News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip

Cerebras WSE-3 Turbo

Wafer-scale AI processor with 4 trillion transistors

The Cerebras WSE-3 Turbo is a wafer-scale artificial intelligence processor released in 2026. It contains four trillion transistors and delivers 250 petaflops of compute power for training and inference tasks.

Share on WhatsApp

  • Contains 4 trillion transistors on a single wafer
  • Delivers 250 petaflops of AI compute performance
  • Features 900,000 AI-optimized cores
  • Clock speed increased to 2.8 GHz in Turbo model
  • Includes 44 GB of on-wafer SRAM memory

What it is

The Cerebras WSE-3 Turbo is an artificial intelligence processor built on a single silicon wafer. Cerebras Systems designed this chip to handle large-scale machine learning workloads. It measures over 46,000 square millimetres and holds four trillion transistors. The device provides 250 petaflops of computational performance. This makes it one of the largest and most powerful processors available for AI tasks. It serves both training and inference needs within data centres.

How it came about

Cerebras launched the original WSE-3 in March 2024 using a five nanometre process. In August 2026, the company introduced the Turbo variant without changing the core count or memory size. This update coincided with the release of the CS-4 system, codenamed Nexus. The new version doubled the clock speed compared to its predecessor. This improvement allowed the chip to perform twice the work of the earlier model while keeping the same physical layout.

Inside the chip

The processor features 900,000 cores optimised for artificial intelligence calculations. It includes 44 gigabytes of static random-access memory located directly on the wafer. The design uses redundant cores and routing paths to manage manufacturing defects. This fail-in-place architecture allows the chip to shut down flawed sections and route around them. The clock speed reaches 2.8 gigahertz, which is double the rate of the standard WSE-3. This higher frequency requires increased power delivery and enhanced cooling solutions.

How it was made

Taiwan Semiconductor Manufacturing Company produces the chip using a five nanometre process node. The entire silicon wafer functions as a single processor rather than being cut into smaller dies. This approach yields a massive surface area of 46,225 square millimetres. The manufacturing strategy relies on redundancy to overcome yield issues common in such large structures. The package integrates the wafer with power modules and cold plates for thermal management.

Where it went

The WSE-3 Turbo powers the Cerebras CS-4 rack-scale AI system. Each CS-4 unit contains three of these processors linked by high-speed connections. The system targets hyperscale deployment for generative AI models. It supports direct wafer-to-wafer links for building larger clusters. Partners such as OpenAI and Amazon Web Services can integrate their own network interfaces. The hardware aims to provide faster inference speeds compared to traditional graphics processing unit systems.

Versions and family

The Turbo version is an overclocked iteration of the standard WSE-3 engine. Both chips share the same 900,000 cores and 44 gigabytes of memory. The primary difference lies in the clock speed, which increased from 1.4 gigahertz to 2.8 gigahertz. Future generations may involve different engines, but the current roadmap focuses on improving throughput within the existing Nexus rack design. The CS-4 system represents the main platform for this specific processor variant.

Why it matters

This processor demonstrates the viability of wafer-scale computing for artificial intelligence. By integrating trillions of transistors on one piece of silicon, it reduces the need for complex interconnects between multiple chips. The architecture offers a simpler path to scaling AI infrastructure. It challenges conventional designs that rely on connecting many smaller processors. The increased clock speed in the Turbo model shows how performance can grow without shrinking the manufacturing node further.

Questions people ask about Cerebras WSE-3 Turbo

How does the WSE-3 Turbo handle manufacturing defects?

The chip uses a fail-in-place architecture with redundant cores and routing. If a section of the wafer has flaws, the system shuts down those specific areas and routes data around them. This allows the large wafer to function correctly despite minor imperfections.

What is the difference between the WSE-3 and WSE-3 Turbo?

Both chips have the same number of cores and memory capacity. The Turbo variant runs at a clock speed of 2.8 GHz, which is double the 1.4 GHz of the original. This increase allows the Turbo model to perform twice the amount of work.

Which company manufactures the Cerebras WSE-3 Turbo?

Taiwan Semiconductor Manufacturing Company (TSMC) produces the chip. They use a five nanometre process node to create the wafer-scale engine. Cerebras Systems designs the processor and integrates it into their CS-4 systems.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Cerebras WSE-3 Turbo yet, so the facts on this page come from the sources listed above: cerebras.ai/chip, nextplatform.com/compute/2026/08/19/cerebras-overclocks-wse-3-wafersca. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.