Software Engineer - Inference Platform
About the job
- At least 3 years of experience in software engineering, including experience building and operating large-scale distributed systems or cloud infrastructure.
- Experience with distributed systems, ideally with Kubernetes.
- Experience building highly available, latency-sensitive systems at scale.
- Experience with security technologies including certificates, Transport Layer Security, and mutual Transport Layer Security.
- Experience optimizing latency, throughput, and efficiency in high-queries-per-second systems.
- Strong proficiency in a backend or systems language such as Go or C++.
- Design, develop, test, and maintain production software spanning testing, continuous development, observability, security, networking, debugging, and productionization.
- Help shape the technical direction for the Inference Platform, Kubernetes custom resource definitions, failure domains, service boundaries, and system evolution over time, and own the roadmap for major technical areas.
- Architect active-active systems with rapid failover, graceful degradation, and clear service-level objectives.
- Drive system-level improvements in latency, throughput, capacity efficiency, and resilience under unpredictable demand.
- Write and review production code in the most important parts of the platform.
- Make high-consequence architectural decisions within the assigned area and set the technical bar through design reviews, code reviews, and sound engineering judgment.
- Lead the hardest production issues and cross-system bottlenecks.
- Drive observability, incident response, capacity planning, and post-incident improvement with a high standard for operational rigor.
- Partner with ML, Product, Infrastructure, and Cloud teams to translate product and business requirements into scalable system designs and drive alignment on shared technical decisions within the domain and adjacent platform surfaces.
- Experience with ML inference infrastructure, model serving systems, or GPU-accelerated workloads.
- Experience with time to first token and tail-latency reduction.
About the company
Cerebras Systems creates AI acceleration hardware and software. Its CS-2 system is designed to replace traditional GPU clusters for AI workloads, speeding up training and inference while simplifying the setup by eliminating the need for parallel programming, distributed training, and cluster management. The product works as a single, large processor-based accelerator with accompanying software and cloud services to run AI models efficiently, reducing latency and time to results. Compared with competitors, Cerebras differentiates itself with the largest processor in the industry and an integrated hardware-software stack that aims to streamline AI workflows rather than relying on multi-GPU clusters. The company's goal is to help research labs, healthcare, finance, and other industries achieve faster, more cost-effective AI development and deployment by offering a turnkey high-performance AI compute solution.
Company Size
501-1,000
Company Stage
IPO
Headquarters
Sunnyvale, California
Founded
2016
Get referred to Cerebras
See people who can refer or advise you
Simplify's Take
What believers are saying
- Q2 2026 core revenue hit $209.9 million, up 103% year over year.
- Cerebras booked $25.4 billion in RPO on June 30, 2026.
- Finland's September 1, 2026 Mikkeli campus adds 165MW contracted capacity.
What critics are saying
- Data-center shortages force Cerebras to rent back customer systems, crushing near-term margins.
- The August 2026 copyright class action in N.D. California targets model-training practices.
- If power builds slip, Cerebras misses 2027 delivery and the backlog becomes stranded paper.
What makes Cerebras unique
- Cerebras' wafer-scale chips still beat cluster-based AI systems for low-latency inference and training.
- OpenAI signed a 750MW deal in December 2025, validating Cerebras' architecture.
- AWS Bedrock integration and AMD disaggregation extend Cerebras beyond one-off hardware sales.
Help us improve and share your feedback! Did you find this helpful?
Benefits
Professional Development Budget
Flexible Work Hours
Remote Work Options
401(k) Company Match
401(k) Retirement Plan
Mental Health Support
Wellness Program
Paid Sick Leave
Paid Holidays
Paid Vacation
Parental Leave
Family Planning Benefits
Fertility Treatment Support
Adoption Assistance
Childcare Support
Elder Care Support
Pet Insurance
Bereavement Leave
Employee Discounts
Company Social Events
Growth & Insights and Company News
6 month growth
-1%1 year growth
-1%2 year growth
0%Cerebras Systems (NASDAQ: CBRS), a lesser-known chipmaker, is drawing attention in the AI boom with a processor built for speed. Its Wafer-Scale Engine puts computing cores, memory, and communication on a single piece of silicon, cutting the need to move data between components. The company targets AI inference, where trained models answer questions, write code, and run tasks for millions of users. Inference is among the fastest-growing areas of the AI market. Cerebras and OpenAI agreed a multi-year deal to deploy 750 megawatts of Cerebras compute capacity, with staged deployment through 2028. In August, Cerebras said it powers the Ultrafast mode for OpenAI's GPT-5.6 Sol, which it says delivers up to 750 output tokens per second and runs up to 14 times faster than standard processing. Cerebras also announced its CS-4 system, which it says is up to twice as fast as its predecessor.
Cerebras Systems holds $25.4 billion in signed contracts yet to be delivered, according to its June 30, 2026 figures. The AI hardware and cloud computing company projects core revenue of $880 million to $890 million for 2026, with expectations to more than triple in 2027. Second-quarter 2026 core revenue reached $209.9 million, up 103% year-over-year, driven primarily by cloud services including OpenAI deployment. The stock trades at 51.1 times sales, compared to 3.1 times for the S&P 500. Data centre capacity remains the key bottleneck. Cerebras has secured over 600 megawatts of capacity through end-2027. The company temporarily rents back some customer-operated systems to meet demand, pressuring gross margins short-term. Management expects AWS integration on Bedrock platform in first quarter 2027, with OpenAI remaining a significant but declining revenue portion.
Cerebras Systems CEO Andrew Feldman will speak at TechCrunch Disrupt 2026 about AI scaling challenges. The company, which raised $5.5 billion in its May IPO, takes a different approach to AI computing through wafer-scale processors rather than conventional chip architectures. Cerebras recently signed a multiyear agreement with OpenAI to deploy 750 megawatts of systems from 2026 through 2028. The company has more than 600 megawatts of data centre capacity live or under contract for delivery by end of 2027 and is increasing manufacturing capacity more than tenfold during 2026. Feldman will discuss compute demand, energy and infrastructure constraints, and potential limits of current AI hardware. The event takes place 13-15 October at Moscone West in San Francisco.
BigBear.ai and Cerebras Systems represent two distinct approaches to the AI market. BigBear.ai provides decision intelligence solutions primarily for US federal and defence agencies, using AI to help navigate supply chains, manage autonomous systems, and enhance cybersecurity. In FY 2025, BigBear.ai generated roughly $127.7 million in revenue, down approximately 19.3% year-over-year. The company reported a net loss of about $293.9 million, yielding a negative 230.2% net margin. Free cash flow was negative $46.3 million. Around 51% of total sales came from customers contributing over 10% of revenue individually. Cerebras Systems builds wafer-scale chips designed for demanding AI workloads. The company serves sectors including medical research, energy, and agentic AI through hardware sales and cloud services. BigBear.ai has 579 employees, many holding high-level security clearances, and maintains a debt-to-equity ratio of 0.0x with a current ratio of approximately 1.8x.
Gimlet Labs and Cerebras Systems announced a collaboration to deliver ultrafast AI inference through Gimlet Cloud. The partnership combines Cerebras' wafer-scale compute with Gimlet's inference technology to achieve speeds of up to 3,000 tokens per second for real-time AI applications. Gimlet Cloud integrates the Cerebras Wafer Scale Engine with GPUs, using advanced disaggregation technology to optimise each phase of inference. The first Cerebras-powered Gimlet Cloud datacentre is expected to launch later this year. The collaboration aims to improve user experience in real-time AI applications such as voice assistants and AI agents, where reduced latency creates more responsive interactions. Gimlet Labs will serve as a launch partner for Cerebras' CS-4 technology, providing developers with production-scale access to the combined infrastructure.