A startup working on unconventional AI hardware is making the case that the next major advance in artificial intelligence may come from a different kind of computer, not simply from scaling up today’s model architectures. In a recent discussion on The Neuron podcast, Great Sky co-founder and chief executive Jeff Shainline outlined why the company is building brain-inspired systems based on superconductors and photons rather than conventional GPUs.
Shainline described Great Sky’s technology as Superconducting Optoelectronic Networks, or SOENs. The idea is to move beyond the standard digital computer design that dominates modern AI infrastructure. According to the company, the same architecture that has powered recent model breakthroughs may not be the best fit for future systems that need to operate more efficiently and handle different kinds of workloads.
Great Sky’s pitch centers on the belief that AI hardware should look more like the brain in some important ways. That means using analog signals, dense connections, and a division of labor between electrons and photons. The company argues this approach could help address the memory and communication bottlenecks that limit current chips when they are pushed to run increasingly complex AI systems.
The company’s ambitions also extend to the environment in which the hardware operates. Shainline discussed the role of cryogenic temperatures, saying that cooling systems are part of the challenge, but not necessarily the biggest one. Great Sky’s approach relies on superconducting components that work at very low temperatures, but the company appears to view that requirement as manageable compared with the broader problem of building a computer architecture suited to AI workloads.
The discussion also touched on whether the systems could handle language models, which are the backbone of many popular AI products. Great Sky does not appear to be positioning its hardware as a simple drop-in replacement for existing model training setups. Instead, it is pursuing a design that could be better matched to tasks where conventional transformers may be overly resource-intensive.
That focus suggests a broader argument about the direction of the AI industry. Rather than assuming that progress will always come from bigger neural networks running on faster GPUs, Great Sky is betting that new physical computing paradigms may unlock different capabilities. The company’s outlook is especially relevant as data centers, chipmakers, and AI developers continue to look for ways to improve performance without endlessly increasing power consumption and infrastructure costs.
Great Sky also framed its work in terms of practical markets. In the conversation, the company pointed to early opportunities in science, cloud computing, and hyperscalers. It also suggested that some near-term use cases, such as analyzing video frame by frame for content moderation, could benefit from new hardware designed for high-throughput inference tasks.
The company’s roadmap includes systems aimed at the 100 million parameter range, along with longer-term plans that would require changes from foundries and advances in wafer-based scaling, optics, and superconducting components. Great Sky’s approach remains experimental, but its message is clear. If AI continues to evolve beyond today’s dominant model designs, the hardware underneath it may need to change as well.