The next billion AI index

The compass for AI utility and adoption in the global majority.

Featured in
The Hindu AI Alliance India AI Impact Summit

"Within the community, there's no consensus on what works for us...the index would help us as a community figure out what's best for us, beyond just building policy documents."

Data scientist, Kenya

What is nexbax?

The Next Billion AI Index (nexbax) addresses the measurement blind spot in next billion markets, in which AI progress cannot be equated with benchmark supremacy. The index evaluates generative AI systems—including models and agentic architectures—not by performance alone, but by their capacity to deliver economically sustainable, inclusive, and developmentally relevant value across constrained environments.

Inspired by the precedent set by Stanford's Foundation Model Transparency Index, nexbax extends that momentum toward inclusivity, accessibility, and innovation for the next billion users, asking not only how powerful systems are, but how meaningfully they serve the global majority.

Nexbax is comprised of 10 dimensions organised under 3 themes: Effective Efficiency, Operational Practicality, and Societal Integrity.

Effective Efficiency Can the system deliver maximum value per unit cost and resource?
Cost Effectiveness and Price PerformanceDoes the system deliver strong performance relative to its total cost of ownership (usage, infrastructure, development, and deployment) for the target use case?
Resource EfficiencyHow efficiently does the system use compute, energy, bandwidth, and hardware resources to perform the task?
Operational Practicality Can the system be adapted and sustained in heterogenous real-world settings?
Adaptability and CustomizabilityHow easily can the system be modified, fine-tuned, or extended to support new tasks, domains, or modalities?
InteroperabilityHow easily does the system integrate with other tools, data sources, and infrastructure through standards, APIs, and plug-and-play components?
Resilience, Reliability, and RobustnessHow reliably does the system operate under real-world conditions such as poor connectivity, noisy inputs, or infrastructure instability?
Usability and AutomationHow easy is the system to deploy and operate, including availability of automation, templates, or low-code workflows?
Education and EmpowermentIs there sufficient documentation, learning resources, and community support to help users build and maintain solutions?
Societal Integrity Does the system respect, represent, and include local values and does it allow for shared stewardship?
Trustworthiness and EthicsHow transparently and responsibly does the system handle safety, security, privacy, bias, data governance, and user control?
Multiculturalism, Inclusivity, and PluralismHow well does the system support diverse languages, cultures, and accessibility needs across global user populations?
Openness and CollaborationHow open is the technology to community participation through open standards, shared resources, and collaborative development?

What do the next billion users have to say?

We tested nexbax with eleven experts, which provided early signal on its efficacy and usefulness in capturing deployment considerations.

11 Experts interviewed
  • Founders / CTOs
  • Developers / Engineers
  • Product / Analytics leads
4 Countries represented
  • India
  • Kenya
  • Ghana
  • United States
3 Top dimensions cited
  • Cost effectiveness
  • Usability & automation
  • Trustworthiness & ethics
"All the tools that we use to evaluate our customers beyond localization, you hit all of them, ease of use, reliability, the cost, the set up time. Those are the same types of things that we use to pick our vendors."Product manager, India

How can nexbax be used in practice?

1 Define rubrics per dimension

Each dimension gets a ladder, e.g. weak, moderate, strong or feasible, potential, proven. Applied to developer facing system properties in this first version.

  • Weake.g., no fine-tuning, cloud-only, English defaults
  • Moderatee.g., some adapters, major languages, basic safety docs
  • Stronge.g., modular, edge-viable, community governance
2 Build on existing benchmarks

Where they exist, established benchmarks inform dimensions. Where they don't, proxy evidence and practitioner judgment fill the gap.

  • Resource efficiency: MLPerf, AI Energy Score, Intelligence per Watt
  • Multiculturalism and inclusivity: FLORES-200, MMTEB, low-resource benchmarks
  • Trustworthiness: MLCommons AILuminate, DecodingTrust
  • Adaptability: Proxy: fine-tuning APIs, adapter ecosystems
3 Co-design with local stakeholders

For the global majority, usefulness is context-dependent. Rubrics must be defined and adapted with local communities.

  • Choose which dimensions matter most for your use case
  • Validate what "strong" means in your deployment context
  • Surface trade-offs that are invisible from the outside
  • Develop a vocabulary to reason together

Next steps

We are continuing to improve nexbax and welcome your feedback.

Ambrish Rawat
Ambrish Rawat
IBM Research
Jessica He
Jessica He
IBM Research
Subhabrata Majumdar
Subhabrata Majumdar
Indian Institute of Management Bangalore
Claudio Pinhanez
Claudio Pinhanez
University of São Paulo
Yann Le Beux
Yann Le Beux
YUX Design
Satyapriya Krishna
Satyapriya Krishna
Harvard University
Rahul Gupta
Rahul Gupta
Amazon
Rumman Chowdhury
Rumman Chowdhury
Humane Intelligence
Hiwot Tadesse
Hiwot Tadesse
Harvard University
Kush Varshney
Kush Varshney
IBM Research

Acknowledgments

We thank Varun Aggarwal, Anthony Annunziata, Ginette Azcona, Daksh Chawla, Nahuel Defosse, Giridhar Ganapavarapu, George K. Githae, Sudeep Gowrishankar, Raghav Gupta, Jaikrishnan Hari, Ronak Khandelwal, Jordan McAfoose, Rohan Sahu, Siddhant Sachdeva, Het Shah, and Amith Singhee for their assistance with the survey process, feedback, and other contributions during the development of this work. Subhabrata Majumdar's research is supported by the Indian Institute of Management Bangalore Young Faculty Research Grant.

If you use nexbax in your work, please cite our white paper:

@misc{rawat2026nexbax,
  title        = {Next-Billion {AI} Index: The compass for {AI} utility and adoption in the global majority},
  author       = {Rawat, Ambrish and He, Jessica and Majumdar, Subhabrata and
                  Pinhanez, Claudio and {Le Beux}, Yann and Krishna, Satyapriya and
                  Gupta, Rahul and Chowdhury, Rumman and Varshney, Kush R.},
  year         = {2026},
  month        = may,
  eprint       = {2606.00359},
  archivePrefix = {arXiv},
  primaryClass = {cs.CY},
  doi          = {10.48550/arXiv.2606.00359}
}
Rawat, A., He, J., Majumdar, S., Pinhanez, C., Le Beux, Y., Krishna, S., Gupta, R., Chowdhury, R., & Varshney, K. R. (2026). Next-Billion AI Index: The compass for AI utility and adoption in the global majority. arXiv. https://doi.org/10.48550/arXiv.2606.00359
Rawat, Ambrish, et al. "Next-Billion AI Index: The Compass for AI Utility and Adoption in the Global Majority." arXiv, 29 May 2026, arxiv.org/abs/2606.00359.