diff --git a/Cargo.lock b/Cargo.lock
index be8e47dbd..0acc5fcc5 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -5906,7 +5906,6 @@ dependencies = [
  "clap",
  "either",
  "fs-err 3.1.1",
- "regex",
  "schemars",
  "serde",
  "thiserror 2.0.12",
diff --git a/crates/uv-torch/Cargo.toml b/crates/uv-torch/Cargo.toml
index 2190598e7..fdaa4653e 100644
--- a/crates/uv-torch/Cargo.toml
+++ b/crates/uv-torch/Cargo.toml
@@ -19,7 +19,6 @@ uv-static = { workspace = true }
 clap = { workspace = true, optional = true }
 either = { workspace = true }
 fs-err = { workspace = true }
-regex = { workspace = true }
 schemars = { workspace = true, optional = true }
 serde = { workspace = true }
 thiserror = { workspace = true }
diff --git a/crates/uv-torch/src/accelerator.rs b/crates/uv-torch/src/accelerator.rs
index 1ae4fa07e..0bcdca246 100644
--- a/crates/uv-torch/src/accelerator.rs
+++ b/crates/uv-torch/src/accelerator.rs
@@ -1,7 +1,6 @@
 use std::path::Path;
 use std::str::FromStr;
 
-use regex::Regex;
 use tracing::debug;
 use walkdir::WalkDir;
 
@@ -64,7 +63,6 @@ impl Accelerator {
     /// 5. `nvidia-smi --query-gpu=driver_version --format=csv,noheader`.
     /// 6. `rocm_agent_enumerator`, which lists the AMD GPU architectures.
     /// 7. `/sys/bus/pci/devices`, filtering for the Intel GPU via PCI.
-    /// 8. `powershell` command querying `Win32_VideoController` to detect Intel GPU.
     pub fn detect() -> Result<Option<Self>, AcceleratorError> {
         // Constants used for PCI device detection
         const PCI_BASE_CLASS_MASK: u32 = 0x00ff_0000;
@@ -202,33 +200,6 @@ impl Accelerator {
             }
         }
 
-        // Query Intel GPU by `powershell` command via `Win32_VideoController`.
-        //
-        // See: https://learn.microsoft.com/en-us/windows/win32/cimwin32prov/win32-videocontroller
-        if let Ok(output) = std::process::Command::new("powershell")
-            .arg("-Command")
-            .arg("Get-WmiObject Win32_VideoController | Select-Object -ExpandProperty PNPDeviceID")
-            .output()
-        {
-            if output.status.success() {
-                let stdout = String::from_utf8(output.stdout)?;
-                let re = Regex::new(r"VEN_([0-9A-Fa-f]{4})").unwrap();
-                for caps in re.captures_iter(&stdout) {
-                    let vendor = u32::from_str_radix(&caps[1], 16)?;
-                    if vendor == PCI_VENDOR_ID_INTEL {
-                        debug!("Detected Intel GPU from PCI, vendor=0x{:04x}", vendor);
-                        return Ok(Some(Self::Xpu));
-                    }
-                }
-            } else {
-                debug!(
-                    "Failed to query Intel GPU with powershell with status `{}`: {}",
-                    output.status,
-                    String::from_utf8_lossy(&output.stderr)
-                );
-            }
-        }
-
         debug!("Failed to detect GPU driver version");
 
         Ok(None)
diff --git a/crates/uv-torch/src/backend.rs b/crates/uv-torch/src/backend.rs
index 646fdb7f3..5ad71b385 100644
--- a/crates/uv-torch/src/backend.rs
+++ b/crates/uv-torch/src/backend.rs
@@ -351,10 +351,11 @@ impl TorchStrategy {
                 }
             },
             TorchStrategy::Xpu { os } => match os {
-                Os::Manylinux { .. } | Os::Windows => Either::Right(Either::Right(Either::Left(
+                Os::Manylinux { .. } => Either::Right(Either::Right(Either::Left(
                     std::iter::once(TorchBackend::Xpu.index_url()),
                 ))),
-                Os::Musllinux { .. }
+                Os::Windows
+                | Os::Musllinux { .. }
                 | Os::Macos { .. }
                 | Os::FreeBsd { .. }
                 | Os::NetBsd { .. }
diff --git a/docs/guides/integration/pytorch.md b/docs/guides/integration/pytorch.md
index 880293a72..f060cbb6b 100644
--- a/docs/guides/integration/pytorch.md
+++ b/docs/guides/integration/pytorch.md
@@ -460,4 +460,12 @@ $ # With an environment variable.
 $ UV_TORCH_BACKEND=cu126 uv pip install torch torchvision
 ```
 
+On Windows, Intel GPU (XPU) is not automatically selected with `--torch-backend=auto`, but you can
+manually specify it using `--torch-backend=xpu`:
+
+```shell
+$ # Manual selection for Intel GPU.
+$ uv pip install torch torchvision --torch-backend=xpu
+```
+
 At present, `--torch-backend` is only available in the `uv pip` interface.