Intel Labs and Sandia National Laboratories Deploy Hala Point: 1.15-Billion Neuron Neuromorphic AI Supercomputing System

The Three-Question Framework

What Happened?

Intel Labs announced the successful deployment and benchmarking of 'Hala Point' at Sandia National Laboratories, the world's largest neuromorphic research system. Integrating 1,152 Loihi 2 processors within a six-rack-unit enclosure, Hala Point supports up to 1.15 billion artificial neurons and 128 billion synapses across 140,544 neuromorphic processing cores, executing 20 petaops of continuous spiked computation on a power budget of just 15 kilowatts.

Why It Matters

As modern LLMs approach multi-megawatt energy footprints that strain local electrical grids, neuromorphic computing offers a biological alternative based on how the human brain functions. Traditional GPUs continuously consume electricity recalculating entire matrix tensors; in contrast, neuromorphic architectures process information event-driven via asynchronous voltage spikes, consuming power only when active biological neurons fire, resulting in 100-fold energy efficiency gains.

What Could Happen Next

Sandia researchers are programming Hala Point to solve complex non-linear physics simulations, smart electrical grid optimization, and autonomous drone swarm path planning in GPS-denied environments.

Executive Key Takeaways

World's largest neuromorphic system: 1.15 billion artificial neurons and 128 billion synapses.

1,152 Loihi 2 processors: fabricated on Intel 4 process with 140,544 event-driven neuromorphic cores.

Extreme energy efficiency: delivers 15 trillion 8-bit operations per watt (15 TOPS/Watt).

Asynchronous spike processing: consumes zero static energy across idle synaptic nodes, mimicking biological cerebral cortex efficiency.

On August 10, 2026, Intel Labs and the US Department of Energy's Sandia National Laboratories announced the operational deployment of Hala Point, establishing a new world record for brain-inspired computing scale.

In modern von Neumann computer architectures, CPUs and GPUs constantly shuttle data back and forth between separate compute units and memory banks, creating the infamous 'memory wall' that accounts for the majority of datacenter energy consumption. The human brain, by comparison, performs complex multimodal reasoning on a power budget of roughly 20 watts. It achieves this by co-locating memory and processing inside billions of biological neurons that communicate via sparse, asynchronous electrical impulses known as spikes.

Hala Point mimics this biological architecture on silicon. The system packs 1,152 Intel Loihi 2 processors fabricated on the Intel 4 process node into a standard six-rack-unit server chassis roughly the size of a microwave oven. When executing spiking neural networks (SNNs), individual artificial neurons only compute and transmit data when incoming synaptic potential crosses an activation threshold.

In standardized workload evaluations, Hala Point solved complex scientific optimization problems—including graph coloring and linear programming—up to 50 times faster and using 100 times less energy than conventional CPU and GPU architectures. The system demonstrated an operational throughput of 20 petaops with an efficiency reaching 15 trillion operations per watt for 8-bit operations.

The deployment provides national laboratory researchers with the computational platform necessary to explore post-GPU computing paradigms for defense, physics, and autonomous robotic systems.