Local AI development,
built for control and scale
Enable AI teams to prototype, fine tune and run inference locally with predictable performance and a clear path to scale.
Why upgrade
AI development is facing practical constraints
As organisations expand AI capability, development teams are encountering real-world limits. Cloud GPU pricing fluctuates. Shared environments introduce queue time. Governance requirements limit how sensitive data can be handled.
These constraints slow iteration and create uncertainty around cost, compliance and delivery timelines.
AI programmes rarely stall due to lack of ambition. They stall when development environments become unpredictable.
The challenge
Iteration speed is reduced by shared cluster contention and remote dependencies.
The risk
Volatile consumption pricing and governance concerns can delay projects and increase operational exposure.
The requirement
Organisations need controlled, always available AI development capacity aligned with production environments.
The solution
Lenovo ThinkStation PGX provides compact, office-ready AI compute powered by NVIDIA Grace Blackwell architecture, designed specifically for local AI workflows.
Why ThinkStation PGX
Explore how ThinkStation PGX combines AI performance, unified memory and deployment flexibility to support local AI development.
Optimised for modern AI workflows
Designed specifically for AI development, PGX combines CPU, GPU, high bandwidth unified memory and networking in a single compact system, supporting experimentation, fine tuning and inference without reliance on shared resources.
Software foundation
Linux native AI
stack
NVIDIA DGX OS based on Ubuntu Linux and the NVIDIA AI software stack, including CUDA 13, providing a consistent development environment aligned with larger scale infrastructure.
Performance
AI performance at the desk
Up to 1 PFLOP of AI performance powered by the NVIDIA GB10 Grace Blackwell Superchip, enabling high performance model development in a controlled local environment.
Compact and
energy efficient
A 1.13L small form factor with a maximum 240W power profile, designed for office and edge environments without specialist facilities.
Unified memory architecture
128GB LPDDR5X coherent unified system memory with 273GB/s bandwidth, designed to support model experimentation, inference and data intensive workflows.
Deployment flexibility
Flexible
deployment modes
Operate PGX in standalone, companion or cluster mode, enabling teams to work locally, offload GPU intensive tasks or link units together for increased model capacity.
Built for practical AI development
ThinkStation PGX supports common AI development and experimentation workflows in controlled local environments.
Development
Prototyping and fine tuning
Experiment with model architectures, refine parameters and validate outputs locally before scaling to larger infrastructure.
Data
Data science and model validation
Use GPU acceleration to support data preparation, analytics and testing workflows that benefit from high performance and unified memory.
Applications
Enterprise AI applications
Develop and test AI agents and internal tools that rely on dynamic context switching, proprietary datasets and governed environments.
Deployment
Edge and inference workloads
Deploy AI capability in office or edge locations where latency, data sensitivity or operational control are priorities, and run inference locally with responsive performance.
Start local. Scale with confidence.
ThinkStation PGX provides an entry point into a broader Lenovo AI infrastructure continuum.
Organisations can:
- Begin with desk side AI development on PGX
- Expand to ThinkStation P Series systems with additional GPU capacity
- Scale further to Lenovo ThinkSystem servers supporting multiple GPUs per server
The consistent Linux native NVIDIA AI stack supports smoother transition from development to production environments.
Why choose us
This is where VitrX supports your AI roadmap
VitrX works with organisations to design AI capability that is practical, governed and aligned with budget cycles.
We support:
- Workload assessment and architecture planning
- Alignment with lifecycle and refresh strategies
- Deployment and standardisation
- A clear path from experimentation to scaled production
The objective is predictable AI capability, not reactive hardware decisions.
Ready to
move forward?
If you are planning AI capability over the next 12 to 18 months, ThinkStation PGX provides a controlled starting point with a defined path to scale.
Questions
Find answers to common questions about ThinkStation PGX and planning AI capability with VitrX.
What is Lenovo ThinkStation PGX designed for?
Lenovo ThinkStation PGX is designed to support local AI development workflows, including prototyping, fine tuning and inference. It provides high performance AI compute at the desk, enabling development teams to experiment and validate models in a controlled environment before scaling to larger infrastructure.
Is PGX a replacement for cloud AI infrastructure?
No. PGX is not positioned as a replacement for cloud or data centre AI infrastructure. It is designed to support the early and iterative stages of AI development locally. Organisations can then scale workloads to larger on-premise or cloud environments using a consistent software foundation when higher throughput is required.
How does PGX support data governance?
PGX enables AI development to take place within a controlled local environment. This can help organisations manage sensitive datasets, proprietary code and regulated information without immediately moving them into shared or external environments. It supports governance discussions by keeping early-stage experimentation within defined boundaries.
Can PGX integrate with existing Lenovo or NVIDIA infrastructure?
Yes. PGX is built on the NVIDIA Grace Blackwell architecture and uses a Linux-based NVIDIA AI software stack. It is designed to align with broader Lenovo ThinkStation and ThinkSystem infrastructure, supporting a smoother transition from local development to scaled production environments.
How does VitrX support deployment?
VitrX works with organisations to assess AI use cases, map workloads to appropriate execution environments and align infrastructure decisions with lifecycle planning. We support selection, deployment and standardisation to ensure AI capability is introduced in a structured and predictable way.
