Cerebras refused to cut the wafer into chips: how wafer-scale computing became an AI company
Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie and Jean-Philippe Fricker founded Cerebras Systems in 2015 with the goal of building an AI processor at wafer scale.
Instead of cutting a processed silicon wafer into many separate dies, the team designed around an enormous interconnected processor spanning most of a wafer.
That starting point is useful context for understanding the modern Cerebras Systems. The company did not arrive at its current position in one jump; it accumulated technology, customers and operating knowledge through a series of distinct transitions.
The WSE-1 made the impossible architecture tangible
Cerebras unveiled its first Wafer Scale Engine and CS-1 system in 2019, demonstrating that a wafer-sized processor could be manufactured, powered, cooled and programmed as a practical system.
In semiconductor and electronics markets, this kind of capability compounds over time. Manufacturing know-how, yields, supplier relationships and process experience are difficult to recreate quickly, so each generation can strengthen the next.
Successive wafer engines increased memory and compute density
Later WSE generations improved transistor count, on-chip memory and performance while Cerebras built software intended to make the unusual architecture accessible to machine-learning developers.
In semiconductor and electronics markets, this kind of capability compounds over time. Manufacturing know-how, yields, supplier relationships and process experience are difficult to recreate quickly, so each generation can strengthen the next.
Cloud inference made the system available without owning the hardware
Cerebras expanded beyond on-premise supercomputers into hosted inference and training services, allowing developers to use wafer-scale systems through APIs.
This shift also changed how customers consumed the technology. Instead of treating software or infrastructure as a one-off purchase, the relationship became more continuous, with updates, capacity and services delivered over time.
The 2026 public listing opened a new capital chapter
Cerebras began trading on Nasdaq in May 2026 and continued expanding data-centre deployments around high-speed AI inference and training.
For Cerebras Systems, access to public markets was more than a branding milestone. It increased financial transparency and created another route to fund expansion, acquisitions or long-term product development.
The wafer itself became the processor
As of 18 September 2026, cerebras remains defined by one radical architectural decision: make one processor vastly larger than a conventional chip package. Its commercial challenge is to convert that technical differentiation into a durable software and cloud ecosystem.
The company’s current position is therefore the result of accumulated technical and commercial choices rather than one breakthrough product. That is the context future coverage should preserve.
For TechnologyBlog.co.za, this page is intended to serve as the company-history cornerstone for future Cerebras Systems coverage, so product and company-news articles can link back to one factual brand background instead of repeating the full origin story.
Primary sources checked include Cerebras’ official company timeline and public-market material. Information reflects public material available on 18 September 2026. Official company identity cross-checked against https://www.cerebras.ai.
