Liquid AI Business Models

This PhD aims to study techno-economic and business impact of distributed AI compute.

AI is the most disruptive technology in the domains covered by HIPEAC in the last decades. Apart from boosting chip development companies like NVIDIA to extreme valuations, it made model developers like OpenAi and Anthropic investing massively in AI data centers, trying to stay ahead of the competition. This led to a range of issues in terms of energy consumption, availably of components, reliability, etc. As an alternative to the systematic request for cloud services, Europe (HiPEAC report) suggests the Next Computing Paradigm (NCP) based on distributed computing on demand, where AI compute shifts from big, centralized data centers towards more distributed "liquid" setups, running on edge devices, on-premise, or even orchestrated across many owners' devices. In the light of this expected shift, the central research question of this PhD is “Which business models make the shift towards liquid AI economically viable?”

As this overarching question should tackled from different angles, the following three sub-questions will be studied.

(1) What drives the cost of the different AI architectures? This question takes the perspective of the AI provider and aims to understand their cost model. It investigates how the costs of AI compare across the spectrum: central cloud vs. edge vs. liquid, big and general-purpose LLMs versus small, task-specialized SAM (small and specialized models). It takes into account the moving importance from training to inference and consider investment as well as operational costs.

(2) What are viable revenue models for AI usage? This question takes the perspective of the AI user and compares the current default of token base pricing to subscription, infrastructure/utility, hardware-as-a-service, compute-sharing marketplaces. What is the impact of the different pricing models on the business model for the AI users? How does it impact AI driven companies' strategic choices like cloud versus self-hosted AI solutions?

(3) What is the impact on the business models for different actors across AI stack? How does value creation and value capture shift along the value chain (model providers, chip/device makers, hyper-scalers, orchestration layers)? Which models are economically viable for the AI providers (trade-off between costs and revenues) so that we get to an economically viable solution when token subsidization stops?

This PhD will develop a novel techno-economic methodology that combines bottom-up cost modeling and multi-actor value-network analysis. The liquid compute exchange is modelled as a two-sided market and the orchestrator can be a distinct value-capture layer.

A suitable PhD candidate combines expertise from both technical and economic perspectives: either an economist/business engineer with an additional background and interest in IT/AI, or a computer scientist with an additional background and interest in economics.

reference: Duranton, M., Blondelle, G., De Bosschere, K., Carpenter, P., Gray, M., Goubier, T., Nackaerts, A., Robinson, C., Vardanega, T., & Zendra, O. (Eds.). (2026). HiPEAC Vision 2026. HiPEAC. ISBN 978-90-78427-08-7.