Nutanix builds $20m AI cluster to cut model costs
Nutanix says it has built a $20 million on-premises AI cluster to curb escalating spending on services such as Microsoft Copilot and Anthropic’s Claude, expecting to recover the investment within a year. The move reflects a wider effort to control AI token costs by using open-weight models internally, while retaining access to frontier models or external infrastructure for workloads that need them.
The company is also moving closer to supporting Arm processors, arguing that lower-cost hardware and wider deployment options could help customers facing constrained availability and high equipment prices. Nutanix reported fourth-quarter revenue of $757 million, up 16 per cent year on year, while full-year revenue rose 12 per cent to $2.85 billion; it added 3,000 customers, many reportedly switching from VMware, but full-year net income was only $1.5 million.
- Nutanix built a $20m AI cluster to reduce token-based AI costs.
- It expects the investment to pay back within a year.
- Arm support could lower hardware costs for customers.
New here? Start with this
Nutanix makes software that helps businesses run and manage computer systems in their own data centres or through cloud services. AI tools such as Microsoft Copilot and Anthropic’s Claude can charge for use, so frequent requests to these systems may add up.
Some businesses are exploring open-weight AI models, whose underlying model files can be used and adapted on their own computers. This can reduce reliance on paid services, though the most capable or demanding tasks may still need external models or computing services.
Nutanix is also considering support for Arm-based processors, which offer another hardware option alongside the chips commonly used in data centres. Its results provide context for these plans: revenue grew, but its reported full-year net income was small.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Building an in-house AI cluster could make costs more predictable and reduce reliance on expensive external services, especially for routine tasks that open-weight models can handle well. If the investment pays back within a year, it may also give Nutanix greater control over data and infrastructure while preserving access to frontier models for work that needs them.
The case against
A $20 million cluster is a substantial commitment, and the projected one-year payback depends on uncertain usage, model performance and ongoing hardware costs. External services may offer stronger models and shift maintenance and upgrade burdens to providers; buying capacity in-house could leave Nutanix with costly, underused equipment as the market changes.
Read the full article at the source →
Originally published by The Register as “Nutanix built $20m AI cluster to reduce use of Copilot and Claude, expects ROI in a year”.