Each line below was checked against the company's own site or a primary release on 7 October 2026, and links to it. Descriptions are mine, kept plain. This field changes quickly, so treat any row as a starting point and follow the source.
The thing these companies share is a teaching problem. The hardware is scarce, so few developers can touch it, so few can learn it. That is the gap this site exists to fill.
Q.ANT
Photonic processors on PCIe cards, sold in rack servers, that compute AI and HPC workloads with light on thin-film lithium niobate.
- Product
- Native Processing Server, NPU Gen 2
- Based
- Stuttgart, Germany
Lumai
Optical computing systems using free space 3D optics to accelerate AI inference, aimed at the prefill stage.
- Product
- Iris
- Based
- Oxford, UK
Akhetonics
Building an all optical processor for reasoning workloads.
- Product
- RPU
Lightmatter
Silicon photonics interconnect and light sources that link AI chips together.
- Product
- Passage
- Based
- Mountain View, USA
Celestial AI
Optical interconnect for scale up links between AI accelerators.
Acquired by Marvell, completed 2 February 2026
- Product
- Photonic Fabric
Salience Labs
All optical silicon photonic circuit switches for networks inside AI datacentres.
- Product
- OC-32M
- Based
- Oxfordshire, UK
Black Semiconductor
Graphene photonics integrated directly onto standard chips so they exchange data with light.
- Based
- Aachen, Germany
iPronics
Silicon photonics optical circuit switches for AI clusters.
- Product
- iPronics ONE
SpiNNcloud Systems
Brain inspired supercomputer systems for low energy AI on sparse workloads.
- Product
- SpiNNaker2
- Based
- Dresden, Germany
Innatera
Spiking neural processors that find patterns in sensor data on low power edge devices.
- Product
- Pulsar
- Based
- Rijswijk, Netherlands
BrainChip
Neuromorphic edge AI processors, sold as chips and as licensed IP.
- Product
- Akida
- Based
- Laguna Hills, USA
Mythic
Analog AI inference processors for automotive, robotics, defence and enterprise.
- Product
- M1
- Based
- Palo Alto, USA
EnCharge AI
In memory computing accelerator chips spanning edge devices to servers.
- Product
- EN100
- Based
- Santa Clara, USA
Axelera AI
AI inference chips plus cards and systems for edge, embedded and server use.
- Product
- Metis, Europa
- Based
- Eindhoven, Netherlands
Tenstorrent
RISC-V based AI accelerator cards and servers with open source software, plus licensed chip IP.
- Product
- Blackhole
Groq
An AI inference cloud running on its own LPU chips.
Non exclusive licensing deal with Nvidia, December 2025
- Product
- GroqCloud
Etched
Inference chips co-designed with racks and software, built as clusters for frontier models.
- Based
- San Jose, USA
What is missing
Plenty. In-memory and neuromorphic research groups in universities, quantum photonics, and several stealth companies are not here yet. If you build in this space and want a line, open an issue on the site's repository.