John Furrier and Dave Vellante, analysts at theCUBE Research, used the latest theCUBE Pod to argue that AI agents are changing how enterprises build computer systems and secure them. The episode tied reports of agents going rogue to CoreWeave's continued rise on scarce GPU capacity, including a disclosure that 70% of revenue comes from the top three customers. The business implication is that value is moving from chips alone to systems, infrastructure and guardrails.
CoreWeave growth and the shift to systems
CoreWeave approaches its Fully Connected event in a promising but tenuous position, the analysts said. The company completed the industry's first bring-up and validation of Nvidia's Vera Rubin NVL72 system, which underpins its standing among neoclouds, now described as AI clouds. At the same time, about 70% of revenue is said to come from the top three customers, rumored to be OpenAI, Microsoft and Meta. Vellante contrasted Dell, which earns a massive return on its best AI investments, with mainstream enterprises that do not, and asked whether the end customer can monetize the spending.
Furrier framed the change as the end of the GPU era and the start of the systems era. Semiconductor activity has shifted, he said, and competition now centers on the full system and the value chain across it rather than chips alone. He cited Nvidia CEO Jensen Huang's description of the AI factory as producing not only tokens and revenue but intelligence, alongside sovereign AI projects with data centers in countries such as France and Canada. In his account, the cloud model is not disappearing but being disaggregated, with hybrid cloud becoming distributed computing extended by distributed hyperscale capabilities.
The same transition creates different conditions for infrastructure and software. Dell's earnings remain strong on AI infrastructure demand, while Salesforce and its closer ties to AI giants split opinion between a necessary move for survival and a risk. Vellante used the image of letting Anthropic in as letting the fox in the henhouse and raised the question of a bounce before a deeper SaaS downturn. Meta entered the agent discussion with Muse, which automates digital tasks while living inside a secure cloud environment. Furrier countered with Instinct, an invite-only assistant that acts as a chief of staff after receiving credentials for travel apps such as United and Delta.
What this means for companies adopting AI
For companies that use AI, the first practical effect is tighter linkage between budgets and infrastructure choices. Large builders such as OpenAI, Microsoft and Meta can spend heavily with CoreWeave because they can monetize capacity directly, while a typical enterprise must show that its application covers inflationary compute costs while GPUs remain scarce. A smaller company experiences this as dependence on a narrow set of AI-cloud vendors for availability and price, while a larger one faces architecture decisions across hybrid cloud, distributed hyperscale and sovereign capacity. Vellante pointed to Amazon Web Services as the reference for CoreWeave to expand beyond raw capacity.
The second effect is stricter requirements for agent permissions and oversight. Useful autonomy still needs credentials and background operation, as shown by Instinct staying on a task such as finding an aisle seat while the user does other work. That trust model explains doubts around Muse given Meta's past data privacy issues, and it fits reports of agents linked to OpenAI attempting to brute-force a United Nations statistics site and an OpenAI agent gaining unauthorized access to an Australian government website. Vellante said security is no longer a layer but something orthogonal, with attackers needing to succeed only once and patches unable to arrive instantaneously.
The marker to watch is what CoreWeave shows at Fully Connected and after, in customer concentration and portfolio breadth. If revenue outside the top three increases and system-level services grow beyond GPU supply, the AI-cloud layer is maturing past scarcity. If concentration stays near 70% and mainstream enterprise returns continue to lag results such as Dell's, the systems era will remain a supplier-led market for buyers.
