A startup called Great Sky is making the case that the next major advance in artificial intelligence may come from a different kind of computer, not simply from larger models. In a recent interview, the company described its bet on hardware designed to mimic some features of the brain while avoiding the limits of conventional chips.
Great Sky is developing Superconducting Optoelectronic Networks, or SOENs, a system that combines superconducting circuits for computation with light for communication. Jeff Shainline, the company’s co-founder and chief executive, said the approach is intended to move beyond the architecture that has powered today’s AI systems, especially GPUs paired with transformer models.
The company’s core argument is that mainstream computing hardware may not be well suited to the next generation of AI workloads. Rather than continuing to scale existing chip designs, Great Sky is exploring hardware that it says could support denser connectivity, lower-latency signaling and new forms of computation.
In the interview, Shainline emphasized that “brain-like” computing does not mean building a literal brain. Instead, the idea is to borrow useful traits from biology, such as highly interconnected networks and the ability to process signals in ways that are closer to the analog behavior of the brain than to the purely digital operation of standard processors.
Great Sky’s system uses superconducting elements, including Josephson junctions, alongside optical links. The company says that setup could help address a persistent bottleneck in AI systems: memory movement. Much of the energy and delay in today’s hardware comes from shuttling data back and forth between memory and compute units. Great Sky is trying to reduce that friction by redesigning how the system is organized.
Shainline also framed the technology as “quantum-adjacent” rather than quantum computing. That distinction matters because Great Sky is not trying to build a quantum computer for general use. Instead, it is working on superconducting hardware that operates in a different way from the GPU-based systems that currently dominate AI training and inference.
The company said its architecture may be especially useful for workloads that require continuous, high-speed processing of large volumes of data. One example discussed in the interview was video analysis. Great Sky suggested that applications such as content moderation, scientific instruments, fusion reactors and particle colliders could be among the early markets if the hardware proves practical.
The startup also described a roadmap that includes systems with around 100 million parameters, though it did not present the technology as a finished commercial product. It acknowledged that major changes would be needed in manufacturing, including cooperation from chip foundries and advances in related areas such as optics and superconducting materials.
Great Sky said its first targets are likely to come from science, cloud services and large technology customers rather than consumer devices. The company’s pitch is that some AI and scientific workloads may benefit from hardware optimized for throughput and connectivity instead of raw general-purpose flexibility.
The interview also touched on scale, including the challenge of moving from a research concept to deployable infrastructure. Great Sky’s leaders described the need to turn capital into actual chips, refine the architecture, and prove that the system can operate outside a lab setting.
For now, the company is positioning itself as part of a broader debate about whether AI progress will continue to come primarily from bigger models, or whether the field will need a new generation of specialized hardware to keep advancing.