diff --git a/v1/providers/launchpad/instancetype.go b/v1/providers/launchpad/instancetype.go
index a66f941..7006114 100644
--- a/v1/providers/launchpad/instancetype.go
+++ b/v1/providers/launchpad/instancetype.go
@@ -242,7 +242,7 @@ func launchpadGpusToGpus(lpGpus []openapi.InstanceTypeGpu) []v1.GPU {
 			Manufacturer:   v1.GetManufacturer(gp.Manufacturer),
 			Count:          gp.Count,
 			Memory:         gbToBytes(gp.MemoryGb),
-			MemoryBytes:    v1.NewBytes(v1.BytesValue(int64(gp.MemoryGb)), v1.Gigabyte),
+			MemoryBytes:    launchpadGPUMemoryBytes(gp.MemoryGb),
 			NetworkDetails: string(gp.InterconnectionType),
 			Type:           strings.ToUpper(gp.Model),
 		}
@@ -347,7 +347,7 @@ func launchpadGputoGpu(node openapi.Node) *v1.GPU {
 	var lpGpuMemoryBytes v1.Bytes
 	if lpGpu.Memory != nil {
 		lpGpuMemory = gbToBytes(*lpGpu.Memory)
-		lpGpuMemoryBytes = v1.NewBytes(v1.BytesValue(*lpGpu.Memory), v1.Gigabyte)
+		lpGpuMemoryBytes = launchpadGPUMemoryBytes(*lpGpu.Memory)
 	}
 
 	gpu := &v1.GPU{
@@ -361,6 +361,11 @@ func launchpadGputoGpu(node openapi.Node) *v1.GPU {
 	return gpu
 }
 
+// Launchpad's GPU memory values are nominal decimal-GB capacities.
+func launchpadGPUMemoryBytes(memoryGB int32) v1.Bytes {
+	return v1.NewBytes(v1.BytesValue(memoryGB), v1.Gigabyte)
+}
+
 func launchpadNodeToSupportedStorage(node openapi.Node) []v1.Storage {
 	if len(node.Storage) == 0 {
 		return nil
diff --git a/v1/providers/launchpad/instancetype_test.go b/v1/providers/launchpad/instancetype_test.go
index 43822f3..51b460e 100644
--- a/v1/providers/launchpad/instancetype_test.go
+++ b/v1/providers/launchpad/instancetype_test.go
@@ -8,6 +8,7 @@ import (
 
 	"github.com/brevdev/cloud/internal/validation"
 	v1 "github.com/brevdev/cloud/v1"
+	openapi "github.com/brevdev/cloud/v1/providers/launchpad/gen/launchpad"
 )
 
 func TestGetInstanceTypes(t *testing.T) {
@@ -85,6 +86,15 @@ func TestInstanceTypeInfo(t *testing.T) {
 	}
 }
 
+func TestLaunchpadGpusToGpusUsesNominalGB(t *testing.T) {
+	t.Parallel()
+
+	gpus := launchpadGpusToGpus([]openapi.InstanceTypeGpu{{MemoryGb: 48}})
+
+	require.Len(t, gpus, 1)
+	require.Equal(t, v1.NewBytes(48, v1.Gigabyte), gpus[0].MemoryBytes)
+}
+
 func TestMakeGenericInstanceTypeID(t *testing.T) {
 	t.Parallel()
 
diff --git a/v1/providers/nebius/instancetype.go b/v1/providers/nebius/instancetype.go
index 07e9a48..4906403 100644
--- a/v1/providers/nebius/instancetype.go
+++ b/v1/providers/nebius/instancetype.go
@@ -209,7 +209,7 @@ func (c *NebiusClient) getInstanceTypesForLocation(ctx context.Context, platform
 					Name:         gpuName,
 					Manufacturer: v1.ManufacturerNVIDIA, // Nebius currently only supports NVIDIA GPUs
 					Memory:       memory,                // Populate VRAM based on GPU type
-					MemoryBytes:  v1.NewBytes(v1.BytesValue(int64(memory)/int64(units.Gibibyte)), v1.Gibibyte),
+					MemoryBytes:  gpuMemoryBytes(memory),
 				}
 				instanceType.SupportedGPUs = []v1.GPU{gpu}
 			}
@@ -615,6 +615,10 @@ func parseBlackwellGPUType(platformName string) string {
 	}
 }
 
+func gpuMemoryBytes(memory units.Base2Bytes) v1.Bytes {
+	return v1.NewBytes(v1.BytesValue(int64(memory)/int64(units.Gibibyte)), v1.Gigabyte)
+}
+
 // getGPUMemory returns the VRAM for a given GPU type in GiB
 func getGPUMemory(gpuType string) units.Base2Bytes {
 	// Static mapping of GPU types to their VRAM capacities
diff --git a/v1/providers/nebius/instancetype_test.go b/v1/providers/nebius/instancetype_test.go
index 4a7915f..e1d41cc 100644
--- a/v1/providers/nebius/instancetype_test.go
+++ b/v1/providers/nebius/instancetype_test.go
@@ -18,6 +18,14 @@ func TestNebiusPlatformArchitecture(t *testing.T) {
 	require.Equal(t, cloudv1.ArchitectureUnknown, nebiusPlatformArchitecture("future-platform"))
 }
 
+func TestNebiusGPUMemoryUsesNominalGB(t *testing.T) {
+	t.Parallel()
+
+	memory := getGPUMemory("L40S")
+
+	require.Equal(t, cloudv1.NewBytes(48, cloudv1.Gigabyte), gpuMemoryBytes(memory))
+}
+
 func TestApplyInstanceTypeFiltersUsesArchitectureMetadata(t *testing.T) {
 	t.Parallel()
 
diff --git a/v1/providers/sfcomputev2/instancetype.go b/v1/providers/sfcomputev2/instancetype.go
index ebe9d82..6feef4a 100644
--- a/v1/providers/sfcomputev2/instancetype.go
+++ b/v1/providers/sfcomputev2/instancetype.go
@@ -42,7 +42,7 @@ var h100InstanceTypeMetadata = func() sfcInstanceTypeMetadata {
 	m := sfcInstanceTypeMetadata{
 		diskBytes:       v1.NewBytes(1500, v1.Gigabyte),
 		memoryBytes:     v1.NewBytes(960, v1.Gigabyte),
-		gpuVRAM:         v1.NewBytes(80, v1.Gibibyte),
+		gpuVRAM:         v1.NewBytes(80, v1.Gigabyte),
 		vcpu:            sfcVCPU,
 		gpuCount:        sfcGPUCount,
 		gpuManufacturer: v1.ManufacturerNVIDIA,
diff --git a/v1/providers/sfcomputev2/instancetype_test.go b/v1/providers/sfcomputev2/instancetype_test.go
new file mode 100644
index 0000000..a6ecd30
--- /dev/null
+++ b/v1/providers/sfcomputev2/instancetype_test.go
@@ -0,0 +1,16 @@
+package v2
+
+import (
+	"testing"
+
+	v1 "github.com/brevdev/cloud/v1"
+	"github.com/stretchr/testify/require"
+)
+
+func TestH100InstanceTypeUsesNominalGBForVRAM(t *testing.T) {
+	t.Parallel()
+
+	instanceType := buildInstanceType(h100InstanceTypeMetadata, true)
+
+	require.Equal(t, v1.NewBytes(80, v1.Gigabyte), instanceType.SupportedGPUs[0].MemoryBytes)
+}