Private AI Infrastructure Takes Shape With Galaxy Blackhole

Organizations that need control over their data and AI infrastructure now have a new path to self-hosted deployments. Tenstorrent and ConfidentialMind announced a partnership on November 16, 2026, bringing hardware, software and operating controls together in one stack validated on Galaxy Blackhole systems.
The offer targets governments, universities, regulated enterprises and inference providers across Europe, the Middle East, Asia-Pacific and the Americas. Instead of assembling separate tools for model serving, data retrieval, user permissions and system monitoring, customers can purchase a combined platform that they operate in their own facilities.
A Self-Hosted Layer for Models and Agents
ConfidentialMind provides the deployment and management layer, while Tenstorrent supplies silicon, systems and regional field teams. The arrangement separates the hardware purchase from the platform-integration work, giving buyers a defined path from physical infrastructure to working AI services.
The platform bundles the functions organizations would otherwise have to build and connect themselves. Its customer-operated layer includes inference serving, vector databases, access control, accelerator management and observability, with multi-tenancy, data connectors and compliance logging added around the underlying compute.
From a self-service portal, customers can deploy open-source large language models, retrieval-augmented generation pipelines and agents. The platform serves those workloads through OpenAI-compatible API endpoints, creating a familiar way to connect applications to models running inside the customer’s own environment.
Tenstorrent software partner ecosystem head Grant Jensen said the intended result is that “teams can run language models, retrieval-augmented generation and agents in their own facilities without first constructing the software platform.” That goal captures the partnership’s central pitch: keep data and infrastructure under customer control without forcing every organization to create a complete AI operations layer from scratch.
The platform’s controls reach beyond inference. Multi-tenancy supports separate users or workloads, data connectors link the system to information sources, and compliance logging records deployment and operating activity in the self-hosted environment.
Galaxy Blackhole Brings Serious Compute
ConfidentialMind’s platform has been validated on Tenstorrent’s 32-processor Galaxy Blackhole systems. Galaxy Blackhole houses Blackhole Tensix Processors in an air-cooled 6U rackmount chassis, giving customers a standalone server option that can expand into a larger private AI installation.
- 23 PFLOPS of Block FP8 compute
- 1 TB of GDDR6 accelerator memory
- 16 TB/s of memory bandwidth
- 8–10 kW of average system power use
- 12 kW of maximum system power use, configurable up to 14.5 kW
- Up to 56 800-GbE QSFP-DD ports for cluster scale-out
That networking capacity gives buyers room to connect systems as their workloads grow. Customers can run one server in their own data center or build a supercluster, with Tenstorrent’s listed configurations extending to 36 or more Galaxy systems for large-scale language-model inference and private AI infrastructure.
The pricing makes the scale of the hardware clear. A standalone Galaxy Blackhole system carries a $160,000 list price, while a four-system Blackhole supercluster starts at $640,000. Those figures cover the hardware purchase, with the partnership placing the platform-integration work alongside the system rather than treating it as an invisible part of the server cost.
Open Software Creates a Deeper Customization Path
Tenstorrent says its full software stack, from compiler to kernel, is open source. Users can compile and run supported models or tune kernels directly, giving technical teams a customization path beneath the ConfidentialMind platform.
That combination creates two layers of control. ConfidentialMind handles deployment, management and the tools needed to operate models, while Tenstorrent’s open software stack gives users access to the lower levels of model execution and accelerator behavior.
For organizations with strict data requirements, that distinction matters. A self-hosted platform keeps workloads inside customer-operated infrastructure, and the open software stack gives users a way to adapt supported models and kernels instead of treating the hardware as a sealed system.
The joint sales plan now connects those pieces across four major regions: Europe, the Middle East, Asia-Pacific and the Americas. Governments, universities, regulated enterprises and inference providers can choose a single Galaxy Blackhole system or scale toward 36 or more systems as private AI demand grows.
The bigger story is the shift from buying compute alone to building an AI environment around it. Tenstorrent brings the processors, systems and field teams; ConfidentialMind brings the deployment and management layer; together, they are packaging the infrastructure needed to run models, retrieval-augmented generation and agents under the customer’s control.
Based on
- Tenstorrent Pairs Galaxy with ConfidentialMind’s Sovereign AI Platform — unite.ai
- European technology leaders form AI Gigafactory Northern consortium to build trusted, sustainable AI infrastructure | Markets Insider — markets.businessinsider.com
- DataKrypto Collaborates with Intel to Make Encrypted Execution a Standard Property of AI Infrastructure | Markets Insider — markets.businessinsider.com
- Nebul Adopts NetApp Novus to Power Sovereign AI Infrastructure | AFP.com — afp.com
- Rackspace and Together AI Announce Strategic Agreement to Support Enterprise AI Inference | Markets Insider — markets.businessinsider.com



