Key Takeaways
- Great Sky, a Boulder, Colorado-based AI hardware startup, closed a $14 million seed round led by Bison Ventures, with Matchstick Ventures, Range Ventures, and notable angel investors including Mark Leslie and Adam Pritzker participating
- The company’s Superconducting Optoelectronic Network (SOEN) platform can process more than 60 million video frames per second, compared to roughly 30 frames per second on conventional GPU-based systems
- Great Sky’s technology is grounded in over a decade of research at the U.S. National Institute of Standards and Technology (NIST), with a founding team that has published more than 20 peer-reviewed papers in the field
- Fresh capital will fund commercialization across high-stakes verticals including defense, energy, and large-scale AI infrastructure
Quick Recap?
Great Sky, a brain-inspired computing startup headquartered in Boulder, Colorado, officially came out of stealth on March 13, 2026, announcing both its Superconducting Optoelectronic Network (SOEN) architecture and a $14 million seed round in the same breath. The raise was led by Bison Ventures and confirmed via a Business Wire press release, with participation from Matchstick Ventures, Range Ventures, and angel investors Mark Leslie, Adam Pritzker, and Ivan Vendrov. The company is positioning itself as a direct challenge to the GPU-centric AI infrastructure that dominates today’s data centers.
Great Sky’s SOEN Technology
Great Sky is not building another GPU variant or a custom ASIC tweak. The company is rebuilding the computing stack from the physics layer up. Its Superconducting Optoelectronic Networks (SOENs) combine three distinct technological pillars: superconducting circuits that operate at cryogenic temperatures, optical communication links that transmit signals at single-photon sensitivity, and semiconductor circuitry that bridges the electronic and photonic layers.
What this architecture achieves is a system that mirrors how biological neural networks actually work. Memory and processing are co-located rather than separated, which eliminates the so-called “von Neumann bottleneck” that plagues conventional chips when shuttling data between processors and memory. The result is a platform built for continuous learning and on-device adaptation, meaning AI models can update from live data streams without expensive retraining cycles.
The numbers are hard to ignore. Great Sky claims its chips can process more than 60 million video frames per second versus the roughly 30 fps of standard GPU systems. Separately, a report from Semafor noted the company claims processing speeds more than 1 million times faster than conventional models for specific video workloads when run at scale.
The founding team, led by CEO Jeff Shainline, a former NIST scientist whose work has been cited over 5,000 times and who received a Presidential Early Career Award for Scientists and Engineers in 2024, brings deep credibility to these claims.
The company’s chips have already completed successful tape-outs, demonstrating measurable improvements in energy efficiency and performance over traditional silicon-based systems, and the SOEN architecture supports high-throughput, low-latency workloads including spoken language processing, real-time video analysis, and multimodal AI applications.
Next AI Hardware Race After GPUs
Great Sky’s announcement lands in the middle of a fast-moving wave of capital flowing toward alternative compute architectures. Tech companies are facing an acute data center capex problem: the energy and infrastructure demands of running transformer-based AI models on GPUs are scaling faster than costs are coming down. That tension is opening doors for startups that can credibly claim orders-of-magnitude efficiency gains.
The broader industry context shows neuromorphic and superconducting chips are capturing serious investor attention. Just nine months before Great Sky’s announcement, Snowcap Compute launched with a $23 million seed round led by Playground Global in June 2025 to build superconducting AI chips, with former Intel CEO Pat Gelsinger joining its board.
In December 2025, Unconventional AI raised a staggering $475 million seed round led by a16z and Lightspeed at a $4.5 billion valuation to pursue neuromorphic, biology-inspired computing. The market signal is clear: investors believe the transformer-on-GPU paradigm is approaching a ceiling. Research backs the energy efficiency argument.
Studies indicate neuromorphic systems achieve 20 to 100 times better energy efficiency than GPUs for inference tasks, while the human brain runs on approximately 0.3 kilowatt-hours per day versus 10 to 15 kWh daily for a high-end GPU. Great Sky’s superconducting approach takes this further by eliminating electrical resistance at the circuit level, theoretically pushing efficiency beyond what any silicon-based design can reach.
From a regulatory and geopolitical lens, defense and national security applications add another dimension of urgency. Great Sky specifically names defense as a target vertical, and the NIST pedigree of its founding team means the company has built-in credibility with U.S. government procurement channels.
Competitive Landscape
Great Sky enters a niche but rapidly heating segment of the AI hardware market. The two most directly relevant seed-stage competitors operating in the superconducting and neuromorphic compute space are Snowcap Compute and Unconventional AI.
| Feature / Metric | Great Sky | Snowcap Compute | Unconventional AI |
| Funding Raised (Seed) | $14M | $23M | $475M |
| Core Technology | Superconducting Optoelectronic Networks (SOENs) with single-photon optical links | Superconducting CMOS-replacement chips (niobium titanium nitride) | Neuromorphic / analog mixed-signal chips with spiking neural networks |
| Architecture Inspiration | Biological neural network with co-located memory and processing | Superconducting logic for AI, quantum, and HPC workloads | Brain-inspired, event-driven spiking neurons targeting 1000x GPU efficiency |
| Key Efficiency Claim | 60M+ video frames per second vs. ~30 fps on GPU; 1M+ times faster for select video | 25x performance per watt over current best chips (including cooling overhead) | ~1000x energy efficiency improvement over GPUs (theoretical) |
| Target Verticals | Defense, energy, large-scale AI infrastructure | AI inference, training, HPC, quantum-classical hybrid | Cloud providers, supercomputing centers |
| Lead Investor | Bison Ventures | Playground Global | Andreessen Horowitz, Lightspeed |
| Notable Backers | Matchstick Ventures, Range Ventures, Mark Leslie, Adam Pritzker | Cambium Capital, Vsquared Ventures, Pat Gelsinger (board) | Jeff Bezos, Lux Capital, DCVC, Databricks |
| Research Foundation | 10+ years at U.S. NIST; 20+ peer-reviewed papers | Team from Cadence, Northrop Grumman, Imec | Founded by former Databricks AI head Naveen Rao |
| First Chip Timeline | Tape-outs already completed | First basic chip targeted by end of 2026 | First chips expected in 2026 |
| HQ | Boulder, Colorado | Palo Alto, California | Undisclosed (U.S.-based) |
Strategic Analysis
Great Sky leads on technical depth and hardware readiness, having already completed chip tape-outs while competitors are still targeting 2026 for their first silicon. However, Snowcap has a funding edge at $23M versus Great Sky’s $14M, and benefits from Pat Gelsinger’s board presence for manufacturing credibility. Unconventional AI’s $475M war chest is in a different league entirely, but its efficiency claims remain largely theoretical at this stage, while Great Sky has empirical benchmark data to back its numbers.
TechnoTrenz’s Takeaway
I will be direct: this is one of the more genuinely exciting seed announcements I have seen in the AI hardware space in recent memory, and I think most of the industry is underestimating what it means.
In my experience tracking deep-tech hardware startups, the biggest red flag is usually a gap between the research pedigree and the commercial execution. Great Sky does not have that problem. Jeff Shainline spent over a decade at NIST, published foundational work that the broader scientific community has cited over 5,000 times, and has already put test chips through tape-out. That is not a PowerPoint company; that is a team that has done the hard physics before asking for money.
