Think launches AI compute service to bypass hyperscaler costs

The new service centers on Think’s hardware and software stack, specifically the Think Fabric platform and its ILM orchestration technology. Each Grid SuperNode features four NVIDIA PRO 6000 Blackwell accelerators, providing 384GB of total VRAM. The ILM software pools this memory into a unified space, enabling multiple AI models to run on shared hardware and reducing the underutilization common in isolated deployments.

Think claims its bare-metal architecture achieves faster token generation and improved training parallelization compared to standard containerized environments. By bundling storage, support, and data egress into a single monthly fee, the company reports its pricing is up to 27% lower than comparable dedicated hyperscaler configurations. This model allows enterprises and government entities to bypass significant upfront capital expenditures.

Currently live in Saudi Arabia, the Riyadh-based deployment allows local organizations to process AI workloads while maintaining data sovereignty within the Kingdom. Founders Ahmed AlSharif and Ammar Enaya position the service as a direct response to the industry's reliance on either expensive, self-managed infrastructure or the unpredictable costs of public cloud providers. Think is now initiating a global rollout of the service, with registration for international markets open through their website.