Think closes MENA's largest AI infrastructure pre-seed round at over $8 million

• 15 Jul 2026
• TM Team
• 0 Comments

Think, the Saudi-based company building a new generation of intelligent, unified hardware and software infrastructure for artificial intelligence, announced it has raised over $8 million in pre-seed funding, marking the largest AI infrastructure and deeptech pre-seed round in MENA to date.

The round is being co-led by RAED Ventures and Wa'ed Ventures, with participation from Dhahran Techno Valley's Venture Capital arm and strategic angel investors. The capital will support team expansion, manufacturing scale-up, product development, and international growth initiatives as Think rapidly accelerates deployments across Saudi Arabia and expands its presence across the GCC and selected global markets.

Think is focused on solving the next major challenge in AI adoption by reducing the cost and complexity of AI infrastructure while dramatically improving efficiency. Its technology combines high-density, liquid-cooled multi-GPU compute nodes with proprietary bare-metal orchestration software, enabling companies of any size to deploy AI models more efficiently, securely, and cost-effectively while maximising all available compute capacity.

Think was founded by Ahmed AlSharif, a technology leader whose career includes senior roles at Meta, Sony PlayStation Europe and EA Games, alongside enterprise technology veteran Ammar Enaya, whose career spans leadership positions at Cisco, HPE Aruba and Vectra AI.

"As the industry moves beyond the race for bigger models and larger data centres, a new age of efficiency is beginning," said CEO Ahmed AlSharif. "AI infrastructure today is expensive, inefficient, and increasingly difficult to scale. Think exists to help organisations do more with the compute they already have, offering an alternative to the industry's current obsession with bigger, faster and more expensive."

Think's approach combines proprietary AI Node hardware with ILM, a software orchestration layer designed to maximise GPU utilisation, lower token costs, and reduce the overall cost of deploying AI. In production benchmark testing, the platform achieved sustained GPU utilisation of more than 90%, compared with industry averages of 30"50%, with a per-million-token cost that's almost 10x lower than the average cost of using frontier models from Google, OpenAI, and Anthropic.

This is all achieved using existing, widely available GPUs, and doesn't require proprietary or specialist inference hardware. The platform will soon support mixed-vendor and specialist inferencing silicon working in tandem for both inferencing and training.