diff --git a/cpp/src/arrow/compute/kernels/vector_hash.cc b/cpp/src/arrow/compute/kernels/vector_hash.cc
index 9a29dcf333f..57061cf413f 100644
--- a/cpp/src/arrow/compute/kernels/vector_hash.cc
+++ b/cpp/src/arrow/compute/kernels/vector_hash.cc
@@ -445,9 +445,11 @@ class NullHashKernel : public HashKernel {
 
 class DictionaryHashKernel : public HashKernel {
  public:
-  explicit DictionaryHashKernel(std::unique_ptr<HashKernel> indices_kernel,
-                                std::shared_ptr<DataType> dictionary_value_type)
+  DictionaryHashKernel(std::unique_ptr<HashKernel> indices_kernel,
+                       std::shared_ptr<DataType> index_type,
+                       std::shared_ptr<DataType> dictionary_value_type)
       : indices_kernel_(std::move(indices_kernel)),
+        index_type_(std::move(index_type)),
         dictionary_value_type_(std::move(dictionary_value_type)) {}
 
   Status Reset() override { return indices_kernel_->Reset(); }
@@ -513,15 +515,25 @@ class DictionaryHashKernel : public HashKernel {
       return first_dictionary_;
     }
 
-    auto out_dict_type = first_dictionary_->type();
+    std::shared_ptr<DataType> out_dict_type;
     std::shared_ptr<Array> out_dict;
     RETURN_NOT_OK(dictionary_unifier_->GetResult(&out_dict_type, &out_dict));
+    // The output keeps the input index type, so the highest index into the unified
+    // dictionary (its length minus one) must fit in it
+    const Int64Scalar max_index(out_dict->length() - 1);
+    if (out_dict->length() > 0 &&
+        !::arrow::internal::IntegersCanFit(max_index, *index_type_).ok()) {
+      return Status::Invalid(
+          "These dictionaries cannot be combined.  The unified dictionary requires a "
+          "larger index type.");
+    }
     return out_dict;
   }
 
  private:
   std::unique_ptr<HashKernel> indices_kernel_;
   std::shared_ptr<Array> first_dictionary_;
+  std::shared_ptr<DataType> index_type_;
   std::shared_ptr<DataType> dictionary_value_type_;
   std::unique_ptr<DictionaryUnifier> dictionary_unifier_;
 };
@@ -597,7 +609,7 @@ Result<std::unique_ptr<KernelState>> DictionaryHashInit(KernelContext* ctx,
   ARROW_ASSIGN_OR_RAISE(auto indices_hasher,
                         GetHashInit<Action>(dict_type.index_type()->id())(ctx, args));
   return std::make_unique<DictionaryHashKernel>(
-      checked_pointer_cast<HashKernel>(std::move(indices_hasher)),
+      checked_pointer_cast<HashKernel>(std::move(indices_hasher)), dict_type.index_type(),
       dict_type.value_type());
 }
 
diff --git a/cpp/src/arrow/compute/kernels/vector_hash_test.cc b/cpp/src/arrow/compute/kernels/vector_hash_test.cc
index b11e0a722a5..8bd907679f4 100644
--- a/cpp/src/arrow/compute/kernels/vector_hash_test.cc
+++ b/cpp/src/arrow/compute/kernels/vector_hash_test.cc
@@ -21,6 +21,7 @@
 #include <functional>
 #include <locale>
 #include <memory>
+#include <numeric>
 #include <stdexcept>
 #include <string>
 #include <utility>
@@ -764,6 +765,35 @@ TEST_F(TestHashKernel, DictionaryUniqueAndValueCounts) {
   }
 }
 
+TEST_F(TestHashKernel, DictionaryUniqueAndValueCountsIndexOverflow) {
+  // GH-45010: unifying the chunk dictionaries can produce more entries than the
+  // index type can address
+  auto dict_ty = dictionary(int8(), int64());
+  auto make_chunk = [&](int64_t first_value, int64_t dict_length) {
+    std::vector<int64_t> values(dict_length);
+    std::iota(values.begin(), values.end(), first_value);
+    std::shared_ptr<Array> dict;
+    ArrayFromVector<Int64Type, int64_t>(values, &dict);
+    auto indices = ArrayFromJSON(int8(), "[" + std::to_string(dict_length - 1) + "]");
+    return std::make_shared<DictionaryArray>(dict_ty, indices, dict);
+  };
+
+  // 64 + 64 = 128 unified entries still fit in int8: the highest index is 127
+  auto fits = *ChunkedArray::Make({make_chunk(0, 64), make_chunk(100, 64)}, dict_ty);
+  ASSERT_OK_AND_ASSIGN(auto uniques, Unique(fits));
+  ASSERT_OK(uniques->ValidateFull());
+  ASSERT_EQ(checked_cast<const DictionaryArray&>(*uniques).dictionary()->length(), 128);
+  ASSERT_OK_AND_ASSIGN(auto counts, ValueCounts(fits));
+  ASSERT_OK(counts->ValidateFull());
+
+  // 64 + 65 = 129 unified entries do not
+  auto overflows = *ChunkedArray::Make({make_chunk(0, 64), make_chunk(100, 65)}, dict_ty);
+  EXPECT_RAISES_WITH_MESSAGE_THAT(Invalid, ::testing::HasSubstr("larger index type"),
+                                  Unique(overflows));
+  EXPECT_RAISES_WITH_MESSAGE_THAT(Invalid, ::testing::HasSubstr("larger index type"),
+                                  ValueCounts(overflows));
+}
+
 /* TODO(ARROW-4124): Determine if we want to do something that is reproducible with
  * floats.
 TEST_F(TestHashKernel, ValueCountsFloat) {