by JL Ecochard
In April 2026 I facilitated a session at the Good Tech Summit in Washington DC. Ninety minutes. Five tables. Fifty practitioners, people actually building and using AI tools in the field. No panels. No keynotes. Three real case studies and honest conversation.
This is not a summary of that session. It is what the room built together, with my commentary added where I have something useful to say.
They advanced thinking on three themes and I promised to share it back. In fact, their insights were so profound that I have not been able to stop thinking about them and that led to the fourth blog. Each theme is a piece in this series.
None of these are comfortable. All of them are actionable.
In 2024, DEEP went dark.
DEEP was a humanitarian data platform that used AI to surface patterns across 40 different information sources (e.g., situation reports, field assessments, news feeds) and turn them into actionable intelligence for crisis response. It worked. People built workflows around it. Organizations depended on it.
Then the grant ended. And DEEP disappeared.
Not because the technology failed. Not because the need disappeared. Because nobody budgeted for what happens after the demo.
This is not a sustainability problem. It is a structural indictment of how the humanitarian sector finances digital infrastructure. We funded DEEP like a meal delivery program, project by project, grant by grant, when what it was building was more like a water system. You do not fund a water treatment plant for 18 months and then wonder why the water stopped. You do not build a road to a community and let it crumble because the project budget expired.
Infrastructure is not a project. It is a commitment.
I put this case to 50 humanitarian practitioners at the Good Tech Summit in Washington DC, five tables, all field experience, no observers. The room recognized itself immediately. Because DEEP is not unique. It is the pattern.
Population mapping tools losing their maintenance teams. Chatbots that cannot sustain localization once the pilot funding ends. Early warning systems that show real promise and then vanish. The room had seen all of it. Lived all of it. Built some of it.
But what came back from the tables was not just diagnosis. It was design. These practitioners had been burned enough times to be actively reaching for alternatives.
Co-investment models between funders and organizations, building longer-term business models that lead to actual sustainability rather than perpetual re-application. Pro-bono contributions from tech industry experts to reduce the cost of maintenance without depending on grant cycles. Public-private synergy applied not just to connectivity but to compute. Software alliances that spread the cost so that at least one tool in each critical category survives the funding cliff. Local ownership structures that give communities and regional organizations a stake in the infrastructure they depend on, not just access to it.
And the sharpest question the room raised: someone makes money from humanitarian AI infrastructure. The SaaS vendors. The cloud providers. The consultants. The tech companies whose tools get embedded in NGO workflows. If you are extracting value from this ecosystem, perhaps you should be funding it as infrastructure. The cost of sustainability belongs with those who accrue the economic benefit, not with the organizations least able to afford it.
The AI gap in our sector does not just run between advanced and lagging organizations. It runs along the entire aid delivery chain.
A large INGO at 90% AI fluency cannot implement effectively if the local community-based organization at the last mile is at 5%. We do not have a tiered product. We have a chain. Chains break at the weakest link.
We saw this with digital transformation. Rich organizations transformed. The end of the supply chain did not. The result was not sector-wide transformation. It was a capability gap that throttled the whole system. We cannot repeat that with AI.
There is also a hardware dimension almost nobody is talking about. Right now I run large language models locally on my MacBook. No cloud. No data leaving the device. That is real today, for me, with current hardware, in the Global North. I am indeed privileged.
In Zambia, DRC, Bangladesh, Colombia: laptops in LMIC NGO contexts tend to be scarce and low-spec purchased for low cost, good enough to run modern workplace tools. A commodity not designed as infrastructure. And here is the critical difference from the last infrastructure challenge that was high speed internet access. You cannot share GPU and VRAM the way you shared bandwidth. One fast internet connection served many users, slowly, but it worked. A billion-parameter model either runs or it does not. Partial resources do not produce partial capability. They produce nothing.
NetHope was founded in 2001 to democratize access to connectivity. Broadband was the infrastructure that determined who could participate in the digital world, and market forces alone were never going to close that gap. GPU and VRAM are the new bandwidth. The next chapter is to democratize access to compute, regional shared inference infrastructure, hotAIspots, that give local NGOs and CBOs the processing power they may not be able to afford individually.
If you are a funder: build operational AI costs into your grants. A pilot without a runway is a pilot that will crash. Ask whether the most useful thing you can fund right now is not a new pilot, but year three of something that already works. Ask who is extracting value from the infrastructure your grantees depend on and whether they are contributing to its sustainability.
If you are deploying: document what you build and share it, including the fail rates. The sector is not starved for success stories. It is starved for honest ones. Find your partners at the last mile/km before you build, not after. Built with not for.
If you are both: stop funding digital infrastructure like meals.
Fund it like roads.
JL Ecochard is Chief Innovation Officer at NetHope and co-founder of its Center for the Digital Nonprofit.
This is Part 1 of a four-part series developed from his session “When Every Hour Counts” started at the Good Tech Summit, April 2026. Part 2: Nobody Goes to Jail. Part 3: Local. Part 4: The Brain Drug Argument.
