Junietta is developing efficient local large-language-model hardware built around transformer silicon and digital compute-in-memory technology. The company is targeting an affordable system for running Llama 3.2 locally, with first-silicon research tests planned for November 2026 through IHP’s 130nm process and software support for PyTorch and tinygrad.
PhynomyPhynomy is a deep-technology company developing specialized silicon for AI inference, with a focus on running agent loops efficiently on a single chip. Its physics-first approach uses computing inside memory so multiple models can remain loaded and pass work between one another without repeatedly moving their weights. Based in New York and Patna, Phynomy has a granted patent covering its core memristor device and has fabricated and measured memory arrays, though its production chip has not yet shipped.
LuxcoreAtlanta-based Luxcore is developing a photonic fabric and integrated infrastructure platform for distributed, sovereign AI computing. Its approach, which it calls “Scale-Across,” connects power-constrained data centers into a logical compute fabric while pairing the network with policy software and infrastructure management. The company’s Secure Fabric remains in development, with performance figures presented as engineering targets rather than validated production results.New York, NY
Efficient ComputerPittsburgh-based Efficient Computer develops energy-efficient general-purpose processors and the software tools that run them, led by the Electron E1 and its companion effcc Compiler. Its Fabric architecture uses spatial dataflow to keep computation close to data, bringing accelerator-like efficiency to programmable devices such as remote sensors, wearables and autonomous systems. Founded by Carnegie Mellon University professors Brandon Lucia and Nathan Beckmann alongside Graham Gobieski, the company pairs semiconductor development with browser-based engineering tools through Efficient Labs.PITTSBURGH, PA
HobbyJetHobbyJet is developing a compact waterjet built specifically for classrooms, makerspaces, FabLabs, and career-and-technical-education programs. Designed to run from a standard 110V outlet, the machine pairs with free, open-source CAD and toolpath software and comes with project-based curriculum, bringing hands-on precision fabrication into school settings without an industrial footprint. HobbyJet plans to launch its public beta in fall 2026.