Agentic AI adoption stretches cloud infrastructure and security models

Enterprises racing to deploy agentic AI, from customer service to data analytics, are hitting infrastructure bottlenecks that demand new compute architectures, orchestration layers and identity security models. In June 2026, a flurry of announcements underscored the rapid shift.

AWS reports that its Arm-based Graviton processors now account for more than half of new CPU capacity deployed over the past three years, signaling a decisive pivot from x86 toward performance per watt. Google Cloud and Microsoft have followed with their own Arm designs, including Axion and Cobalt.

Orchestration startups fill the middleware gap

Startups are capitalizing on the complexity of managing multi-agent AI systems. Dutch-founded Optiak emerged from stealth with a EUR4 million pre-seed round to build what it calls a modular operating system for enterprise AI orchestration. The company is routing agent requests across models, data sources and API endpoints.

Malaysian messaging platform Respond.io raised USD62.5 million to scale its agent-powered customer service software, charging per conversation rather than per seat. The company plans acquisitions in North America and Europe to capture enterprise demand for autonomous customer interaction.

Infrastructure changes are evident across hyperscalers. Google Cloud says its Axion instances deliver up to 65 percent better price performance and 60 percent greater energy efficiency than comparable x86 systems. IDC reports that AI-ready datacenter racks have risen from 5 to 10 kW to 30 kW or more, and in some cases exceed 100 kW.

Machine identity risk grows with agent autonomy

As AI agents gain autonomy, they create a new attack surface. Chris Hughes, a cybersecurity expert involved with OWASP’s non-human identity security work, warns that machine identities now outnumber human identities in many enterprises, and agentic AI amplifies that trend. Unmanaged agent identities can lead to privilege escalation or data leakage.

Enterprises must adopt unified orchestration and identity management systems to prevent agents from operating outside governance controls. The convergence of Arm-based efficiency, agentic software, and security is forcing organizations to rethink how they design, deploy and monitor AI at scale.