Redefining Compute Physics

We hypothesize that heterogeneous compute orchestration and technological breakthroughs in compute physics are the keys to unlocking AGI.

The Symphony Challenge

Our hypothesis aligns with a significant school of thought—often called hardware-software co-design—which argues that AGI is primarily a resource and orchestration problem rather than a purely algorithmic one.

While the "Transformer moment" provided the logic, experts now point to a "Symphony Challenge" where AGI emerges from a complex ecosystem of hardware and software working in harmony.

1. Breakthroughs in Compute Physics

Standard transistor scaling (Moore’s Law) is hitting physical limits, making breakthroughs in alternative physics essential for the "brute force" side of AGI.

• Neuromorphic & Optical Computing: Future AGI may require Neuromorphic chips (brain-inspired) or Optical processors (using light instead of electrons) to achieve the 2-3 orders of magnitude efficiency gains needed for real-time human-level reasoning.
• Quantum Acceleration: While not a standalone path, Quantum Computing is projected to provide the "computational muscle" for AGI to evolve autonomously by solving complex optimization subtasks that currently bottleneck classical systems.

2. Heterogeneous Compute Orchestration

Modern consensus, backed by reports from MIT Technology Review and Arm, suggests AGI will not run on a single "super-chip" but on heterogeneous architectures.

• Task-Specific Allocation: AGI workloads require integrating CPUs, GPUs, NPUs, and ASICs into a single system where each core handles what it does best (e.g., CPUs for logic, NPUs for neural acceleration).
• The Orchestration Layer: The "intelligence" of the system will increasingly reside in the orchestration software—the middleware that dynamically partitions and schedules workloads across these diverse processors from the cloud to the edge.

3. The "Idea Bottleneck" vs. Compute

Our hypothesis shifts the focus from Algorithms (new architectures) to Physical Realities.

• Constraint-Led Development: Some researchers argue that AGI progress is currently limited by the physical constraints of computation and energy costs.
• Efficiency Gains: Breakthroughs in compute physics could lower the "barrier to entry" for AGI by making the massive data and power requirements economically feasible.

Join Us

We are a small team of systems engineers, compiler researchers, and kernel hackers. We are building the foundation for the next decade of AI infrastructure. If you want to redefine the physics of compute, join us.