What's happening
A new bare metal technology lab called ESTROLABS has announced operations with an unusual pitch: modern computing stacks are too fragile, and the industry needs to rebuild from the hardware up.
The lab operates on bare metal infrastructure - no virtualization, no cloud instances - and publishes research while keeping code under commercial licenses. Think of it as open documentation with closed source.
The approach
ESTROLABS uses a version-based development model where every project starts as an MVP and spawns new versions rather than modifying originals. Work moves non-linearly through stages from research to deployment, with all decisions documented publicly.
Three stated priorities: design as system architecture (not just UI), security as foundational (not bolted on), and performance as efficiency (not raw speed). The criticism of layered systems echoes longstanding debates in high-performance computing circles about virtualization overhead.
The business model
Funding comes from donations and commercial licenses for finished products. Code can be inspected but not freely used - a departure from traditional open source that will limit community contributions. No third-party integrations by design.
The lab explicitly states it's "not optimised for general consumers" and warns work is "slow, experimental and imperfect."
What to watch
The bare metal positioning matters if you're evaluating infrastructure. For machine learning workloads or Kubernetes deployments, bare metal can deliver 15-30% performance improvements over cloud instances - but at higher operational complexity.
The harder question: can a single lab materially improve on decades of systems engineering? History suggests fundamental improvements come from research institutions or well-funded companies, not self-funded experiments. The Linux kernel, after all, started as one person's project but succeeded through thousands of contributors.
ESTROLABS has published principles but no shipped products yet. We'll see if the theory survives contact with production workloads.
Worth noting: The rejection of cloud infrastructure and open source licensing makes this more philosophical statement than pragmatic alternative for enterprise buyers evaluating bare metal options in 2025.