diff --git a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.cpp b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.cpp
index 90ecc7f87f..a16c00b9d4 100644
--- a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.cpp
+++ b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.cpp
@@ -946,6 +946,13 @@ namespace Dml
         assert(!m_closed);
         m_context->Flush();
     }
+    
+    void __stdcall ExecutionProviderImpl::FlushAndSync() const
+    {
+        Flush();
+        m_context->GetCurrentCompletionEvent().WaitForSignal();
+        m_context->ReleaseCompletedReferences();
+    }
 
     void ExecutionProviderImpl::SetDefaultRoundingMode(AllocatorRoundingMode roundingMode)
     {
diff --git a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.h b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.h
index b9ac772095..8881347936 100644
--- a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.h
+++ b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/ExecutionProvider.h
@@ -125,6 +125,7 @@ namespace Dml
 
         STDMETHOD_(D3D12_COMMAND_LIST_TYPE, GetCommandListTypeForQueue)() const override;
         STDMETHOD_(void, Flush)() const override;
+        STDMETHOD_(void, FlushAndSync)() const override;
 
         void SetDefaultRoundingMode(AllocatorRoundingMode roundingMode);
 
diff --git a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/FusedGraphKernel.cpp b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/FusedGraphKernel.cpp
index b7f24d49d1..4f72d1dc12 100644
--- a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/FusedGraphKernel.cpp
+++ b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/FusedGraphKernel.cpp
@@ -72,6 +72,9 @@ namespace Dml
                 m_persistentResourceBinding = DML_BUFFER_BINDING { m_persistentResource.Get(), 0, persistentResourceSize };
             }
 
+            // For temporary prototyping, allow weight upload to complete before issuing initialization work which reads those weights
+            m_provider->FlushAndSync();
+
             ORT_THROW_IF_FAILED(m_provider->InitializeOperator(
                 m_compiledExecutionPlanOperator.Get(),
                 m_persistentResourceBinding ? &*m_persistentResourceBinding : nullptr,
diff --git a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/IExecutionProvider.h b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/IExecutionProvider.h
index d7a0a607cd..f5556ae2ab 100644
--- a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/IExecutionProvider.h
+++ b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/IExecutionProvider.h
@@ -64,6 +64,7 @@ namespace Dml
 
         STDMETHOD_(D3D12_COMMAND_LIST_TYPE, GetCommandListTypeForQueue)() const noexcept = 0;
         STDMETHOD_(void, Flush)() const noexcept = 0;
+        STDMETHOD_(void, FlushAndSync)() const noexcept = 0;
 
         STDMETHOD_(ID3D12Resource*, DecodeResource)(void* allocation) const noexcept = 0;
         STDMETHOD(AllocatePooledResource(size_t size, AllocatorRoundingMode roundingMode, ID3D12Resource **d3dResource, IUnknown* *pooledResource)) const noexcept = 0;