diff --git a/torchx/pipelines/kfp/adapter.py b/torchx/pipelines/kfp/adapter.py
index efb439a7e..8da9e1c7a 100644
--- a/torchx/pipelines/kfp/adapter.py
+++ b/torchx/pipelines/kfp/adapter.py
@@ -7,7 +7,7 @@
 
 import copy
 import os
-from typing import Callable, Dict, List, Optional, Type
+from typing import Callable, Dict, List, Optional, Protocol, Tuple, Type
 
 import yaml
 from kfp import components, dsl
@@ -115,7 +115,7 @@ def output(self) -> dsl.PipelineParam:
         ...
 
 
-def component_spec_from_app(app: api.Application) -> str:
+def component_spec_from_app(app: api.Application) -> Tuple[str, api.Resource]:
     assert len(app.roles) == 1, f"KFP adapter only support one role, got {app.roles}"
 
     role = app.roles[0]
@@ -126,9 +126,6 @@ def component_spec_from_app(app: api.Application) -> str:
 
     container = role.container
     assert container.base_image is None, "KFP adapter does not support base_image"
-    assert (
-        container.resources == api.NULL_RESOURCE
-    ), "KFP adapter requires you to specify resources in the pipeline"
     assert len(container.port_map) == 0, "KFP adapter does not support port_map"
 
     command = [role.entrypoint, *role.args]
@@ -144,10 +141,36 @@ def component_spec_from_app(app: api.Application) -> str:
             }
         },
     }
-    return yaml.dump(spec)
+    return yaml.dump(spec), container.resources
 
 
-# pyre-fixme[24]: Generic type `Callable` expects 2 type parameters.
-def component_from_app(app: api.Application) -> Callable:
-    spec = component_spec_from_app(app)
-    return components.load_component_from_text(spec)
+class ContainerFactory(Protocol):
+    def __call__(self, *args: object, **kwargs: object) -> dsl.ContainerOp:
+        ...
+
+
+def component_from_app(app: api.Application) -> ContainerFactory:
+    resources: api.Resource
+    spec, resources = component_spec_from_app(app)
+
+    assert (
+        len(resources.capabilities) == 0
+    ), f"KFP doesn't support capabilities, got {resources.capabilities}"
+    component_factory: ContainerFactory = components.load_component_from_text(spec)
+
+    def factory_wrapper(*args: object, **kwargs: object) -> dsl.ContainerOp:
+        c = component_factory(*args, **kwargs)
+        container = c.container
+        if (cpu := resources.cpu) >= 0:
+            cpu_str = f"{int(cpu*1000)}m"
+            container.set_cpu_request(cpu_str)
+            container.set_cpu_limit(cpu_str)
+        if (mem := resources.memMB) >= 0:
+            mem_str = f"{int(mem)}M"
+            container.set_memory_request(mem_str)
+            container.set_memory_limit(mem_str)
+        if (gpu := resources.gpu) >= 0:
+            container.set_gpu_limit(str(gpu))
+        return c
+
+    return factory_wrapper
diff --git a/torchx/pipelines/kfp/test/adapter_test.py b/torchx/pipelines/kfp/test/adapter_test.py
index 42909b7bf..4e2180d87 100644
--- a/torchx/pipelines/kfp/test/adapter_test.py
+++ b/torchx/pipelines/kfp/test/adapter_test.py
@@ -11,6 +11,7 @@
 from typing import Callable, Optional, TypedDict
 
 from kfp import compiler, components, dsl
+from kubernetes.client.models import V1ResourceRequirements
 from torchx.apps.io.copy import Copy
 from torchx.pipelines.kfp.adapter import (
     TorchXComponent,
@@ -117,7 +118,14 @@ class KFPSpecsTest(unittest.TestCase):
     """
 
     def _test_app(self) -> api.Application:
-        container = api.Container(image="pytorch/torchx:latest")
+        container = api.Container(
+            image="pytorch/torchx:latest",
+            resources=api.Resource(
+                cpu=2,
+                memMB=3000,
+                gpu=4,
+            ),
+        )
         trainer_role = (
             api.Role(name="trainer")
             .runs(
@@ -135,8 +143,9 @@ def _test_app(self) -> api.Application:
     def test_component_spec_from_app(self) -> None:
         app = self._test_app()
 
-        spec = component_spec_from_app(app)
+        spec, resources = component_spec_from_app(app)
         self.assertIsNotNone(components.load_component_from_text(spec))
+        self.assertEqual(resources, app.roles[0].container.resources)
         self.assertEqual(
             spec,
             """description: KFP wrapper for TorchX component test, role trainer
@@ -160,6 +169,22 @@ def test_pipeline(self) -> None:
 
         def pipeline() -> dsl.PipelineParam:
             a = kfp_copy()
+            resources: V1ResourceRequirements = a.container.resources
+            self.assertEqual(
+                resources,
+                V1ResourceRequirements(
+                    limits={
+                        "cpu": "2000m",
+                        "memory": "3000M",
+                        "nvidia.com/gpu": "4",
+                    },
+                    requests={
+                        "cpu": "2000m",
+                        "memory": "3000M",
+                    },
+                ),
+            )
+
             b = kfp_copy()
             b.after(a)
             return b