Development & the Indo-Pacific
Real money, a policy gap on AI, and governments asking for sovereignty in their own words. A very different buyer from a charity.
Real money, a policy gap on AI, and governments asking for sovereignty in their own words. A very different buyer from a charity.
The opening. There is no DFAT AI-for-development strategy, programme line or tender. A small, credible, Australian pilot that respects Pacific ownership could help shape the approach rather than chase a tender later.
| Australian not-for-profits | International development | |
|---|---|---|
| Who holds the money | The organisation itself | Donors (DFAT, MFAT, JICA, EU), development banks (ADB, World Bank), UN agencies, foundations |
| Who buys | CEO, COO, board | Managing contractors (Palladium, DT Global, Abt, Tetra Tech), INGOs, UN country offices, ministries using donor money |
| Who uses it | Staff | Officials, health workers, teachers, churches, local NGOs |
| Procurement | Relationship-led | Formal: Commonwealth Procurement Rules, contractor manuals, ADB and World Bank rules |
| Timeline | Weeks to months | Design to contract commonly 6–18 months |
| Buying hardware | Painful: tied grants, little capex | Normal. Equipment lines are routine. The hard question is who pays in year four. |
| Also required | Little | Gender and disability inclusion, safeguarding, localisation, climate, fraud and anti-corruption policies |
Managing contractors hold multi-year, multi-hundred-million DFAT "facility" contracts and can fund a pilot inside them without a new tender.
High Commissions and Embassies (Suva, Dili) hold small direct funds. Below A$125K, purchasing is simpler.
The cyber and critical tech programme historically gave up to A$200K per grant. A company usually needs a university or NGO to lead.
A$150m of flexible DFAT money to Australian INGOs, who could build Selma into their Pacific programmes.
UNDP Pacific already convenes AI readiness work. Slow buyers, strong partners.
Typically US$0.5–2m, procured through consultants.
| Line | A$ | Why it's relevant |
|---|---|---|
| Fiji digital package (Vuvale Union, announced 30 Aug 2026) | FJ$146.3m | Data storage, digital skills and cyber security for Fiji. Probably the money behind the Fiji hub; bought through DFAT's contractor or the Fiji Government Secondary source |
| REnew Pacific (off-grid renewable energy) | 75m | A natural partner for solar-powered sites |
| ANCP (Australian NGO Cooperation Program) | 150m | Flexible money through Australian INGOs |
| Pacific Church Partnerships | n/a | Seven PNG churches reaching ~70% of the population; they run many rural clinics and schools |
| Australian Volunteers Program | 31m | Could place local "AI custodians" to support sites |
| Export Market Development Grants (for EthicAI) | up to 80K/yr | 1:1 matching for Pacific and Timor-Leste marketing and travel |
| AIFFP (infrastructure financing) | +550m | Climate-resilient and digital infrastructure |
| SEA-PAC Cyber (was Cyber and Critical Tech Cooperation) | 83m since 2016 | "Critical technology" is in scope |
| Type | Examples | Readiness | Why |
|---|---|---|---|
| Managing contractors | Palladium, DT Global, Abt Global, Tetra Tech | High | Can buy inside existing budgets. Also have their own sensitive data (M&E, safeguarding, reporting). |
| Catalpa International | Timor-Leste-founded digital developer | High, as partner | DFAT partner on maternal mHealth; co-wrote UNESCO's 2025 AI readiness assessment for Timor-Leste. |
| Australian INGOs | World Vision, Save the Children, Oxfam, CARE, Plan, Caritas, Red Cross | Medium-high | Look like our NFP customers at home. Pacific offices hold child protection and GBV case data. |
| Red Cross / IFRC Pacific | Suva office | Medium | Cyclones cut connectivity; strict data protection rules. |
| Churches | PNG mainline churches | Medium | Trusted, physical sites, new DFAT funding. |
| UN agencies | UNDP Pacific, UNICEF, UNESCO | Medium | Already doing AI readiness in Tonga, Fiji, Samoa, Vanuatu. |
| Regional bodies | Pacific Islands Forum Secretariat, SPC, USP | Medium | PIFS owns the new AI principles; SPC holds sensitive regional datasets. |
| Partner governments | Fiji Ministry of Communications, PNG DICT, Timor-Leste | Medium, slow | Strongest ownership signal; funded by donors in practice. |
| Development banks | ADB, World Bank | Low, short term | Large digital public infrastructure programmes; Selma only as a sub-component. |
| Country | Online | Grid electricity | 1GB mobile data (US$) | AI and data policy | Starlink |
|---|---|---|---|---|---|
| Papua New Guinea | ~24% | 21% | 1.36 | Most explicit sovereignty language: draft sovereign AI framework, National AI Register tracking "offshore data processing" | Live May 2026 |
| Fiji | 79% | 99% | 0.09 | National Digital Strategy 2025-30, AI Hub (2024), data protection and cloud policy due 2026 | Yes |
| Timor-Leste | 40% | ~100% (reliability varies) | 1.92 | Timor Digital 2032; joined ASEAN Oct 2025; Malaysian firm building a "national sovereign AI cloud" | Rolling out to 450 villages |
| Solomon Islands | 43% | 81% | 6.96 | No AI strategy found | Yes |
| Vanuatu | 46% | 70% | 1.54 | UNDP readiness assessment done | Resellers |
| Samoa | 58% | 100% | 0.36 | ICT Policy 2025-30 draft, digital ID | Yes |
| Tonga | 59% | 100% | 1.19 | Building AI tools for citizens in Tongan (UNDP) | Licensed Dec 2024 |
| Kiribati | 88%* | 96% | 3.19 | None found | Yes |
| Tuvalu | 74% | 100% | 6.08 | None found | Yes |
DataReportal Digital 2026, PNG NRI, World Bank electrification data, 2023 data prices (dated). *Kiribati figure looks high; treat with caution.
| Substitute | What it is | Price | How much of a threat |
|---|---|---|---|
| Starlink + ChatGPT / Copilot | Satellite link and cloud AI seats | Starlink kit ~A$665, 50GB ~A$152/month (Fiji). 20 Copilot users + Starlink ≈ A$12.6K/yr | High where data isn't sensitive and links hold |
| Project NOMAD | Open-source offline knowledge server with optional local LLM; recommended hardware nearly identical to Selma | Free | High for the box, none for governance and support |
| RACHEL / RACHEL AI | Offline education server, now with AI search | RACHEL 5 US$500 | Medium, education only |
| SolarSPELL (Arizona State) | Solar offline library, ~600 units, deployed in Vanuatu, Samoa, Tonga, Micronesia; has since built an offline Llama health chatbot (piloted in Africa so far) | Not published | Low. A strong potential partner. |
| Kolibri, Kiwix, Internet-in-a-Box | Offline learning and content | Free | Low; can share a network with Selma |
| Zchwantech | National sovereign AI cloud for Timor-Leste | National scale | Medium at national level |
| Huawei and Chinese vendors | Networks, data centres, "safe city" packages | Loan-financed | Infrastructure, not organisations |
| OpenAI for Countries | National ChatGPT and data centres | Deal-based | Low now; high if a Pacific deal appears |
| TalanoaAI, Tetun.org | Community language AI (44 Pacific and First Nations languages; Tetum translation) | Free | Partners, not competitors |
No company was found selling a packaged, supported, governed offline AI appliance for development settings in the Indo-Pacific. Medium confidence
Development buyers compare against consulting day rates and cloud seats. One reading: sell the hardware near cost as a signal of good faith and value for money, and price the co-design, governance, capability and support.
| Package | Contents | A$ |
|---|---|---|
| Scoping and co-design | 2–3 weeks in-country, talanoa and workshops, use-case and data audit, draft Constitution | 40K–70K |
| Pilot | One country, 2–3 sites, 6–9 months, local custodians trained, baseline and endline measured, independent learning report | 150K–280K |
| Per site, grid | Unit, UPS, commissioning, 1 year support | 18K–30K |
| Per site, off-grid solar | As above plus solar, battery, enclosure | 28K–45K |
| Annual support per site | Updates, models, data packs, local partner retainer | 6K–15K |
| Capacity building | Train-the-trainer, local AI custodian certification, ministry governance training | 50K–120K |
| Hub and spoke | e.g. Fiji hub + 10–20 spokes over 3 years, regional custodians, language work, M&E | 0.8M–2.5M |
Selma tailored to the place it's used, with a community's Constitution written by locals for local use cases, is a strong idea. Others are already doing versions of it, which is both validation and a caution: we would be arriving as outsiders.
Te reo Māori speech recognition at 92% accuracy, trained on community-contributed recordings and governed by its Kaitiakitanga Licence, which keeps the data for Māori benefit.
1,000+ participants across 30+ countries building language AI, and explicitly rejecting "parachute research".
Launched May 2026: community-contributed agents for 44 Pacific and First Nations languages.
Community-involved translation and speech tools for Bislama, Pijin, Tok Pisin and Nafsan from small datasets.
What this suggests. Partner before using the phrase publicly (USP, TalanoaAI, Catalpa, Tetun.org, with Te Hiku as a possible adviser), and commit that community data, Constitutions and fine-tunes stay community property.