Saudi-backed Humain taps MIS to build private AI data center; JV with stc targets up to 1 GW
Humain, backed by Saudi Arabia's Public Investment Fund, has awarded Al Moammar Information Systems Co. (MIS) a contract to design and build a private data center dedicated to AI. The project value exceeds 155% of MIS's 2024 revenues. With MIS reporting SR1.21 billion ($320 million) last year, the deal implies nearly SR1.88 billion (about $500 million).
The contract is expected to be signed on Feb. 15, 2026. MIS will handle end-to-end design and construction. The company noted there are no related parties tied to the deal.
Context: stc partnership and scale targets
Humain also entered a joint venture with Saudi Telecom Co. this month to develop and operate AI-focused data centers in the Kingdom through stc's subsidiary, center3. Humain will own 51% and stc 49%.
The facility roadmap calls for advanced infrastructure capable of up to 1 gigawatt of power, starting with 250 megawatts based on demand. That level points to hyperscale GPU clusters and serious grid planning.
What this means for engineers and builders
- Compute density: Expect 30-80 kW per rack (and higher) for H100/B200/MI300-class GPU pods; liquid cooling will likely be required.
- Network fabric: 400/800G Ethernet with RoCE vs. InfiniBand NDR-design choices here will define training throughput and efficiency.
- Cooling strategy: Rear-door heat exchangers, direct-to-chip liquid, or immersion-CDU capacity and redundancy will matter.
- Power delivery: New substations, dual feeds, UPS architecture, and generator strategy; watch PUE targets and water usage.
- Storage tiers: NVMe-oF for hot data, scalable object storage for checkpoints and datasets, plus high-speed caching layers.
- Security & isolation: Private environment for Humain suggests strict segmentation; air-gapped enclaves may be on the table.
- Interconnect & latency: Proximity to center3 PoPs improves east-west bandwidth and external connectivity for data ingest and delivery.
Saudi AI momentum
Saudi Arabia continues to push AI infrastructure and talent. The Kingdom ranked fifth globally and first in the Arab region for AI sector growth, according to the Global AI Index-backing the country's economic diversification under Vision 2030.
For reference on national priorities, see Vision 2030's official overview: Vision 2030. For network and facility backbone context, explore center3: center3.
What we know so far
- MIS contract value: nearly SR1.88 billion, exceeding 155% of MIS's 2024 revenues.
- Scope: MIS will design and construct a private AI data center for Humain.
- stc-Humain JV: 51% Humain, 49% stc, executed via center3; up to 1 GW target, starting at 250 MW.
- National standing: Saudi ranks fifth globally for AI sector growth (Global AI Index).
- Separate MIS win: SR114.43 million, 36-month contract with the Saudi Central Bank for IT license renewals and support.
Timeline, risks, and procurement notes
With a 2026 signing, long-lead items should be scoped early. Transformers, switchgear, and generators can run 12-18 months; accelerators and optical modules often take 6-12 months depending on allocation.
- Site & permits: Land, zoning, and environmental approvals drive the critical path.
- Grid interconnect: Substation design, dual feeds, and protection schemes set the upper limit for expansion speed.
- Modular builds: Shell-first with incremental MEP and white space lets teams land early 10-50 MW blocks.
- Liquid cooling readiness: CDU capacity planning, facility water systems, and containment must be nailed early.
- Accelerator supply: Secure GPU/AI accelerator allocation, NICs, and optics alongside the fabric decision.
- Software stack: Kubernetes or Slurm for scheduling, GPUDirect/NCCL optimization, observability, and cost allocation baked in.
What this could enable
Large-scale training and fine-tuning within Saudi Arabia, including multilingual LLMs, video and 3D generation, and sector-specific models where data residency is essential. Expect growing demand for onshore inference capacity, retrieval pipelines, and secure data platforms tied to local compliance.
Actions for IT and development teams
- Plan for 800G Ethernet or NDR InfiniBand topologies; validate with cluster-scale NCCL microbenchmarks, not just single-node tests.
- Adopt container-native HPC: GPU operators, SR-IOV/DPDK, Multus/Calico, and storage classes for NVMe-oF and object tiers.
- Implement strong tenancy: network policies, eBPF, SPIFFE/SPIRE, KMS-backed secrets, and dataset lineage controls.
- Design for portability: training on-prem with burst-to-cloud inference or vice versa; bake in CI/CD for models and data.
- Upskill teams on MLOps, GPU performance, and data governance. Useful starting point: AI courses by job.
Bottom line
This is a serious capacity build with national backing, a private AI facility for Humain, and a JV pipeline that targets up to 1 GW. If you run infrastructure, start modeling power, cooling, and fabric choices now-supply chains and design choices made in the next 6-12 months will dictate what's feasible by 2026 and beyond.
Your membership also unlocks:
