diff --git a/packages/typegpu/src/data/index.ts b/packages/typegpu/src/data/index.ts
index 9aeb7edc16..5eebe77415 100644
--- a/packages/typegpu/src/data/index.ts
+++ b/packages/typegpu/src/data/index.ts
@@ -48,11 +48,6 @@ assignInfixOperator(MatBase, 'mul', Operator.star);
   };
 }
 
-// TODO: Remove this side-effect once we have shaderbits
-import { f32 } from './numeric.ts';
-// oxlint-disable-next-line typescript/no-explicit-any
-(globalThis as any).__TYPEGPU_F32__ = f32;
-
 export { bool, f16, f32, i32, u16, u32 } from './numeric.ts';
 export {
   isAlignAttrib,
diff --git a/packages/typegpu/src/data/numeric.ts b/packages/typegpu/src/data/numeric.ts
index d49368843b..85fbc2bb4e 100644
--- a/packages/typegpu/src/data/numeric.ts
+++ b/packages/typegpu/src/data/numeric.ts
@@ -2,22 +2,6 @@ import { $internal } from '../shared/symbols.ts';
 import type { AbstractFloat, AbstractInt, Bool, F16, F32, I32, U16, U32 } from './wgslTypes.ts';
 import { callableSchema } from '../core/function/createCallableSchema.ts';
 
-export const abstractInt = {
-  [$internal]: {},
-  type: 'abstractInt',
-  toString() {
-    return 'abstractInt';
-  },
-} as AbstractInt;
-
-export const abstractFloat = {
-  [$internal]: {},
-  type: 'abstractFloat',
-  toString() {
-    return 'abstractFloat';
-  },
-} as AbstractFloat;
-
 const boolCast = callableSchema({
   name: 'bool',
   schema: () => bool,
@@ -299,3 +283,21 @@ export const f16: F16 = Object.assign(f16Cast, {
   [$internal]: {},
   type: 'f16',
 }) as unknown as F16;
+
+export const abstractInt = {
+  [$internal]: {},
+  type: 'abstractInt',
+  toString() {
+    return 'abstractInt';
+  },
+  concretized: i32,
+} as AbstractInt;
+
+export const abstractFloat = {
+  [$internal]: {},
+  type: 'abstractFloat',
+  toString() {
+    return 'abstractFloat';
+  },
+  concretized: f32,
+} as AbstractFloat;
diff --git a/packages/typegpu/src/data/wgslTypes.ts b/packages/typegpu/src/data/wgslTypes.ts
index dc85031925..f8757d90b5 100644
--- a/packages/typegpu/src/data/wgslTypes.ts
+++ b/packages/typegpu/src/data/wgslTypes.ts
@@ -129,6 +129,7 @@ export interface matInfixNotation<T extends matBase> {
  */
 export interface AbstractInt extends BaseData {
   readonly type: 'abstractInt';
+  readonly concretized: I32;
   // Type-tokens, not available at runtime
   readonly [$repr]: number;
   readonly [$invalidSchemaReason]: 'Abstract numerics are not host-shareable';
@@ -140,6 +141,7 @@ export interface AbstractInt extends BaseData {
  */
 export interface AbstractFloat extends BaseData {
   readonly type: 'abstractFloat';
+  readonly concretized: F32;
   // Type-tokens, not available at runtime
   readonly [$repr]: number;
   readonly [$invalidSchemaReason]: 'Abstract numerics are not host-shareable';
diff --git a/packages/typegpu/src/tgsl/conversion.ts b/packages/typegpu/src/tgsl/conversion.ts
index 7c83aec720..f410e53e6d 100644
--- a/packages/typegpu/src/tgsl/conversion.ts
+++ b/packages/typegpu/src/tgsl/conversion.ts
@@ -5,6 +5,7 @@ import { derefSnippet, RefOperator } from '../data/ref.ts';
 import { schemaCallWrapperGPU } from '../data/schemaCallWrapper.ts';
 import { snip, type Snippet } from '../data/snippet.ts';
 import {
+  type AbstractFloat,
   type AnyWgslData,
   type BaseData,
   type F16,
@@ -123,10 +124,7 @@ function getImplicitConversionRank(src: BaseData, dest: BaseData): ConversionRan
   if ((trueSrc.type === 'u32' || trueSrc.type === 'i32') && trueDst.type === 'abstractFloat') {
     // When one of the types is a float (abstract or not), we don't want to cast it to a non-float type,
     // which would cause it to lose precision. We instead choose the common type to be f32.
-
-    // TODO: Remove this side-effect once we have shaderbits
-    // oxlint-disable-next-line typescript/no-explicit-any
-    return { rank: 1, action: 'cast', targetType: (globalThis as any).__TYPEGPU_F32__ };
+    return { rank: 1, action: 'cast', targetType: (trueDst as AbstractFloat).concretized };
   }
 
   if (trueSrc.type === 'abstractFloat') {