Snapshot 2026-11-12. This page reads the company's own job board, where 124 full-time requisitions were advertised on that date. The role and skill counts below cover the 83 engineering-titled roles that carried a usable description; every count is an open posting, not a hire.
Is Cerebras actively hiring?
Yes, and the pace increased through 2026. Ten requisitions were first published in the seven days before this snapshot, twenty in the previous thirty, and seventy inside ninety days. The median requisition is 71 days old, so the board is a live pipeline rather than a stale archive.
| Signal | Value |
|---|---|
| Open requisitions | 124 |
| Published in the last 7 days | 10 |
| Published in the last 30 days | 20 |
| Published in the last 90 days | 70 |
| Median requisition age | 71 days |
| Engineering-titled roles analysed | 83 |
New requisitions by month show the ramp: 6 in May, 10 in June, 15 in July, 11 in August, 34 in September, 19 in October, and 10 in the first eleven days of November.
Where the roles are
The company's own departments are the clearest view of what it is building toward.
| Department | Open roles | Share | Median age (days) |
|---|---|---|---|
| Software Engineering | 59 | 48% | 114 |
| Datacenters | 16 | 13% | 25 |
| IT & Security | 16 | 13% | 60 |
| Hardware | 14 | 11% | 62 |
| Corporate | 8 | 6% | 72 |
| Product | 6 | 5% | 34 |
| University Departments | 4 | 3% | 148 |
| Sales & Partnerships | 1 | 1% | 120 |
Software engineering is the largest block but also the oldest: its median requisition was first published 114 days ago, and only 26 of its 59 roles are inside the last quarter. Datacenters is the opposite: 13 of 16 roles were published inside ninety days and the median is 25 days. Seven of the ten newest postings on the whole board belong to that department, including four staff technical program manager roles for data-center new-product introduction, supply chain, end-to-end delivery, and data tooling.
Which skills the roles name
A technique is counted once per role when the role's own text names it. Company-wide marketing and culture boilerplate is removed first, so a phrase every posting repeats is not scored as a requirement.
| Technique | Roles | Share |
|---|---|---|
| Kubernetes / cloud infrastructure | 37 | 45% |
| Distributed systems | 34 | 41% |
| GPU / inference optimization | 28 | 34% |
| Latency / serving / caching | 27 | 32% |
| Low-level performance (concurrency, SIMD, memory) | 25 | 30% |
| LLM application | 22 | 26% |
| Infrastructure as code | 14 | 17% |
| Deep learning / transformers | 14 | 17% |
| Search agents / tool use | 13 | 16% |
| Training / fine-tuning / RL | 6 | 7% |
| Crawl / index / data pipeline | 5 | 6% |
| Hybrid retrieval / reranking | 4 | 5% |
The shape is infrastructure and serving work rather than model research: distributed systems, kubernetes, serving latency and low-level performance together describe the majority of roles, while training and fine-tuning appear in six. Classical search and ranking skills are a minority — lexical retrieval, embeddings and reranking sit in single digits — so a search-background candidate should expect the loop to test systems and serving depth instead.
Seniority and locations
| Level stated in the title | Roles | Share |
|---|---|---|
| Intern / new grad | 4 | 5% |
| Mid or unspecified | 41 | 49% |
| Senior | 22 | 27% |
| Staff | 12 | 14% |
| Principal | 4 | 5% |
| Primary location | Roles |
|---|---|
| Sunnyvale, CA | 66 |
| Toronto, Canada | 8 |
| United States and Canada | 3 |
| Bengaluru, India | 3 |
| Abu Dhabi, UAE | 1 |
| Other or multiple locations | 2 |
Roughly half the engineering roles ask for senior or above in the title, and the entry point is narrow: four intern or new-grad requisitions against 124 total. Sunnyvale is the default; Toronto is the only meaningful second hub, with a handful of roles open to both countries and a small Bengaluru footprint.
What this means if you are preparing
- Expect serving and systems questions: batching, streaming, routing, latency and throughput trade-offs
on an unusual accelerator, rather than a GPU-cluster baseline.
- The data-center roles are program and operations interviews — capacity planning, supply chain, vendor
and construction coordination — not algorithm rounds.
- Hardware roles name their discipline precisely (RTL, design verification, packaging, power and
manufacturing test), so match the posting line by line.
- New-grad candidates should target the named new-grad tracks directly rather than applying broadly.
Method and limits
- Source: the company's own Ashby job board, read through its public posting API on 2026-11-12.
- Counts: requisitions as advertised that day. A requisition is not a hire, an approved headcount slot,
or a filled position, and nothing here observes offers, starts or attrition.
- Skills: a title-and-description census. It measures what a description names, not the depth a role
requires, and a technique counts once per role regardless of how often it appears.
- Scope: engineering-titled roles with a usable description. The 41 non-engineering roles on the board
are counted in the requisition total and excluded from the skill counts.
- Freshness: requisition age is time since first publication, so a reposted or evergreen role looks
older than the vacancy is.