/.buildkite/pipeline.yml
@@ -93,6 +93,10 @@ steps:
         key: data_opt_similar
         command: "julia --color=yes --check-bounds=yes --project=.buildkite test/DataLayouts/opt_similar.jl"
 
+      - label: "Unit: opt_universal_size"
+        key: opt_universal_size
+        command: "julia --color=yes --check-bounds=yes --project=.buildkite test/DataLayouts/opt_universal_size.jl"
+
       - label: "Unit: data_ndims"
         key: unit_data_ndims
         command: "julia --color=yes --check-bounds=yes --project=.buildkite test/DataLayouts/unit_ndims.jl"
diff --git a/ext/ClimaCoreCUDAExt.jl b/ext/ClimaCoreCUDAExt.jl
index 28cc705328..167696e93d 100644
--- a/ext/ClimaCoreCUDAExt.jl
+++ b/ext/ClimaCoreCUDAExt.jl
@@ -16,6 +16,7 @@ import ClimaCore.Utilities: half
 import ClimaCore.Utilities: cart_ind, linear_ind
 import ClimaCore.RecursiveApply:
     ⊠, ⊞, ⊟, radd, rmul, rsub, rdiv, rmap, rzero, rmin, rmax
+import ClimaCore.DataLayouts: get_N, get_Nv, get_Nij, get_Nij, get_Nh
 
 include(joinpath("cuda", "cuda_utils.jl"))
 include(joinpath("cuda", "data_layouts.jl"))
diff --git a/ext/cuda/data_layouts_copyto.jl b/ext/cuda/data_layouts_copyto.jl
index 3e53d05c46..5439a61527 100644
--- a/ext/cuda/data_layouts_copyto.jl
+++ b/ext/cuda/data_layouts_copyto.jl
@@ -84,11 +84,11 @@ function Base.copyto!(
 end
 
 import ClimaCore.DataLayouts: isascalar
-function knl_copyto_flat!(dest::AbstractData, bc)
+function knl_copyto_flat!(dest::AbstractData, bc, us)
     @inbounds begin
-        n = size(dest)
         tidx = thread_index()
-        if valid_range(tidx, prod(n))
+        if tidx ≤ get_N(us)
+            n = size(dest)
             I = kernel_indexes(tidx, n)
             dest[I] = bc[I]
         end
@@ -98,14 +98,15 @@ end
 
 function cuda_copyto!(dest::AbstractData, bc)
     (_, _, Nv, Nh) = DataLayouts.universal_size(dest)
+    us = DataLayouts.UniversalSize(dest)
     if Nv > 0 && Nh > 0
-        auto_launch!(knl_copyto_flat!, (dest, bc), dest; auto = true)
+        auto_launch!(knl_copyto_flat!, (dest, bc, us), dest; auto = true)
     end
     return dest
 end
 
 # TODO: can we use CUDA's luanch configuration for all data layouts?
-# Currently, it seems to have a slight performance degredation.
+# Currently, it seems to have a slight performance degradation.
 #! format: off
 # Base.copyto!(dest::IJFH{S, Nij},          bc::DataLayouts.BroadcastedUnionIJFH{S, Nij, Nh}, ::ToCUDA) where {S, Nij, Nh} = cuda_copyto!(dest, bc)
 Base.copyto!(dest::IFH{S, Ni, Nh},        bc::DataLayouts.BroadcastedUnionIFH{S, Ni, Nh}, ::ToCUDA) where {S, Ni, Nh} = cuda_copyto!(dest, bc)
diff --git a/ext/cuda/data_layouts_fill.jl b/ext/cuda/data_layouts_fill.jl
index f140250db2..9999c65a8a 100644
--- a/ext/cuda/data_layouts_fill.jl
+++ b/ext/cuda/data_layouts_fill.jl
@@ -1,8 +1,8 @@
-function knl_fill_flat!(dest::AbstractData, val)
+function knl_fill_flat!(dest::AbstractData, val, us)
     @inbounds begin
         tidx = thread_index()
-        n = size(dest)
-        if valid_range(tidx, prod(n))
+        if tidx ≤ get_N(us)
+            n = size(dest)
             I = kernel_indexes(tidx, n)
             @inbounds dest[I] = val
         end
@@ -12,8 +12,9 @@ end
 
 function cuda_fill!(dest::AbstractData, val)
     (_, _, Nv, Nh) = DataLayouts.universal_size(dest)
+    us = DataLayouts.UniversalSize(dest)
     if Nv > 0 && Nh > 0
-        auto_launch!(knl_fill_flat!, (dest, val), dest; auto = true)
+        auto_launch!(knl_fill_flat!, (dest, val, us), dest; auto = true)
     end
     return dest
 end
diff --git a/ext/cuda/operators_finite_difference.jl b/ext/cuda/operators_finite_difference.jl
index 1b790f1e16..980e5a813a 100644
--- a/ext/cuda/operators_finite_difference.jl
+++ b/ext/cuda/operators_finite_difference.jl
@@ -29,17 +29,11 @@ function Base.copyto!(
     (li, lw, rw, ri) = bounds = Operators.window_bounds(space, bc)
     Nv = ri - li + 1
     max_threads = 256
+    us = DataLayouts.UniversalSize(Fields.field_values(out))
     nitems = Nv * Nq * Nq * Nh # # of independent items
     (nthreads, nblocks) = _configure_threadblock(max_threads, nitems)
-    args = (
-        strip_space(out, space),
-        strip_space(bc, space),
-        axes(out),
-        bounds,
-        Val(Nv),
-        Val(Nq),
-        Nh,
-    )
+    args =
+        (strip_space(out, space), strip_space(bc, space), axes(out), bounds, us)
     auto_launch!(
         copyto_stencil_kernel!,
         args,
@@ -49,20 +43,16 @@ function Base.copyto!(
     )
     return out
 end
+import ClimaCore.DataLayouts: get_N, get_Nv, get_Nij, get_Nij, get_Nh
 
-function copyto_stencil_kernel!(
-    out,
-    bc,
-    space,
-    bds,
-    ::Val{Nv},
-    ::Val{Nq},
-    Nh,
-) where {Nv, Nq}
+function copyto_stencil_kernel!(out, bc, space, bds, us)
     @inbounds begin
         gid = threadIdx().x + (blockIdx().x - 1) * blockDim().x
-        if gid ≤ Nv * Nq * Nq * Nh
+        if gid ≤ get_N(us)
             (li, lw, rw, ri) = bds
+            Nv = get_Nv(us)
+            Nq = get_Nij(us)
+            Nh = get_Nh(us)
             (v, i, j, h) = cart_ind((Nv, Nq, Nq, Nh), gid).I
             hidx = (i, j, h)
             idx = v - 1 + li
diff --git a/src/DataLayouts/DataLayouts.jl b/src/DataLayouts/DataLayouts.jl
index 16c50a29db..c8f0cb0b16 100644
--- a/src/DataLayouts/DataLayouts.jl
+++ b/src/DataLayouts/DataLayouts.jl
@@ -76,6 +76,41 @@ corresponding to `UniversalSize`.
 """
 @inline universal_size(::UniversalSize{Ni, Nj, Nv, Nh}) where {Ni, Nj, Nv, Nh} =
     (Ni, Nj, Nv, Nh)
+
+"""
+    get_N(::AbstractData)
+    get_N(::UniversalSize)
+
+Statically returns `prod((Ni, Nj, Nv, Nh))`
+"""
+@inline get_N(::UniversalSize{Ni, Nj, Nv, Nh}) where {Ni, Nj, Nv, Nh} =
+    prod((Ni, Nj, Nv, Nh))
+
+@inline get_N(data::AbstractData) = get_N(UniversalSize(data))
+
+"""
+    get_Nv(::UniversalSize)
+
+Statically returns `Nv`.
+"""
+get_Nv(::UniversalSize{Ni, Nj, Nv}) where {Ni, Nj, Nv} = Nv
+
+"""
+    get_Nij(::UniversalSize)
+
+Statically returns `Nij`.
+"""
+get_Nij(::UniversalSize{Nij}) where {Nij} = Nij
+
+"""
+    get_Nh(::UniversalSize)
+
+Statically returns `Nh`.
+"""
+get_Nh(::UniversalSize{Ni, Nj, Nv, Nh}) where {Ni, Nj, Nv, Nh} = Nh
+
+get_Nh(data::AbstractData) = Nh
+
 @inline universal_size(data::AbstractData) = universal_size(UniversalSize(data))
 
 function Base.show(io::IO, data::AbstractData)
diff --git a/test/DataLayouts/opt_universal_size.jl b/test/DataLayouts/opt_universal_size.jl
new file mode 100644
index 0000000000..f54d5a6330
--- /dev/null
+++ b/test/DataLayouts/opt_universal_size.jl
@@ -0,0 +1,68 @@
+#=
+julia --project
+using Revise; include(joinpath("test", "DataLayouts", "opt_universal_size.jl"))
+=#
+using Test
+using ClimaCore.DataLayouts
+using ClimaCore: DataLayouts, Geometry
+import ClimaComms
+using StaticArrays: SMatrix
+ClimaComms.@import_required_backends
+using JET
+using InteractiveUtils: @code_typed
+
+function test_universal_size(data)
+    us = DataLayouts.UniversalSize(data)
+    # Make sure results is statically returned / constant propagated
+    ct = @code_typed DataLayouts.get_N(us)
+    @test ct.first.code[1] isa Core.ReturnNode
+    @test ct.first.code[end].val == DataLayouts.get_N(us)
+
+    ct = @code_typed DataLayouts.get_Nv(us)
+    @test ct.first.code[1] isa Core.ReturnNode
+    @test ct.first.code[end].val == DataLayouts.get_Nv(us)
+
+    ct = @code_typed DataLayouts.get_Nij(us)
+    @test ct.first.code[1] isa Core.ReturnNode
+    @test ct.first.code[end].val == DataLayouts.get_Nij(us)
+
+    ct = @code_typed DataLayouts.get_Nh(us)
+    @test ct.first.code[1] isa Core.ReturnNode
+    @test ct.first.code[end].val == DataLayouts.get_Nh(us)
+
+    ct = @code_typed size(data)
+    @test ct.first.code[1] isa Core.ReturnNode
+    @test ct.first.code[end].val == size(data)
+
+    ct = @code_typed DataLayouts.get_N(data)
+    @test ct.first.code[1] isa Core.ReturnNode
+    @test ct.first.code[end].val == DataLayouts.get_N(data)
+
+    # Demo of failed constant prop:
+    ct = @code_typed prod(size(data))
+    @test ct.first.code[1] isa Expr # first element is not a return node, but an expression
+end
+
+@testset "UniversalSize" begin
+    device = ClimaComms.device()
+    device_zeros(args...) = ClimaComms.array_type(device)(zeros(args...))
+    FT = Float64
+    S = FT
+    Nf = 1
+    Nv = 4
+    Nij = 3
+    Nh = 5
+    Nk = 6
+#! format: off
+    data = DataF{S}(device_zeros(FT,Nf));                        test_universal_size(data)
+    data = IJFH{S, Nij, Nh}(device_zeros(FT,Nij,Nij,Nf,Nh));     test_universal_size(data)
+    data = IFH{S, Nij, Nh}(device_zeros(FT,Nij,Nf,Nh));          test_universal_size(data)
+    data = IJF{S, Nij}(device_zeros(FT,Nij,Nij,Nf));             test_universal_size(data)
+    data = IF{S, Nij}(device_zeros(FT,Nij,Nf));                  test_universal_size(data)
+    data = VF{S, Nv}(device_zeros(FT,Nv,Nf));                    test_universal_size(data)
+    data = VIJFH{S,Nv,Nij,Nh}(device_zeros(FT,Nv,Nij,Nij,Nf,Nh));test_universal_size(data)
+    data = VIFH{S, Nv, Nij, Nh}(device_zeros(FT,Nv,Nij,Nf,Nh));  test_universal_size(data)
+#! format: on
+    # data = DataLayouts.IJKFVH{S, Nij, Nk, Nv, Nh}(device_zeros(FT,Nij,Nij,Nk,Nf,Nv,Nh)); test_universal_size(data) # TODO: test
+    # data = DataLayouts.IH1JH2{S, Nij}(device_zeros(FT,2*Nij,3*Nij));                     test_universal_size(data) # TODO: test
+end