diff --git a/library/core/src/slice/mod.rs b/library/core/src/slice/mod.rs
index 5bb7243c4491b..058515efdfd74 100644
--- a/library/core/src/slice/mod.rs
+++ b/library/core/src/slice/mod.rs
@@ -8,7 +8,7 @@
 
 use crate::cmp::Ordering::{self, Equal, Greater, Less};
 use crate::intrinsics::{exact_div, unchecked_sub};
-use crate::mem::{self, SizedTypeProperties};
+use crate::mem::{self, MaybeUninit, SizedTypeProperties};
 use crate::num::NonZero;
 use crate::ops::{OneSidedRange, OneSidedRangeBound, Range, RangeBounds, RangeInclusive};
 use crate::panic::const_panic;
@@ -4589,7 +4589,7 @@ impl<T> [T] {
         // or generate worse code otherwise. This is also why we need to go
         // through a raw pointer here.
         let slice: *mut [T] = self;
-        let mut arr: mem::MaybeUninit<[&mut I::Output; N]> = mem::MaybeUninit::uninit();
+        let mut arr: MaybeUninit<[&mut I::Output; N]> = MaybeUninit::uninit();
         let arr_ptr = arr.as_mut_ptr();
 
         // SAFETY: We expect `indices` to contain disjunct values that are
@@ -4774,6 +4774,55 @@ impl<T> [T] {
     }
 }
 
+impl<T> [MaybeUninit<T>] {
+    /// Transmutes the mutable uninitialized slice to a mutable uninitialized slice of
+    /// another type, ensuring alignment of the types is maintained.
+    ///
+    /// This is a safe wrapper around [`slice::align_to_mut`], so inherits the same
+    /// guarantees as that method.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// #![feature(align_to_uninit_mut)]
+    /// use std::mem::MaybeUninit;
+    ///
+    /// pub struct BumpAllocator<'scope> {
+    ///     memory: &'scope mut [MaybeUninit<u8>],
+    /// }
+    ///
+    /// impl<'scope> BumpAllocator<'scope> {
+    ///     pub fn new(memory: &'scope mut [MaybeUninit<u8>]) -> Self {
+    ///         Self { memory }
+    ///     }
+    ///     pub fn try_alloc_uninit<T>(&mut self) -> Option<&'scope mut MaybeUninit<T>> {
+    ///         let first_end = self.memory.as_ptr().align_offset(align_of::<T>()) + size_of::<T>();
+    ///         let prefix = self.memory.split_off_mut(..first_end)?;
+    ///         Some(&mut prefix.align_to_uninit_mut::<T>().1[0])
+    ///     }
+    ///     pub fn try_alloc_u32(&mut self, value: u32) -> Option<&'scope mut u32> {
+    ///         let uninit = self.try_alloc_uninit()?;
+    ///         Some(uninit.write(value))
+    ///     }
+    /// }
+    ///
+    /// let mut memory = [MaybeUninit::<u8>::uninit(); 10];
+    /// let mut allocator = BumpAllocator::new(&mut memory);
+    /// let v = allocator.try_alloc_u32(42);
+    /// assert_eq!(v, Some(&mut 42));
+    /// ```
+    #[unstable(feature = "align_to_uninit_mut", issue = "139062")]
+    #[inline]
+    #[must_use]
+    pub fn align_to_uninit_mut<U>(&mut self) -> (&mut Self, &mut [MaybeUninit<U>], &mut Self) {
+        // SAFETY: `MaybeUninit` is transparent. Correct size and alignment are guaranteed by
+        // `align_to_mut` itself. Therefore the only thing that we have to ensure for a safe
+        // `transmute` is that the values are valid for the types involved. But for `MaybeUninit`
+        // any values are valid, so this operation is safe.
+        unsafe { self.align_to_mut() }
+    }
+}
+
 impl<T, const N: usize> [[T; N]] {
     /// Takes a `&[[T; N]]`, and flattens it to a `&[T]`.
     ///