A new NSF-funded initiative is making brain-inspired computing hardware freely available to researchers across the United States. THOR: The Neuromorphic Commons launched February 23, 2026 at the University of Texas at San Antonio, housing the largest publicly accessible neuromorphic computing system in the country.
What THOR Offers
At the core of THOR sits the SpiNNaker2 system, developed by German company SpiNNcloud. The platform contains roughly 400,000 energy-efficient ARM-based processing elements capable of simulating over 393 million neurons. This ranks among the top four largest neuromorphic computing platforms globally.
The hardware works fundamentally differently from conventional computers. Rather than running continuously, SpiNNaker2 activates only when new data arrives, mimicking how biological neurons fire in response to stimuli. The chips integrate memory directly with processors using specialized components called memristors, eliminating the energy-hungry data transfers that plague traditional systems.
The numbers tell the story: where the human brain runs on roughly 20 watts (equivalent to two LED bulbs), training GPT-4 consumed approximately 50 times the energy required for GPT-3. As AI electricity consumption is projected to double by 2026, neuromorphic approaches offer a potential path forward.
“THOR is the US national hub for neuromorphic computing,” says founding director Dhireesha Kudithipudi, who holds the Robert F. McDermott Chair in Engineering at UTSA. “We are democratizing the technology.”
How Access Works
The hub operates like a public library. Researchers nationwide can apply for free access, entering a queue to run experiments. When one user finishes, the system becomes available for the next. The model dramatically lowers barriers for emerging engineers who previously had no access to such specialized hardware.
The initiative brings together principal investigators from across the country: Kudithipudi at UTSA, Catherine Schuman at UT Knoxville, Gert Cauwenberghs at UC San Diego, and senior personnel Vijay Janapa Reddi at Harvard. THOR supports research across physics, life sciences, AI/machine learning, and neuroscience.
Neuromorphic Systems Can Do More Than Expected
Recent research from Sandia National Laboratories demonstrates neuromorphic hardware can tackle problems that surprised even experts. Computational neuroscientists Brad Theilman and James Aimone published an algorithm in Nature Machine Intelligence showing neuromorphic systems can solve partial differential equations, the mathematical foundation for modeling weather, fluid dynamics, electromagnetic fields, and structural mechanics.
“You can solve real physics problems with brain-like computation,” says Aimone. “That’s something you wouldn’t expect because people’s intuition goes the opposite way.”
This capability matters for national security applications. The Department of Energy funded the work through its Office of Science and the National Nuclear Security Administration, with an eye toward nuclear deterrent simulations that consume far less power than traditional supercomputers.
The Fine Print
Despite promising results, neuromorphic computing faces real obstacles to mainstream adoption. No “killer application” has emerged to drive widespread investment. Major tech companies have billions sunk into existing GPU and TPU infrastructure. Integration with current AI systems remains incomplete.
And the technology isn’t a drop-in replacement for conventional computing. Training neuromorphic systems requires fundamentally different approaches than the backpropagation techniques powering today’s large language models.
Yet for specific applications, adaptive pacemakers, intelligent hearing aids, real-time medical monitoring, and autonomous vehicles, the technology shows genuine promise. THOR provides the infrastructure for researchers to find out where those boundaries lie.