PhD Stipend in Energy-Efficient Agentic AI: Measuring When AI Is Worth Its Energy -DEIS
The Technical Faculty of IT and Design, Department of Computer Science, Section for Distributed, Embedded and Intelligent Systems
Get a reply from this employer
Your resume and cover letter, tailored to exactly what they’re hiring for.
Benefits offered by this job
Job summary
The Technical Faculty of IT and Design, Aalborg University, invites applications for a PhD stipend focused on measuring and reducing the energy cost of agentic AI. The position is three years with an open start and involves collaboration across computer science and business school.
You will run experiments, quantify energy per task, and develop evidence-based comparisons between agentic AI and traditional software, contributing to open-source benchmarks and policy-relevant findings.
Qualifications
- Master's degree required in CS/SE or related field.
- Strong interest in efficiency and sustainability of computing systems.
- Experience with one or more of LLMs, agent frameworks, energy or performance measurement is an advantage.
Responsibilities
- Define and conduct experiments to measure energy usage of agentic AI.
- Develop datasets and methods to compare energy cost of agentic AI vs deterministic software.
- Collaborate with social scientists to frame industrial relevance and dissemination of results.
- Publish in international venues and contribute to open-source measurement tooling.
Skills
Education
Tools
Job description
At the Technical Faculty of IT and Design, the Department of Computer Science at Aalborg University invites applications for a PhD stipend on measuring and reducing the energy cost of agentic AI. The stipend is open for appointment from 1 March 2027 or soon hereafter. The duration of the position is three years.
The position is part of the project Energy, Sovereignty and the Responsible Use of Agentic AI, funded by the Independent Research Fund Denmark (DFF) under its thematic programme on artificial intelligence. The project is a collaboration between the Department of Computer Science and Aalborg University Business School. A second PhD student, recruited at the Business School, studies how organisations decide on, govern and maintain their use of AI; the two PhD students are trained together as a paired cohort.
Every call to a large language model spends energy, and an AI agent calls a model many times to finish a single task: it plans, uses tools, retries and checks its own results. Early measurements suggest that agentic coding uses on the order of 1000 times more tokens than ordinary chat, and that much of the energy goes into failed and retried steps. As organisations move from chat to agents, AI becomes a recurring operational cost per task rather than a fixed background cost.
The project Energy, Sovereignty and the Responsible Use of Agentic AI asks when the use of AI is worth its energy cost, and who is able to judge this. The PhD stipend covers the engineering side of the project. You will build the evidence base: how much energy agentic AI actually spends per successfully completed task, and under which conditions cheaper alternatives do the same job.
A central comparison is between calling an agent every time a task recurs and having AI generate conventional, deterministic software once. Generated software costs energy to produce but almost nothing to run; an agent invocation costs the same every time. The project seeks to measure the break-even point between the two, and how it depends on the model, the framework and the stability of the task.
Within this frame, you develop your own PhD project. Indicative questions include:
- How much energy do LLM and agentic-coding workloads consume per successful task, for frontier models and for small, open-weight models, and how is that energy split between planning, retries and failed steps?
- When is it energy-rational to call a model or agent at runtime, and when is it better to generate a deterministic program once? Where does the break-even lie in practice?
- How different DevOps practices benefit or penalize the energy cost of running the agents?
- How much do levers such as prompt caching, model cascades, small-model routing and distillation reduce energy per successful task, and at what cost in quality?
- How can the energy use of models accessed only through an API be estimated reliably, and validated against direct measurements on local hardware?
- How do changes in providers' models affect the energy cost and the stability of AI-based systems over time, and how portable are agent setups across providers?
The work is experimental. You will run agent frameworks on curated task sets, and measure energy directly on hardware (for example Intel RAPL, performance counters and rack-level power meters) for locally run models. For frontier models you will develop and validate telemetry- and FLOP-based estimates. You will have access to the energy‑measurement testbeds and tooling developed in the ITEA4 projects GreenCode and MAST, on‑premise GPU/CPU infrastructure, and AAU's CLAAUDIA cloud and HPC services.
The PhD project is carried out in close collaboration with the social‑science researchers at the Business School. Your measurements inform their study of how organisations decide on, procure and maintain AI, and their findings help frame which comparisons matter. You will take part in shared project seminars, undertake a short secondment into the organisational work package, and contribute to pilots with Danish public and private organisations, drawing on the GreenCode and MAST industrial partners.
The project follows open‑science practice: protocols are pre‑registered, null and negative results are reported, and the measurement harness and benchmarks are released as open source and contributed to community efforts such as the Hugging Face AI Energy Score.
As a PhD student you will complete PhD courses, gain experience with teaching or other forms of knowledge dissemination, and take part in the research environment at the Department of Computer Science. You will publish your research at international venues in software engineering and sustainable computing, such as HotCarbon, ICSE, FSE, MSR, IEEE Transactions on Software Engineering and Empirical Software Engineering.
You hold a Master's degree in computer science or software engineering, or a related field.
We are looking for a candidate with a strong interest in the efficiency and sustainability of computing systems, who enjoys designing careful experiments and wants to find out what AI systems actually cost when they run. You should also be curious about how organisations use AI, and willing to work closely with social scientists. We are particularly interested in candidates who can formulate a clear, original and feasible research problem, and present a realistic plan for the experiments needed to address it.
Solid programming skills (for example Python) and experience with Linux-based systems are expected. Experience with one or more of the following is an advantage: LLMs and agent frameworks, energy or performance measurement, benchmarking, empirical software engineering, distributed or embedded systems, and GPU or HPC environments. You communicate clearly in English. Danish is not a requirement.
PhD stipends are allocated to individuals who hold a Master's degree. PhD stipends are normally for a period of 3 years. It is a prerequisite for allocation of the stipend that the candidate will be enrolled as a PhD student at the Technical Doctoral School of IT and Design in accordance with the regulations of Ministerial Order No. 872 of 6 October 2026 on the PhD Programme at the Universities and Certain Higher Artistic Educational Institutions. According to the Ministerial Order, the progress of the PhD student shall be assessed at regular points in time. As part of the PhD study, you are among other things required to complete PhD courses corresponding to 30 ECTS, gain experience with teaching or other forms of knowledge dissemination and complete an external research stay outside of Aalborg University, preferably 3-6 months at a foreign research institution.
Aalborg University wants to reflect the surrounding society and has diversity as a core value. Therefore, everyone, regardless of personal background and orientation, is encouraged to apply for the position.
If you have any questions about the position, you are more than welcome to contact us. You will find contact persons at the bottom of the jobpost.
Read more about our recruitment process here.
The assessment of candidates for the position will be carried out by qualified experts. Shortlisting will be applied. This means that after the application deadline, the head of the department, with the assistance of the hiring committee, will select the applicants to be assessed. All applicants will be informed whether they have been shortlisted for assessment or not.
The hiring process at Aalborg University may include a risk assessment as a tool to identify potential risks associated with new hires, ensuring the safety, compliance, and integrity of the workplace.
Read more about The Technical Doctoral School of IT and Design
The employment is in accordance with the Ministerial Order on the Appointment of Academic Staff at Universities and the Ministerial Order on Job Structure for Academic Staff at Universities. Furthermore, your terms of employment and salary are in accordance with the collective agreement between the Danish Confederation of Professional Associations and the state (AC collective agreement).
There is a mutual probationary period of 3 months for the position.
Aalborg University - Knowledge for the world
Type of employment
Contract type
Contract type Full time
Number of positions
Number of positions 1
Full-time equivalent 37
City
City Aalborg
County *
Country
Reference number
Reference number 2026/1289
Contact
'/%3e%3cpath%20d='M101.744%2012.4726L102.894%2016.1947L99.0021%2013.2201L101.744%2012.4726ZM105.3%208.40937H100.488L99.0026%203.59473L97.5116%208.41047L92.7%208.40385L96.5965%2011.3834L95.1055%2016.1942L99.0021%2013.2196L101.408%2011.3834L105.3%208.40992V8.40937Z'%20fill='white'/%3e%3cdefs%3e%3clinearGradient%20id='paint0_linear_5593_1381'%20x1='90'%20y1='9.44471'%20x2='99'%20y2='9.44471'%20gradientUnits='userSpaceOnUse'%3e%3cstop%20stop-color='%2300AD70'/%3e%3cstop%20offset='0.998594'%20stop-color='%2300B67A'/%3e%3cstop%20offset='1'%20stop-color='%23DCDCE6'/%3e%3c/linearGradient%3e%3c/defs%3e%3c/svg%3e){kind=small}
