Cerebras Built AI Chip Supply Chain With 5nm Process
The company reported a $25.4 billion sales backlog as it scales its unique on-chip memory architecture.
Updated on Oct. 4, 2026 in Semiconductors
Live Poll
Do you trust that major tech companies will maintain stable service despite reliance on specific suppliers?
Cerebras Systems has established an AI chip supply chain utilizing 5nm process technology and on-chip SRAM to circumvent common industry chokepoints. The company currently reports a $25.4 billion sales backlog, with OpenAI accounting for more than $20 billion of that total.
Why it matters
By integrating memory directly as SRAM rather than relying on external high-bandwidth memory stacks, the company avoids production delays in CoWoS packaging and 3nm manufacturing. This strategic design choice enables the firm to maintain operations while managing significant capacity commitments.
The WSE-3 chip integrates 900,000 AI cores and 44 GB of on-chip SRAM into a 46,225 mm² area, featuring approximately 4 trillion transistors. These chips utilize TSMC's 5nm process node to achieve performance metrics independent of external HBM stacks.
The players
Cerebras Systems
This hardware company focuses on developing specialized AI computing systems designed to accelerate machine learning workloads.
OpenAI
This artificial intelligence research organization is responsible for creating large-scale models like GPT and maintains a significant hardware backlog with suppliers.
Andrew Feldman
He serves as the CEO of Cerebras Systems and has been a primary spokesperson for the company's technical performance metrics.
Gimlet Labs
This technology firm entered into a supply agreement with Cerebras for 100 MW of data-center hardware capacity.
TSMC
This semiconductor foundry provides the manufacturing process technology used to produce the 5nm chips described in this report.
The details
Cerebras has secured over 600 MW of data-center capacity, including a recent agreement to supply CS-4 systems for 100 MW of capacity to Gimlet Labs. Following its May 2026 Nasdaq IPO, the company continues to see growth, with Q2 revenue doubling compared to the previous year.
Timeline
March 2024: Cerebras announced the WSE-3 chip.
May 2026: The company completed an initial public offering on the Nasdaq exchange.
June 2026: CEO Andrew Feldman discussed the chip's inference performance.
September 2026: Cerebras agreed to supply Gimlet Labs with CS-4 systems.
Q1 2027: Deployment is planned for the General Compute agreement.
The Tech Race
Cerebras's decision to use on-chip SRAM marks a strategic departure from the industry-wide reliance on the high-bandwidth memory (HBM) supply chain. This move positions the firm as a direct alternative to competitors who remain tethered to traditional packaging bottlenecks.
The firm's ability to maintain a massive supply chain impacts the availability and cost of large-scale AI computing resources for institutional clients. Users may see changes in the accessibility of high-compute hardware as the company expands its manufacturing capacity in the United States.
The takeaway
By moving memory directly onto the processor, Cerebras successfully isolates its production line from common semiconductor industry constraints. This strategy demonstrates that alternative architectures can provide viable scaling paths for large-scale AI infrastructure.
What happens next
Cerebras is scheduled to begin deployment for its General Compute agreement with Gimlet Labs in Q1 2027.
Further reading
For more information on the current state of the industry, visit the Semiconductors section.
Source note: This article includes information reported by Crypto Briefing.
Live Poll
Do you trust that major tech companies will maintain stable service despite reliance on specific suppliers?
