Nutanix has built a $20 million AI cluster to reduce its use of tools including Copilot and Claude and expects the investment to pay back in about a year, according to The Register. The report is based on remarks from CEO Rajiv Ramaswami during a media briefing around the company’s Q4 results. Ramaswami told The Register that Nutanix’s software teams had been using AI across the coding lifecycle, including Copilot and Claude. He said usage rose sharply, and with it, costs. Nutanix has therefore moved toward open-weight models running on an on-premises cluster. The operational change is straightforward: Nutanix is trying to avoid paying per token for internal AI coding workloads where it believes an owned cluster and open-weight models are sufficient. Ramaswami said the company is “no longer paying on a per-token basis,” according to The Register, while noting that some users still occasionally turn to frontier models or external clusters when required. The Register frames that mix-and-match approach as increasingly common: organizations are matching models and infrastructure to particular workloads rather than assuming large language models from external providers are the default for every AI application. In Nutanix’s case, the shift is also a live example of the kind of enterprise AI infrastructure posture it sells to customers. The company also updated its Enterprise AI suite with a Model Context Protocol gateway. The Register describes the gateway as a layer between AI agents and MCP servers that provides identity management and security controls over which data and services agents can access. The Register also wrote that MCP gateways are quickly becoming standard issue in packaged AI infrastructure stacks, and that Nutanix added one to stay on par with competitors. The AI discussion came alongside a broader hardware-cost message. Ramaswami told The Register that Nutanix customers are highly focused on hardware costs and availability, which can slow software purchases. Nutanix is responding by expanding its hardware compatibility list and supporting external storage devices, moves Ramaswami said can help users move away from VMware without necessarily replacing existing hardware. Nutanix is also closer to porting its stack to Arm architecture, according to The Register. Ramaswami said Arm support could help the company’s software run on lower-cost hardware in environments where customers are reluctant to buy more expensive systems. That marks a shift from Nutanix’s 2024 stance, when an Arm port was described as a future possibility rather than an active development priority. The Register put the remarks alongside a briefing about the company’s Q4 results, reporting revenue of $757 million, up 16% year over year, and full-year revenue of $2.85 billion, up 12%. Full-year net income was $1.5 million. CFO Rukmini Sivaraman said the company remained focused on sustainable growth and profitability, while Ramaswami pointed to 3,000 new customers in the financial year and predicted Nutanix would continue winning customers from VMware for another five years. Who benefits: Nutanix benefits if customers see its own AI infrastructure choices as proof that on-premises or hybrid AI stacks can control costs. Customers with existing hardware may also benefit if expanded compatibility and external storage support reduce migration friction. Who's exposed: Per-token AI coding-tool providers are exposed where large enterprise users can shift sustained internal workloads to open-weight models. VMware remains the named displacement target in Nutanix’s customer-growth narrative, according to Ramaswami’s comments to The Register.