The pitch landed somewhere between audacious and pragmatic: custom-built AI supercomputers designed to run inside government ministries and enterprise data centers, not in hyperscale clouds owned by Amazon or Microsoft. And on July 15, 2026, Riyadh-based Think managed to convince investors to back that vision with over $8 million—what the company claims is the largest AI infrastructure pre-seed round the Middle East has seen.
The raise—co-led by RAED Ventures and Wa'ed Ventures, Saudi Aramco's venture arm—suggests that sovereign AI infrastructure has moved from abstract policy goal to concrete investment thesis. At least in Saudi Arabia.
RAED manages over $550 million in assets under management. Wa'ed commands a $500 million mandate. Both are placing bets that enterprises and governments across the Gulf Cooperation Council states will pay a premium for compute they control entirely, rather than rent by the hour from Silicon Valley. Dhahran Techno Valley's new venture arm joined the round, alongside a cluster of strategic angels whose names Think declined to disclose.
Building the Hardware Nobody Asked For (Yet)
Think emerged from stealth earlier this year after a 2025 founding, and its core product is not subtle. The AI SuperNode is a liquid-cooled, multi-GPU unit that the company claims pushes up to 8 petaflops of FP8 compute and up to 1 terabyte of VRAM at 4.8 TB/s bandwidth—all inside a 6,000-watt thermal envelope. Think says the design achieves 135% higher power density than comparable enterprise GPU servers, though "comparable" is doing some heavy lifting there.
The hardware runs alongside ILM, Think's orchestration software, which handles hardware-aware task scheduling, dynamic inference routing, and adaptive fine-tuning across LoRA, QLoRA, and reinforcement learning frameworks. Then there's Constellation, which orchestrates multi-node clusters—including, intriguingly, office PCs commandeered "by night" for distributed training and inference workloads.
In production tests, Think reports sustained GPU utilization above 90%, well north of the 30-50% industry averages that have become something of an embarrassment for enterprises running their own AI infrastructure. The company also claims per-million-token costs nearly 10 times lower than frontier models from OpenAI, Google, and Anthropic. Those figures come directly from Think's July 15 announcement and haven't been independently verified. Four patents are pending.
Perhaps the more interesting question is whether anyone actually wants what Think is selling.
The Founders and the Timing
The co-founder and CEO brings a resume that reads like a tour of Silicon Valley's gaming and metaverse hardware: Meta's Reality Labs platform engineering, Sony PlayStation Europe, Electronic Arts, and a CTO stint at Sandsoft. His co-founder logged senior engineering years at Cisco, HPE/Aruba, and Vectra AI before running Ajlan & Bros Technology as CEO. Think's LinkedIn page lists somewhere between 11 and 50 employees as of late July 2026—exact headcount remains fluid, as it tends to be in fast-growing startups.
The timing, though, looks deliberate. Saudi Arabia's Public Investment Fund launched HUMAIN in May 2025, a state-backed AI company charged with building next-generation data centers, hyper-performance cloud infrastructure, and Arabic-language multimodal large language models. By August 2025, HUMAIN had released its ALLaM 34B model and a consumer chat application. Think positions itself as complementary to that effort—providing the infrastructure layer for enterprises that want sovereign AI but lack the capital or expertise to build it themselves.
The kingdom's broader AI ambitions are no secret. What remains unclear is how many enterprises are willing to shoulder the upfront capital expense and operational complexity of on-premises AI infrastructure when hyperscale clouds offer usage-based pricing and near-infinite elasticity.
A Crowded Few Weeks
The first half of 2026 saw $1.7 billion deployed across 242 deals in the Middle East and North Africa, according to Wamda's H1 report. June alone accounted for $148.2 million spread across 41 rounds. Think's raise came just days after Dubai-based AVELIN AI closed a $3.7 million pre-seed targeting sovereign AI for regulated sectors, and a few weeks after 1001 landed a $30 million Series A led by Lux Capital.
Sovereign AI infrastructure, it seems, has become a category.
Think says it's engaged in multiple proofs-of-concept, production deployments, and strategic partnerships across Saudi Arabia. Customer names, deployment counts, and revenue figures remain undisclosed. So does the valuation.
What Happens Next
The capital will go toward expanding the engineering team, scaling manufacturing, and accelerating product development—the usual destinations for pre-seed proceeds. Think also plans selective expansion into GCC markets over the next 18 to 24 months and expects to showcase its platform at LEAP 2026, Riyadh's flagship tech conference, scheduled for late August and early September.
Think frames its go-to-market strategy around a 12 to 24-month ROI window compared to cloud alternatives—a bold claim that hinges on sustained utilization rates, predictable workloads, and capital costs that pencil out against hyperscale pricing. For enterprises with data residency requirements or genuine sovereignty concerns, the calculus may work. For everyone else, the math gets harder.
Whether Think's bet pays off depends less on the elegance of its hardware and more on how quickly sovereign AI moves from policy talking point to procurement line item. In Saudi Arabia, at least, the infrastructure is starting to take shape.
