100 AI DIFFUSION PATHWAYS BY 2030

AI capability is no longer scarce. Frontier models can reason, translate, summarise, and infer across virtually every domain. And yet, for most people — especially those at the base of the pyramid — AI remains invisible. The crisis is not invention. The crisis is implementation.

Consider a farmer in Maharashtra. Even with access to a frontier model, the interaction would likely default to English; the advice would not be grounded in local crop data or soil conditions; there would be no institutional accountability behind the answer. Research across hundreds of AI deployments in agriculture, health, education, and governance found a consistent pattern: technology accounts for roughly 30% of what reaching scale requires. The other 70% — data readiness, language support, workforce adaptation, accountability guardrails, and organisation-wide alignment — is what actually determines whether a deployment reaches people or stalls.

There are two races underway. The race to the bottom is driven by private capital maximising returns through engagement, advertising, and extraction. It requires no coordination and it is fast. The race to the top requires deliberate institutional effort to route AI’s gains to the many rather than concentrating them among the few. 100 AI Diffusion Pathways by 2030 is the counter-architecture: demonstrating in practice that AI can scale upward in capability while scaling outward in benefit.

The deeper problem is the learning cost. Jeffrey Ding’s research on general-purpose technologies shows that the electric dynamo took fifty years to produce widespread economic gains — not because the technology was unavailable, but because adopters independently rediscovered how to organise around it. Without pathway infrastructure, every new AI adopter starts from zero, independently rediscovering architecture patterns, safety frameworks, data integrations, and coordination models. Diffusion pathways make learning cumulative rather than repetitive.

The proof is already visible. MahaVISTAAR (Maharashtra’s agricultural AI system) took nine months to deploy from scratch. Ethiopia’s ATI drew on that pathway and was operational in three months. Amul’s Sarlaben built on two full deployment cycles and was running in three weeks. That compression — from months to weeks — is what structured pathways make possible. At one hundred pathways, diffusion becomes systematic.

A pathway is a structured, reusable record of how AI was deployed at scale in a specific context. It has two mandatory components: a deployment playbook (the step-by-step operational guide) and a reusable toolkit (the assets, code, benchmarks, and safety guardrails that the next deployer can use directly). A pathway compresses the learning time, cost, and risk for the next adopter.

A pathway is not a case study, a white paper, a technology specification, or a best-practice guide. A case study tells you what an organisation did; a pathway tells you what you would need to do. If it does not make another adopter’s AI implementation journey easier, it is not a pathway.

MahaVISTAAR (Maharashtra’s agricultural AI system) is the most fully documented pathway to date. Ethiopia’s ATI deployment drew directly from this pathway and was operational within three months. Amul’s Sarlaben built on two complete deployment cycles and was running within three weeks. Additional pathways are being actively documented and will be made available through the initiative’s shared repository.

The target means 100 distinct, validated deployment experiences documented to the two-component standard, spanning different sectors, geographies, and organisational contexts. The number reflects the scale of diversity in real-world AI deployment — across agriculture, health, education, governance, and other domains. A small set of well-resourced deployments, however thoroughly documented, would not satisfy the target. This number is also such that we can bend and accelerate the arc of AI towards those whose lives it can transform the most.

A pathway counts toward the target when it is sufficiently complete for another organisation to use as a genuine starting point. Progress will be reported publicly, with the shared repository serving as the primary record.

Safety is a structural element of every pathway, not a standalone section. The reusable toolkit that each pathway must include contains safety guardrails: evaluation benchmarks for harm detection, documented failure modes and how they were identified and addressed, accountability mechanisms, and criteria for when a deployment should be paused or reviewed. Safety work done in one deployment context travels to the next — a new deployer inherits the safety checkpoints developed through prior deployments, and adds to them.

Pathways serve three primary audiences. First, deploying organisations — governments, cooperatives, ministries, and agencies that want to deploy AI at population scale and need a starting point grounded in real experience. Second, builders and orchestrators — organisations that have deployed or helped deploying organisations deploy AI at scale and can document their experience for others. Third, enabling organisations — those with expertise in AI safety, data governance, language support, or workforce adaptation who can strengthen existing pathways.

There are four contribution modes. If you have deployed AI at scale, you can Illustrate: document the lived experience and create a pathway. If you are ready to deploy, you can Use: adopt an existing pathway to reduce your learning cost and time — with the responsibility to pay it forward by strengthening the pathway or documenting your own experience. If your expertise is in AI safety, data governance, or workforce adaptation, you can Strengthen: improve existing pathways. If you see a sector or community with no pathway yet, you can Build one from scratch. To connect, reach out to any of the eleven founding organisations or to GIS, which serves as the secretariat.

The answer depends on who you are:

For a deploying government or institution: you gain access to proven deployment playbooks and reusable toolkits that compress your timeline from years to months.

For a technology company or frontier lab: your tools get used in real-world deployments at population scale, building evidence, increasing adoption, and demonstrating responsible use.

For a funder or development organisation: you invest in a declining marginal cost curve — each successive batch of pathways costs less because shared infrastructure accumulates. Your funding builds permanent public goods, not one-off projects.

For a knowledge partner or think tank: you gain access to the most detailed, operationally grounded documentation of AI deployment at scale that exists anywhere.

The primary ask is knowledge and time. The four contribution modes — Illustrate, Use, Strengthen, Build — are principally about documenting deployment experience, using and improving existing pathways, and building new ones where none exist. None of these require a financial commitment to participate.

Pathways can spread along sector lines (agriculture to agriculture across countries) or along institutional lines (state government to state government across sectors). Both patterns are already visible. The six-dimension framework makes cross-sector transfer possible because the non-technology dimensions — institutional transformation, ecosystem design, workforce absorption, operating model — are not sector-specific. The sector-specific parts sit primarily in the technology shifts, while the 70% non-technology dimensions transfer more readily across sectors.