Week 43 - 2026Published 25 Oct 2026

HUMAIN Expands Saudi AI Compute Amid Global AI Shifts

Humain set the pace this week. It tendered a $1 billion contract for a 100MW data centre in Riyadh, expanded its NVIDIA partnership to deploy up to 84MW of AI computing capacity, and, with DataVolt, announced a preliminary framework with Nebius for up to 50MW at Oxagon, NEOM. Elsewhere, Moonshot released Kimi K3.1 Pro with open weights, and Revolut showcased PRAGMA, a single foundation model now behind its banking technology. DeepSeek open-sourced the DeepInfer engine behind V4.1-Flash, and ACE Robotics launched ACE-Brain-0.6 for spatial perception.

The week in AI

Arab

Arab AI Scene

Humain tenders $1 billion contract for 100MW Riyadh data centre

19 Oct 20269 Jumada I 1448 AH

Arab AI Scene

19 Oct 20269 Jumada I 1448 AH

Humain tenders $1 billion contract for 100MW Riyadh data centre

Humain, owned by the Public Investment Fund, has issued a tender for a $1 billion contract to build a 100MW artificial intelligence data center in Riyadh. The tender was issued in September, with bids due by November 8. The facility will occupy a 350,200 square meter site in Al-Saad, east of Riyadh, with an option to expand capacity to 200MW. Construction is expected to begin in January 2027, with operations starting in October 2028 and full capacity reached by early 2029. The project includes specialized cooling and electrical systems to support the power demands of AI computing.

Arab

Arab AI Scene

Saudi Arabia's Humain expands Nvidia deal with 84MW AI computing plan

22 Oct 202612 Jumada I 1448 AH

Arab AI Scene

22 Oct 202612 Jumada I 1448 AH

Saudi Arabia's Humain expands Nvidia deal with 84MW AI computing plan

Saudi Arabia's Humain has announced an expansion of its partnership with US chip maker Nvidia to develop up to 84 megawatts of AI computing capacity. The agreement includes 27,000 advanced graphics processing units (GPUs), with deployments set to begin later this year. This initiative is part of the kingdom's efforts to establish itself as a global AI hub, with the announcement made at Nvidia's GTC conference in Berlin. The first phase of the agreement will comprise 14 megawatts of Blackwell systems, with plans to expand capacity to 500 megawatts over five years.

Arab

Arab AI Scene

HUMAIN and DataVolt advance AI infrastructure framework with Nebius for planned giga campus in NEOM’s Oxagon

21 Oct 202611 Jumada I 1448 AH

Arab AI Scene

21 Oct 202611 Jumada I 1448 AH

HUMAIN and DataVolt advance AI infrastructure framework with Nebius for planned giga campus in NEOM’s Oxagon

HUMAIN and DataVolt announced a preliminary framework agreement with Nebius on October 21, 2026, to secure data center capacity at Oxagon, NEOM. The agreement includes an allocation of up to 50 MW from the initial 100 MW under construction, reflecting the growing demand for AI infrastructure. The framework combines HUMAIN's AI capabilities with DataVolt's infrastructure expertise, aiming to support next-generation AI workloads. The proposed capacity is expected to become operational in phases starting Q4 2027.

Global

Models

[AINews] Kimi 3.1 Pro 104BA-2.8T: Fable-5.1-Class Open Weights narrows Closed-AI lead to 1 month... and Pro Highspeed API with 2-3× speed

24 Oct 202614 Jumada I 1448 AH

Models

24 Oct 202614 Jumada I 1448 AH

[AINews] Kimi 3.1 Pro 104BA-2.8T: Fable-5.1-Class Open Weights narrows Closed-AI lead to 1 month... and Pro Highspeed API with 2-3× speed

Moonshot has announced the launch of Kimi K3.1 Pro, featuring a significant increase in model size and near-frontier independent benchmark results. The new model boasts 2.8 trillion total parameters, with 104 billion active parameters and a one-million-token context window. Moonshot describes K3.1 Pro as a "3T-class" mixture-of-experts model, including strengths in full-project coding, Blender/CAD work, and financial modeling. The model has been optimized to deliver better performance across multiple domains, enhancing its efficiency in use.

Global

Business

Leading European Financial Services Organizations Embrace Open Models to Build Proprietary Intelligence

22 Oct 202612 Jumada I 1448 AH

Business

22 Oct 202612 Jumada I 1448 AH

Leading European Financial Services Organizations Embrace Open Models to Build Proprietary Intelligence

At GTC Berlin, banks and payment companies showcased the PRAGMA model, which rebuilds Revolut's technology stack on a single foundation model powered by NVIDIA. Serving an initial 80 million customers, this model is designed to scale threefold, enabling efficient processing of billions of transactions. Revolut employs NVIDIA's cuDF toolkit and Dynamo-Triton software on Blackwell Ultra GPUs, proving that foundation models can operate at scale. Pavel Nesterov, head of AI at Revolut, states that this model is clear evidence of bridging the gap between research breakthroughs and production banking systems.

Global

Models

Deepseek open-sources the inference engine that powers its fastest and cheapest AI model

20 Oct 202610 Jumada I 1448 AH

Models

20 Oct 202610 Jumada I 1448 AH

Deepseek open-sources the inference engine that powers its fastest and cheapest AI model

Deepseek has announced the open-sourcing of its inference framework, DeepInfer, which powers the fast V4.1-Flash model. This vLLM-based system significantly reduces compute costs and energy consumption when running AI models. The release adds more pricing pressure on Western providers like OpenAI and Anthropic. DeepInfer represents a strategic advancement in the AI field, offering a cost-effective solution for the energy demands of running models. This system can enhance companies' ability to provide AI services at competitive prices, potentially reshaping market dynamics.

Global

Models

ACE Robotics Releases ACE-Brain-0.6, a 9B Model Tuned for Visual Grounding on Spatial Data

19 Oct 20269 Jumada I 1448 AH

Models

19 Oct 20269 Jumada I 1448 AH

ACE Robotics Releases ACE-Brain-0.6, a 9B Model Tuned for Visual Grounding on Spatial Data

ACE Robotics has launched ACE-Brain-0.6, an embodied foundation model with approximately 9 billion parameters, focusing on mapping language instructions to precise locations and object parts in physical scenes. The model is based on Qwen3.5-9B, with the vision encoder frozen and the language model fine-tuned on about one million spatial supervision samples, covering roughly 940,000 images and 2.7 million question-answer pairs. According to the company's results, ACE-Brain-0.6 outperformed OpenAI's GPT-6 Astra in six out of eight grounding evaluations.