The AI & Market Power was the first Fellowship programme run by the European AI & Society Fund. It brought together top researchers and investigators to closely examine market power dynamics in the current AI ecosystem.
The fellows studied diverse yet complementary topics, including but not limited to investment strategies and dependencies in the value chain of AI start-ups, the use of automated trading platforms in Europe’s energy market, and the potential of open cloud computing. Read more about the Fellows and their research.
Learnings for funders
Here are our learnings on what worked, what could be improved and what other funders should consider when designing and delivering a Fellowship scheme.
1. Fund flexibility, not just proposals
AI technology is evolving rapidly, and so is the global AI market. Every Fellowship project changed shape significantly during its lifespan.
Bostoen surveyed four market models instead of two. Thomas pivoted from a narrow generative-biology focus to broad movement capacity-building across food sovereignty, climate justice, and indigenous rights movements. Boyd-Madsen et al. abandoned their initial algorithmic-collusion hypothesis in favour of a “flexibility rents” frame. Au and Katta had to narrow a “huge scope” mid-project.
Several fellows explicitly credited the funder’s willingness to accommodate pivots as essential to impact. In a fast-moving field, a proposal written 12-18 months before the work is finalised will inevitably become partly obsolete.
The lesson is to fund people and questions, not locked-in proposals, while also ensuring that projects are released in a timely way to ensure policy impact.
2. Budget for time and travel, not just outputs
Multiple Fellows reported that the most impactful parts of their fellowship were not the written outputs but the relationships, presentations, and convenings that the funding made possible.
The funding structure that worked best combined a meaningful time stipend (so the Fellow could treat the project as a core professional commitment rather than a side project) with a flexible travel and dissemination budget.
Grant structures that require fellows to specify all conferences and meetings in advance are poorly suited to the pace at which political opportunities open and close.
3. Invest in data access as a precondition for research
A consistent finding is that several fellows were constrained not by analytical capacity but by access to the datasets they needed, sometimes losing access mid-project.
Future grant design should treat data access as a budget line, including commercial datasets, FOIA costs, conference registration fees (which are often the price of admission to industry intelligence), and subscriptions to paywalled trade publications and analyst reports.
4. Impact planning at the start changes the work
The February 2025 workshop on impact planning appears to have been genuinely formative. Multiple Fellows describe shifting their approach as a result. The workshop framing was important: it did not treat impact as dissemination (publishing and hoping someone reads it) but as a theory of change (who has the power to make the desired change, what would make them act, when is the policy window, who is the messenger).
Future programmes should build this into the grant design from the start, ideally with more time for engagement after the project’s final output has been published.
5. The public and policy discourse is still disconnected from findings
A recurring theme is the gap between the sophistication of the research and where public debate and regulatory understanding currently sit. The gap is not only one of communication but of narrative: public debate remains anchored in framings (e.g. “no abuse of dominance,” “AI national champions”) that the fellowship research has empirically dismantled. Translation work therefore requires not just simplification but active narrative challenge.
Future grants could explicitly fund translation activities: short policy briefs, journalist briefings, CSO-facing visuals, podcast appearances. Some Fellows will be natural communicators; others will need institutional support or co-authors with translation skills. Budgeting for a communications partner or for professional editing and design (as Au and Katta did with Open Tech Fund money for illustrations and typesetting) could be an option.
6. The fellowship credential itself has impact value
Several Fellows noted that “one of the most useful things has been the ability to use the title of a Fellow with the European AI & Society Fund as an opening,” or “important for legitimacy” in survey recruitment. The fellowship affiliation with a recognisable institution gave researchers access to regulators, industry conferences, and respondents that they might not have reached as independent scholars.
This suggests funders should think carefully about the branding and institutional weight of fellowship programmes, and should make it easy for Fellows to continue using the affiliation after the formal period ends (as EAISF encouraged).
7. Movement-building and research are not separate activities
A striking pattern in the reflection sessions is that several Fellows found their largest impact was not in policy corridors but in reframing debates and supporting movements. The December workshop captured this tension: “Are policymakers the right audience? Is there other narrative shaping work that needs to happen?”
Fellows identified untapped allies beyond the usual policy audiences: independent cloud providers (who are rarely consulted but share interests in breaking hyperscaler dominance), affected communities that are now seeing the physical impacts of the AI stack for the first time, and cross-movement coalitions (food sovereignty, climate justice, indigenous rights).
Future programmes could explicitly fund ally-mapping as part of impact planning.
8. Think about what comes after the fellowship
Several projects generated findings that clearly need follow-up but lack a natural funding home. The December workshop’s “Future Directions” discussion was full of concrete next steps that have no funding attached. Future programmes should build in either a bridge-funding mechanism for promising projects that need an additional 6-12 months, or a structured handoff process that connects Fellows with other funders who might support the next phase. The alternative is that promising research lines die at the point of maximum potential.
9. The cohort model works, but needs more connective tissue
Both workshop sessions produced cross-cutting insights that individual projects would not have surfaced alone: the observation that cloud concentration appears even in non-cloud projects, that military and biology are state-underwritten domains, that sovereignty framings are being captured. Fellows valued the cohort. One Fellow said that the fellowship “allowed me to build relationships and learn from those who are active in digital fights.”
Others described it as “an invaluable opportunity to identify key areas of overlap.” However, several cross-cutting themes only became visible at the December final workshop. Future programmes could invest in more frequent (even informal) touchpoints between Fellows, joint working sessions, and structured opportunities for Fellows working on related questions to collaborate or co-author.
The AI & Market Power Fellowships were supported by the Open Society Foundations and administered by the Institute of International Education. The European AI & Society Fund was responsible for the selection of the fellowships and for hosting the fellowships community.
Find out more about the Fellows and their findings on AI and market power.