AI Accelerators: The What, the Why, and Open Source Today

AI has outgrown general-purpose compute — yet we’re still stuck in a world dominated by closed, GPU-heavy ecosystems, both in common-awareness and practice.

This session takes a systems-level look at AI accelerators: what they are, basic principles about how they work, and why open source efforts around them matter more than ever.

We’ll cover:

• The rise of specialized hardware in the AI stack—and why CPUs and GPUs alone don’t cut it

• What makes an AI accelerator, including:

  • Compute paradigms

  • Memory models

  • Dataflow architectures

  • Specialization around the particulars of an AI model

  • And more

• A reality check on the state of open source hardware and tooling for AI

  • Why existing open-source efforts mostly build on proprietary hardware—and why building open from the ground up is still a missing piece

  • The rare, often hobbyist, efforts toward replicating or reimagining open AI hardware

• Why the NVIDIA monopoly isn’t just a market issue—it’s a bottleneck for innovation and autonomy

• The need for open, auditable, community-driven alternatives in AI acceleration

• Emerging efforts, gaps, and future directions worth watching (or contributing to)

  • To cite a few;

    • Neuromorphic Hardware

    • Optical Computation

    • Alternative Fabrication methodologies

Whether you're an ML engineer, systems nerd, or just curious about what powers the models behind the curtain, this talk aims to offer a deeper technical and ecosystem-level understanding of where we are—and where we need to go.

• the importance of specialized hardware in the AI landscape

• looking beyond GPUs and the NVIDIA monopoly

• an understanding of the basic principles behind AI accelerators

• the dismal state of open hardware for AI acceleration

• the criticality for open and transparent hardware to come up