Century Report arc
AI Chips on Non-Silicon Substrates
The assumption that artificial intelligence requires silicon is dissolving, as researchers find computation latent in glass interconnects, exotic material physics, and living neural tissue, each substrate offering a different physical grammar for inference.
- Status
- Active
- Developments
- 6
- Last 30 days
- 0
- Last covered
Current synthesis
Where it stands
Non-silicon substrates are beginning to influence commercial AI without entering the rack: rat neurons now tune a software layer distributed through AWS (09-23), extending earlier biological-computing demonstrations. The preview moves bio-derived computation toward commercial distribution, BUT its claimed fivefold speed and 80% cost gains lack independent benchmarks, and deployed inference still runs on conventional GPUs or Amazon Trainium.
Complete record
Development timeline
6 qualifying developments, newest first
September 2026
-
[DD+POD] The Biological Computing Company opened AWS preview access to a video model tuned through rat-neuron cultures, claiming five-times-faster generation and 80% lower inference cost through a software layer under 0.1% of the base model. [Source](https://www.wired.com/story/ai-models-built-from-rat-brains-are-about-to-become-a-reality/)
August 2026
-
[DD+POD] MIT engineered the bacterium Pantoea agglomerans into working transistors and assembled them into functioning circuits, including a two-input adder coordinating 24 colonies over an 8-hour cycle, envisioning circuits printed onto plant leaves to sense and respond to drought or pests.
June 2026
-
[DD+TCR] MIT team embedded GaN transistors into lab-grown diamond via femtosecond laser dicing and cavity placement at commercially viable scale, fabricating a power amplifier outperforming every comparable device in the literature: thermal ceiling on high-power wireless electronics lifted by routing heat onto the highest-conductivity material known.
April 2026
-
Cortical Labs opened the first biological data center in Melbourne, with CL1 units housing 200,000 lab-grown human neurons on silicon microelectrode arrays for reservoir computing; Singapore expansion planned.
-
USC memristor (Nature) survives 700°C, 1B+ switching cycles using tungsten-graphene interface; performs matrix multiplication through physics rather than digital computation: extreme-environment AI inference architecture demonstrated.
March 2026
-
MIT Technology Review documents glass substrates as emerging pathway for future AI chip fabrication, enabling denser interconnects and lower power at scales silicon cannot reach: enters mainstream technology press as credible near-term alternative architecture.