diff --git a/recipes/fvdb-core/build.sh b/recipes/fvdb-core/build.sh
new file mode 100644
index 0000000000000..4322e6d63a03f
--- /dev/null
+++ b/recipes/fvdb-core/build.sh
@@ -0,0 +1,79 @@
+#!/bin/bash
+
+setup_parallel_build_jobs() {
+  # Calculate the optimal number of parallel build jobs based on available RAM
+  RAM_GB=$(free -g | awk '/^Mem:/{print $7}')
+  if [ -z "$RAM_GB" ]; then
+      echo "Error: Unable to determine available RAM"
+      exit 1
+  fi
+  JOB_RAM_GB=3
+
+  # Get number of processors
+  NPROC=$(nproc)
+
+  # count the number of ';' in the TORCH_CUDA_ARCH_LIST
+  NUM_ARCH=$(echo "$TORCH_CUDA_ARCH_LIST" | tr ';' '\n' | wc -l)
+  if [ "$NUM_ARCH" -lt 1 ]; then
+    NUM_ARCH=1
+  fi
+  NVCC_THREADS=$NUM_ARCH
+
+  # Check if we have enough RAM for even one job with full NVCC_THREADS
+  # Requirement: JOB_RAM_GB * NVCC_THREADS
+  MIN_RAM_REQUIRED=$((JOB_RAM_GB * NVCC_THREADS))
+
+  if [ "$RAM_GB" -lt "$MIN_RAM_REQUIRED" ]; then
+      NVCC_THREADS=1
+  fi
+
+  # Limit NVCC_THREADS to NPROC to ensure we don't oversubscribe
+  if [ "$NVCC_THREADS" -gt "$NPROC" ]; then
+      NVCC_THREADS=$NPROC
+  fi
+
+  # Determine max jobs based on CPU:
+  # We want CMAKE_BUILD_PARALLEL_LEVEL * NVCC_THREADS <= NPROC
+  MAX_JOBS_CPU=$((NPROC / NVCC_THREADS))
+
+  # Determine max jobs based on RAM:
+  # Assume each job requires JOB_RAM_GB * NVCC_THREADS
+  MAX_JOBS_RAM=$((RAM_GB / (JOB_RAM_GB * NVCC_THREADS)))
+
+  # Take the minimum
+  PARALLEL_JOBS=$((MAX_JOBS_CPU < MAX_JOBS_RAM ? MAX_JOBS_CPU : MAX_JOBS_RAM))
+
+  # Ensure at least 1 job
+  if [ "$PARALLEL_JOBS" -lt 1 ]; then
+    PARALLEL_JOBS=1
+  fi
+
+  # if CMAKE_BUILD_PARALLEL_LEVEL is set, use that
+  if [ -n "$CMAKE_BUILD_PARALLEL_LEVEL" ]; then
+    echo "Using CMAKE_BUILD_PARALLEL_LEVEL=$CMAKE_BUILD_PARALLEL_LEVEL"
+  else
+
+    CMAKE_BUILD_PARALLEL_LEVEL=$PARALLEL_JOBS
+
+    echo "Setting nvcc --threads to $NVCC_THREADS based on the number of CUDA architectures ($NUM_ARCH)"
+    echo "Setting CMAKE_BUILD_PARALLEL_LEVEL to $CMAKE_BUILD_PARALLEL_LEVEL"
+    echo "  Constraint: Total Threads ($((CMAKE_BUILD_PARALLEL_LEVEL * NVCC_THREADS))) <= NPROC ($NPROC)"
+    echo "  Constraint: Estimated RAM ($((CMAKE_BUILD_PARALLEL_LEVEL * NVCC_THREADS * JOB_RAM_GB))) GB <= Available RAM ($RAM_GB GB)"
+
+    export CMAKE_BUILD_PARALLEL_LEVEL
+    export NVCC_THREADS
+  fi
+}
+
+
+setup_parallel_build_jobs
+export CMAKE_GENERATOR=Ninja
+# GCC 14 false positive: -Wstringop-overflow in NanoVDB headers with deep template inlining at -O3
+# See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118817
+export CXXFLAGS="${CXXFLAGS} -Wno-error=stringop-overflow"
+$PYTHON -m pip install \
+    --no-deps \
+    --no-build-isolation \
+    -vv \
+    -C 'skbuild.ninja.make-fallback=false' \
+    .
diff --git a/recipes/fvdb-core/meta.yaml b/recipes/fvdb-core/meta.yaml
new file mode 100644
index 0000000000000..a274d26d71792
--- /dev/null
+++ b/recipes/fvdb-core/meta.yaml
@@ -0,0 +1,88 @@
+{% set name = "fvdb-core" %}
+{% set version = "0.3.0" %}
+{% set torch_proc_type = "cuda" if cuda_compiler_version != "None" else "cpu" %}
+
+package:
+  name: {{ name }}
+  version: {{ version }}
+
+source:
+  url: https://pypi.org/packages/source/{{ name[0] }}/{{ name }}/fvdb_core-{{ version }}.tar.gz
+  sha256: f6ec0d5e6a2b601632720ed0be8aa7c8ba6f2d9bd63fac477cef2cfd088b5f16
+
+build:
+  skip: true  # [cuda_compiler_version == "None" or win or osx or py!=312]
+  number: 0
+  script_env:
+    - TORCH_CUDA_ARCH_LIST=7.5;8.0;9.0;10.0;12.0+PTX
+
+
+requirements:
+  build:
+    - python                                 # [build_platform != target_platform]
+    - cross-python_{{ target_platform }}     # [build_platform != target_platform]
+    - {{ stdlib('c') }}
+    - {{ compiler('c') }}
+    - {{ compiler('cxx') }}
+    - {{ compiler('cuda') }}
+    - cuda-version {{ cuda_compiler_version }}
+    - cuda-command-line-tools
+    - cmake >=3.25
+    - ninja
+    - pkg-config
+    - git
+    - procps-ng  # provides 'free'
+    - coreutils  # provides 'nproc', 'awk', etc.
+  host:
+    - python
+    - scikit-build-core
+    - pip
+    - pytorch
+    - pytorch =*=*{{ torch_proc_type }}*
+    - numpy
+    - pybind11 >=2.13
+    - gitpython
+    - cuda-version {{ cuda_compiler_version }}
+    - cuda-cudart-dev
+    - cuda-nvrtc-dev
+    - cuda-nvtx-dev
+    - libcublas-dev
+    - libcudnn-dev
+    - libcusolver-dev
+    - libcusparse-dev
+    - libpng
+    - zlib
+  run:
+    - python
+    - pytorch
+    - pytorch =*=*{{ torch_proc_type }}*
+    - {{ pin_compatible('numpy') }}
+    - cuda-version {{ cuda_compiler_version }}
+    - gitpython
+    - git
+    - parameterized
+
+test:
+  commands:
+    - pip check
+    - python run_test.py
+  source_files:
+    - tests/
+  requires:
+    - pip
+    - pytest
+    - parameterized
+    - git
+
+about:
+  home: https://www.openvdb.org/
+  license: Apache-2.0
+  license_family: Apache
+  license_file: LICENSE
+  summary: "A deep learning framework for sparse, large-scale, high-performance spatial intelligence"
+  doc_url: https://fvdb.ai/
+  dev_url: https://github.com/openvdb/fvdb-core
+
+extra:
+  recipe-maintainers:
+    - swahtz
diff --git a/recipes/fvdb-core/run_test.py b/recipes/fvdb-core/run_test.py
new file mode 100644
index 0000000000000..14ae0831c35e8
--- /dev/null
+++ b/recipes/fvdb-core/run_test.py
@@ -0,0 +1,32 @@
+import sys
+
+# Check for fvdb (without importing)
+import pkgutil
+pkgutil.find_loader("fvdb")
+
+# Try to import fvdb
+try:
+    import fvdb
+except ImportError:
+    print("Failed to import fvdb. Exiting without running tests.")
+    sys.exit(0)
+
+print(f"fvdb {fvdb.__version__} imported successfully")
+
+# Check for GPU availability
+import torch
+
+if not torch.cuda.is_available():
+    print("No CUDA GPU available. Exiting without running fvdb's tests.")
+    sys.exit(0)
+
+print(f"CUDA is available: {torch.cuda.get_device_name(0)}")
+
+# Run fvdb's test suite
+# Ignore tests with unavailable optional dependencies (OpenImageIO, point_cloud_utils, torch_scatter)
+import pytest
+sys.exit(pytest.main([
+    "tests/unit", "-v", "--tb=short",
+    "--ignore=tests/unit/test_gsplat.py",
+    "--ignore=tests/unit/test_jagged_tensor.py",
+]))