UAE's G42 Just Finished the Largest AI Concrete Pour Outside the US, and That Matters More Than the Chips
In the industrial backlots of Abu Dhabi's Khazna free zone, G42 has just completed what may be the most consequential piece of Gulf AI infrastructure yet poured. The state-backed compute champion has finished laying 100,000 cubic metres of concrete across a data-centre campus roughly the size of the city of Berkeley, California, forming the physical spine of what the UAE is calling its sovereign AI platform. It is the largest AI-dedicated build outside the United States, and the message it sends to Washington, Beijing, and Riyadh is louder than any ministerial press release.
The campus sits at the intersection of three parallel sovereign AI plays now reshaping the Gulf. G42's partnership with NVIDIA, extended through new infrastructure agreements, feeds the compute. Oracle Cloud Infrastructure supplies the software layer. And a region-specific instance of ChatGPT, built on licensed weights and tuned for Arabic governance workloads, is already being pushed across UAE government departments, banks, and healthcare providers through the TAMM platform and adjacent channels.
This is not a pilot. It is production AI at nation scale.
Why the Build Size Is the Story
Most coverage of Gulf AI in the last year has obsessed over chip supply, export licences, and whether H100s or Blackwells can clear US Commerce. But the chips were always going to arrive one way or another. The bottleneck is power-dense real estate, water-cooled halls, and the year-long lead times on switchgear and transformers. By physically finishing the concrete, G42 has locked in timing advantages that no rival in the Gulf can replicate before 2028.
Saudi Arabia's Humain project, which broke ground on a similarly ambitious campus in Riyadh with Qualcomm, Amazon Web Services, and Cisco support, is understood to be at least twelve months behind on buildout, according to contractors active on both sites.
That timing matters for one reason: whichever Gulf state hosts the first working hyperscale Arabic-native AI model will set the reference architecture for the next decade of MENA public services. The infrastructure is the policy.

The Africa Expansion Nobody Is Tracking
The UAE has also committed $1 billion to expand AI infrastructure and services across Africa, a pledge that sits alongside G42's earlier $1.5 billion cooperation framework with Microsoft and its growing data-centre footprint in Kenya and Egypt. Taken together with a similar G42-led agreement in Vietnam, the Abu Dhabi playbook is now explicit: build sovereign AI stacks in the Gulf, then export the template to the Global South as a package of compute, cloud, and governance tooling.
That matters because it repositions the UAE from a consumer of frontier AI into a wholesaler of it, competing with China's DeepSeek-adjacent infrastructure push and the US hyperscaler bundle. It also gives G42's Inception Institute spinouts, including Jais and Falcon variants, real distribution channels rather than benchmark leaderboards.
By The Numbers
- 100,000 cubic metres: volume of concrete poured at G42's Abu Dhabi campus, the largest AI build outside the United States
- $1 billion: UAE commitment to expand AI infrastructure across Africa in 2026
- $1.5 billion: G42's cooperation framework with Microsoft underwriting the UAE sovereign AI stack
- $100 billion: projected annual Middle East capex on AI, energy, data centres, and digitisation by 2026
- 40%: share of Saudi GDP still derived from oil and gas, the gap Humain's AI build is meant to narrow
What the Corporate Stack Actually Looks Like
Under the hood, the UAE sovereign AI platform is a stack, not a product. NVIDIA supplies the silicon. Oracle handles infrastructure software. Microsoft contributes Azure-linked model access and enterprise integration.
And a regional OpenAI relationship routes ChatGPT-class inference through local endpoints with data residency inside UAE borders. Bank and healthcare customers never see the US hyperscaler logos. They see TAMM, M42 portals, or their own enterprise applications.
| Layer | Supplier | UAE Partner | Workload |
|---|---|---|---|
| Silicon | NVIDIA | G42 | Training and inference |
| Cloud | Oracle, Microsoft | G42, Core42 | Hosting, orchestration |
| Foundation models | OpenAI (regional), Inception | MBZUAI, G42 | Arabic and English LLMs |
| Applications | Various | TAMM, M42, ADQ portfolio | Gov services, health, finance |
The Middle East is building in weeks what the US and Europe still debate in quarterly earnings calls. That is not hype. That is what we are seeing in the plant schedules.
Whoever finishes the concrete first sets the standard. That is the only rule in sovereign AI right now.
Risks Still Looming Over the Programme
This is not a clean story. The ongoing Iran war, stretching from Hormuz to the Red Sea shipping lanes, is adding friction to every major Gulf infrastructure project. Insurance costs for specialised switchgear shipments have climbed. Key expat engineers are being rotated out of Abu Dhabi for security windows.
And the US-Gulf asymmetric interdependence, trading American chips and cloud code for Gulf energy and compute, remains politically fragile. A single Commerce Department ruling could redraw the entire stack.
There is also a rising question about whether the UAE model is durable or simply first. Saudi Arabia's HUMAIN project, which we covered in HUMAIN and the 600,000 GPU Play, is pursuing the same goal at comparable scale with different partners. Qatar's Ooredoo Oracle Alloy sovereign cloud is building a competing enterprise channel. And Iraq's Manara and Bahrain's National AI Portal show the GCC is now fragmenting into multiple sovereign AI stacks rather than consolidating into one.
When an AI model processes input and produces output. The actual 'thinking' step.
Graphics Processing Unit, the powerful chips that AI models run on.
A standardized test used to compare AI model performance.
The processing power needed to train and run AI models.
A massive cloud computing provider like AWS, Azure, or Google Cloud.
National initiatives to develop domestic AI capabilities independent of foreign providers.