Saudi Arabia's national AI company has widened its chip deal with Nvidia, and the capacity involved is large enough to reshape planning conversations well beyond data centres.
The change matters most for teams in FMCG, Pharma and Healthcare, Retail and Modern Trade, and Logistics and Field Services.
This post covers what was announced, why computing power is only half of an AI plan, and what distributors can do about the other half.
What HUMAIN Announced on 22 October
- HUMAIN, the AI company owned by the Public Investment Fund, expanded its Nvidia partnership at Nvidia's GTC conference in Berlin.
- The plan covers up to 84MW of AI computing capacity, about 27,000 GPUs, delivered in two phases.
- It builds on the May 2025 partnership, which already involved 18,000 Blackwell processors.
| Phase | Capacity | Approx. GPUs | Timing |
|---|---|---|---|
| Phase 1 (Blackwell HGX B300) | 14MW | 6,000 | From Q4 2026 |
| Phase 2 (Vera Rubin) | 70MW | 21,000 | From mid-2027, subject to final agreements |
Why Compute Is Only Half the Plan
Extra GPUs raise what a model can process. They do not raise the quality of what it is given.
In distribution, the inputs are orders, stock positions, price lists, scheme claims and visit records.
If those sit in spreadsheets and chat threads, the model sees a delayed and partial picture.
What Distributors Can Do Now
The useful work happens before any model is bought: capturing each transaction once, in one place.
A structured record has a clear owner, a timestamp and a status, so it can be trusted by people and by software alike.
Cloud platforms with offline-first apps keep that record complete across long routes between cities.
Building the Data Layer With 1Channel
1Channel puts orders, stock, claims and field activity on one cloud platform, with automation validating each order against stock, schemes, price and credit.
That gives distributors a clean, current record before any AI layer is added on top.
Give Your AI Plans Data Worth Computing
See how distributors in Saudi Arabia keep orders, batches and collections in SAR on one synced backend, with a chat assistant on top.
Explore AI-Powered DMS →Final Suggestions
- Treat the HUMAIN timeline as a planning signal: Phase 1 begins in Q4 2026, so data work started now has a head start.
- Pick one record type, such as secondary orders, and make it complete before widening the scope.
- Assign an owner for every master list, including SKUs, outlets and price lists.
- Judge any AI tool by the freshness of the data it can reach, not by the hardware behind it.
Source: The National
