Skip to content

Blog post: The Open Source Infrastructure Crisis Nobody Talks About: conda-forge Needs Help #1

@khuyentran1401

Description

@khuyentran1401

Target trend: conda-forge sustainability crisis: GPU build server decommissioned, CZI grant ended, companies that depend on conda-forge not contributing back.

Background:
conda-forge is a community-run service that builds and hosts tens of thousands of binary packages for the scientific Python ecosystem. When someone runs conda install pytorch or pixi add tensorflow, those packages come from conda-forge.

conda-forge relied on a dedicated GPU CI server (6x NVIDIA Tesla V100 GPUs, 24-core AMD EPYC with 48 threads, ~500GB RAM) provided by Quansight via the open-gpu-server project. This server was essential for building and testing GPU-enabled packages like PyTorch, TensorFlow, CUDA libraries, cuDNN, etc. The server was decommissioned on 2026-03-13 and the service is no longer available, meaning conda-forge has lost the ability to reliably build and test GPU packages against real hardware.

The CZI (Chan Zuckerberg Initiative) EOSS 5 grant that funded conda-forge infrastructure work from December 2022 to January 2025 has also ended, leaving a gap in maintenance and operational funding. conda-forge is volunteer-maintained with no permanent paid staff (though some contributors were funded through grants like CZI, NumFOCUS, and GSoC). The project now needs companies that depend on it to contribute infrastructure or funding to keep the ecosystem healthy.

Outline: The Business Case

Tone: Pragmatic, ROI-focused. "This isn't charity. It's risk management."

  1. The hidden dependency in your stack

    • If your team uses conda/pixi/Nebi, you depend on conda-forge
    • Tens of thousands of packages, zero SLA, no permanent paid staff
    • Companies like Quansight, prefix.dev, Anaconda, QuantStack, and OVH have stepped up with infrastructure and funding. But the list of contributors is far shorter than the list of users.
  2. Risk assessment: what's at stake

    • GPU build server decommissioned on 2026-03-13
    • CVE response time: slowing down
    • Supply chain integrity: dependent on volunteer capacity
  3. The cost of doing nothing

    • Teams waste hours on broken GPU installs
    • Security vulnerabilities go unpatched
    • Reproducibility breaks across environments
  4. The cost of fixing it: surprisingly low

    • A GPU build server: ~$1K/mo
    • Maintenance funding: fraction of what companies spend on cloud bills
    • Compare to cost of one engineer debugging GPU install issues for a week
  5. How to contribute: practical next steps

    • Open Collective sponsorship
    • Dedicated infrastructure donations
    • Cloud credit programs

Target audience: Engineering leaders at companies using conda-forge (ML teams, quant funds, pharma, research orgs), open source program offices, cloud providers.

Metadata

Metadata

Labels

No labels
No labels

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions