Niger, the world's 3rd AI power? We checked.

The verdict at a glance

The video makes nine claims. Here is our verdict on each.

The 9 claims

“No evidence” describes the available material, not proof that an invention is impossible. Context links are not tests of the phone shown.

Evidence 1: offline AI on phones has existed since 2023

AI starts with a model: the program’s brain. Models that run on a phone without the internet have existed since 2023. Source

Earlier still: llama.cpp was running a GPT-3-style model on computers, phones and Raspberry Pi in March 2023. Source

Evidence 2: nobody has solved hallucinations

Researchers have already developed several methods to reduce them:

• RAG (Retrieval-Augmented Generation), 2020. Researchers at Facebook AI, now Meta: before answering, the AI retrieves information from real documents. Source
  Question Real documentsAnswer
• Neuro-symbolic AI. The model is combined with rules and knowledge bases. That is the idea behind the “hybrid AI” described in the video. A major scientific survey reviewed related approaches in November 2023. Source
• September 2025. OpenAI, the company behind ChatGPT, publishes a full study, “Why language models hallucinate”, explaining why models still hallucinate. Source

These methods reduce errors

They eliminate errors

They guarantee zero errors

A genuine solution would be a world-class breakthrough, backed by a paper and independent tests. A WhatsApp video alone cannot establish that.

Evidence 3: the US and China are not alone

They do lead the field. But they are not alone, and Africa already has models of its own. Source (Bloomberg)Source (Lelapa AI)Source (Nigeria)

Who builds AI models?

Niger would therefore not be the third. And that is actually good news, as we explain below.

The other claims, briefly

The timeline

1. 2003

   Skype, internet calls Source

2. August 2012

   VoLTE, voice over IP from mobile carriers Source

3. 2017

   Google Pixel 2 with eSIM Source

4. 2018

   iPhone XS with eSIM Source

5. 2020

   RAG, published by Facebook AI (now Meta) Source

6. Sept. 2022

   eSIM-only iPhone 14 in the US Source

7. March 2023

   llama.cpp, a GPT-3-style model on a phone, offline Source

8. 2023

   Mistral (France) and Falcon (UAE) release their models Source (Mistral AI)Source (TII)

9. Nov. 2023

   Scientific survey on knowledge graphs and hallucinations Source

10. Dec. 2023

    Gemini Nano in Pixel 8 Pro Source

11. Apr. 2024

    Microsoft Phi-3 mini on iPhone Source

12. August 2024

    InkubaLM, Lelapa AI (South Africa) Source

13. Apr. 2025

    IEA report “Energy and AI” Source

14. May 2025

    Google Gemma 3n Source

15. Sept. 2025

    OpenAI, “Why language models hallucinate” Source

16. Sept. 2025

    Nigeria presents N-ATLAS at the UN Source

17. Oct. 2026

    The viral video

Why this article

I support the initiative, and especially the effort. But the video asks for backing from the authorities and public money. Before offering support, we ask for evidence. That is all.

• Where is the demo, tested by someone independent?
• Where are the numbers?
• Where is the code, the paper or the patent?
• Who makes the phone, and where?

The real good news

The technologies described already exist, often with openly available models (Gemma, Llama, Mistral, Qwen, DeepSeek). Niger does not need to reinvent AI. We can start building AI in Hausa, Zarma and Tamasheq today, and train thousands of young people to use it. That is real leadership. Source (Google)Source (Mistral AI)Source (Stanford AI Index)

Join the team

I want to make this project real: AI in our languages. Whether you are a developer, linguist, teacher or simply eager to help, join us.

Every Sunday, our masterclass helps you use AI in practical ways in your work.

All sources

1. Google Gemini Nano, Pixel 8 Pro (2023)https://9to5google.com/2023/12/06/pixel-8-pro-gemini-nano/
2. Microsoft, Phi-3 Technical Report (2024)https://arxiv.org/abs/2404.14219
3. Google, Introducing Gemma 3n (2025)https://developers.googleblog.com/en/introducing-gemma-3n/
4. llama.cpp, local models (2023)https://lifeboat.com/blog/2023/03/you-can-now-run-a-gpt-3-level-ai-model-on-your-laptop-phone-and-raspberry-pi
5. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020)https://arxiv.org/abs/2005.11401
6. Unifying Large Language Models and Knowledge Graphs: A Roadmap (2023)https://arxiv.org/abs/2311.07914
7. OpenAI, Why language models hallucinate (2025)https://openai.com/index/why-language-models-hallucinate/
8. Lelapa AI, InkubaLM (2024)https://arxiv.org/html/2408.17024v1
9. Nigeria, N-ATLAS at UNGA80 (2025)https://www.opensourceforu.com/2025/09/nigeria-unveils-open-source-n-atlas-ai-model-to-champion-african-languages-at-unga80/
10. Bloomberg, US and China dominate the model race (2025)https://www.bloomberg.com/news/articles/2025-05-06/mideast-titans-step-back-from-ai-model-race-as-us-china-dominate
11. Fortune, Mistral and European AI sovereignty (2026)https://fortune.com/2026/08/05/mistral-europe-ai-sovereignty-us-openai-anthropic-google-deepmind/
12. PCWorld, MetroPCS Voice-over-LTE service (2012)https://www.pcworld.com/article/460507/metropcs_claims_worlds_first_voiceoverlte_service.html
13. eSIM, device historyhttps://en.wikipedia.org/wiki/ESIM
14. 9to5Mac, iPhone 14 eSIM-only in the US (2022)https://9to5mac.com/2022/09/07/iphone-14-esim-only/
15. Notebookcheck, TSMC 3 nm wafer costshttps://www.notebookcheck.net/TSMC-3nm-wafer-costs-surge-to-18-000.943017.0.html
16. IEA, Energy and AI (2025)https://www.iea.org/reports/energy-and-ai/executive-summary
17. Skype, launch in August 2003https://en.wikipedia.org/wiki/Skype
18. TII, Falcon modelshttps://falconllm.tii.ae/
19. Mistral AI, Mistral 7B (2023)https://mistral.ai/news/announcing-mistral-7b
20. Stanford AI Index, model development by countryhttps://hai.stanford.edu/ai-index/2025-ai-index-report
21. Britannica, Niameyhttps://www.britannica.com/place/Niamey
22. Carl Sagan, Extraordinary claims require extraordinary evidencehttps://en.wikipedia.org/wiki/Extraordinary_claims_require_extraordinary_evidence