Airon.ai — next-generation AI infrastructure
Be part of building the systems that power the next generation of AI.
Why Airon, why now
Every wave of AI capability rests on the infrastructure beneath it. Airon is building that foundational layer — the compute and capacity that the world's most demanding AI workloads run on.
We are growing fast, working with sophisticated international enterprise customers and building the infrastructure required to support increasingly large and demanding AI workloads. This creates engineering challenges that are complex and evolving in real time.
This is a rare window to join. The market is being defined as we build, the technical decisions matter, and the engineers who join us now will have a real opportunity to shape our technology, architecture and engineering culture from the ground up.
Who we are
A tight-knit, high-energy team where people care about what they build, help each other out and have fun along the way.
We build, own and operate our own datacenters in Sweden, running on renewable power, and sell dedicated bare-metal NVIDIA GPU compute to enterprise customers. Everything from the physical facilities to the software platform is ours, which is why the engineering here is unusually broad.
The role
As Systems Engineer, Inference, you will own how open-weight models actually run on our hardware — from GPU kernels up to the gateway in front of them.
You'll be building on our open-source serving stack for open-weight large language models: production-grade inference that scales across nodes automatically, with day-zero support for what ships next week.
You will work hands-on with vLLM and the layers above and below it, taking ownership from profiling and kernel work through deployment, service levels and cost per million tokens. You will not simply execute against predefined specifications; you will help identify the right problems to solve and influence how we build as Airon scales.
You will work closely with our platform, network and operations teams, and with the customers running models on our infrastructure.
What you'll do
- Own solutions end-to-end — from architecture and implementation to deployment, monitoring and continuous improvement.
- Contribute to technical decisions and help shape our architecture, engineering principles and ways of working.
- Improve the reliability, performance and scalability of our technology as the business grows.
- Use AI and modern engineering tools to increase development speed, quality and operational efficiency.
- Own the serving path for large language models in production, from GPU kernels up to the gateway that fronts them.
- Profile inference workloads end to end: prefill vs decode behavior, KV cache growth, memory bandwidth limits, SM occupancy, and arithmetic intensity (FLOPs per byte).
- Write and optimize kernels, cut launch overhead with CUDA graphs, and prove the gains in Nsight.
- Tune and extend serving engines, mostly vLLM: continuous batching, paged KV cache, scheduler and chunked-prefill policy, and the throughput/latency tradeoff behind them.
- Cut memory and cost through prefix caching, KV quantization and offloading, and weight quantization, while measuring quality regression honestly against the speed gained.
- Ship latency wins with speculative decoding and the right tensor/pipeline parallelism split for each model and cluster.
- Scale inference across nodes: NVLink within the node, Spectrum-X Ethernet across them, disaggregated prefill/decode, MoE expert parallelism.
- Keep it reliable: gateway routing and fallback chains, rate limits, TTFT and ITL SLOs, observability, chaos testing.
- Run the economics: queue-based autoscaling, cold start mitigation, spot GPU strategy, cost per million tokens, and MFU.
- Publish benchmarks and writeups that others can reproduce.
We believe you
- Have taken an LLM serving stack into production and kept it there — not only benchmarked one.
- Are fluent in the CUDA execution model, and can read a profile and act on what it actually says rather than on what you expected.
- Know vLLM, SGLang, TensorRT-LLM or a comparable engine well enough to change its behaviour, not only to configure it.
- Understand distributed inference — tensor and pipeline parallelism, MoE expert parallelism — and what the interconnect does to each of them.
- Measure honestly, including when a quantization win costs more quality than it is worth.
- Are comfortable owning something in production: service levels, observability and the reliability work that comes with real customers.
- Are comfortable taking work from problem definition to a reliable, working solution.
- Can make thoughtful trade-offs between speed, scalability, reliability, security and technical complexity.
- Are comfortable working hands-on in an international, fast-moving environment.
- Have a genuine interest in using AI and modern engineering tools to improve how software and infrastructure are built.
Bonus points for
- Kernel work in CUDA, Triton or CUTLASS.
- Upstream contributions to vLLM, SGLang or other open-source serving projects.
- Speculative decoding, prefix caching or KV offloading shipped in production rather than in a notebook.
- Experience with NVLink and high-performance Ethernet fabrics such as Spectrum-X or RoCE.
- Published benchmarks, writeups or talks we can read.
- Experience from AI infrastructure, cloud infrastructure, data centers, distributed systems, high-performance computing or another relevant technical environment.
- Experience working in an early-stage, high-growth or rapidly changing company.
- Experience building systems that support enterprise customers or business-critical workloads.
Who you are
- Curious and technically ambitious, with a genuine interest in understanding how things work.
- Low-ego and action-oriented, with a strong sense of ownership.
- Analytical and pragmatic, with the ability to turn complex problems into working solutions.
- Comfortable navigating ambiguity.
- Collaborative and direct, with the ability to communicate technical decisions clearly.
- Motivated by solving difficult problems and building technology that matters.
What's in it for you
A chance to build the foundational infrastructure behind the next generation of AI — while the market and the technology are still taking shape.
You will work on complex, real-world engineering challenges with significant room to influence architecture, technical direction and how Airon builds. You will join early enough to make a visible impact, take meaningful ownership and help shape the engineering culture as the company scales.
If you have been looking for an engineering role where the problems are genuinely difficult, the pace is high and what you build has a direct impact — this is it.
Practical
- Location: Remote, or Lidköping or Gothenburg if you would rather sit in an office. Occasional travel to Lidköping.
- We ask all candidates to consent to a background check before starting with us.
About your application
Send your application to career@airon.ai. Tell us a little about yourself, what you are great at and what you would like to build with us.
Please include your GitHub profile. If you have published benchmarks, writeups or upstream contributions, link them — we will read them.
We know that great candidates do not always match every requirement on paper. If this opportunity excites you and you believe you could make an impact at Airon, we would still like to hear from you.
Ready to join us?
Help us build the infrastructure that powers the next generation of AI.