Organisation
Cerebras
AI-chip maker known for wafer-scale processors the size of a dinner plate, pitched as an alternative to Nvidia GPUs particularly for fast model inference.
Cerebras Systems is an AI-chip maker, founded in 2015 and led by Andrew Feldman, known for an unusual design: wafer-scale processors roughly the size of a dinner plate, each carrying trillions of transistors, pitched as an alternative to Nvidia GPUs particularly for fast model inference. It has grown quickly on the back of a large inference deal with OpenAI and other customers, but its revenue has been heavily concentrated — filings disclosed that most of it came from a small number of clients linked to the United Arab Emirates. After withdrawing an IPO in 2025, it listed on Nasdaq in May 2026 at a valuation of roughly $56 billion, making it one of the clearest public-market tests of whether investors will price a challenger to Nvidia on a comparable scale.
- Category
- Compute, chips & infrastructure
- Founded
- 2015
- HQ
- Sunnyvale, US
- Funding
- ~$56bn valuation at its 2026 Nasdaq IPO
- Key people
- Andrew Feldman
Appears alongside
Featured in threads
Tracks
- Compute & infrastructure 5
- Models & capabilities 2
- Money & business 2
- Open weights & ecosystem 1
OpenAI adds a much faster response tier for its flagship model, built on Cerebras chips
The 'Ultrafast' preview reaches up to 750 tokens per second, as much as 14 times OpenAI's standard speed, at the same intelligence level as GPT-5.6 Sol.
Models & capabilities · Compute & infrastructure
Cerebras completes IPO on Nasdaq at ~$56bn valuation
Shares priced at $185 opened near $350 on the first day of trading, and Cerebras disclosed that roughly 86% of its 2025 revenue came from two UAE-linked customers.
Compute & infrastructure · Money & business
OpenAI releases GPT-5.3-Codex-Spark, an ultra-low-latency coding model
Served on Cerebras' Wafer Scale Engine 3 rather than OpenAI's usual infrastructure, the smaller model hit over 1,000 tokens per second — about 15 times the standard Codex model's speed.
Models & capabilities
OpenAI partners with Cerebras for 750MW of compute
The multi-year, reportedly $10bn-plus deal covers wafer-scale chips for fast inference — not model training — deployed in phases through 2028.
Compute & infrastructure
Cerebras withdraws IPO filing days after $1.1bn Series G
The AI chipmaker told the SEC its year-old prospectus had gone stale, days after a Fidelity-led $1.1 billion round valued it at $8.1 billion; it listed on Nasdaq the following May.
Compute & infrastructure · Money & business
Cerebras releases seven open Cerebras-GPT models
Cerebras released seven open GPT-style models (111M-13B) trained on its wafer-scale systems under Apache 2.0, with an open, reproducible scaling-law study.
Open weights & ecosystem · Compute & infrastructure