Dive Brief:
- Enterprise demand for AI agents is driving significant changes in infrastructure needs, according to an S&P Global report published Wednesday.
- “Agentic systems consume significantly more IT capacity than chat-based systems as they break free of human pacing and launch multiple prompts and cascade into other agents,” S&P Global said in a release accompanying the report.
- GPU shipment projections from leading suppliers have increased more than 500% for the 2025-2026 period compared with initial estimates released in 2023. The rapid investment growth comes as most organizations aim to deploy agentic tools.
Dive Insight:
As more enterprises actively pursue agentic AI adoption, the infrastructure considerations needed to support the technology are coming into clearer view.
A capacity crunch is already at play, pushing CIOs to plan as far as five years in advance for infrastructure needs. Costs are also on the rise, both in terms of the price tag for deploying the technology and for running it.
Further change is on the horizon.
“While agentic AI promises significant operational improvements, organizations face a critical infrastructure overhaul to support autonomous systems that can initiate actions independently without human prompting,” S&P Global said. The tools are bringing a new set of capacity demands and security considerations.
Most infrastructure strategies moving forward will require data flows and interaction across hybrid environments, S&P Global said in the report.
“For some, that will mean integration across on-premises and in the cloud, a development that has already required more networking capacity to handle the movement of models and data,” the report said. Enterprises will also need to consider improving partnerships with key providers to establish trust.
After all, agentic AI offerings were not all created equally. IT leaders grapple with varying definitions of the technology from vendor to vendor, technical immaturity and hype-filled claims. Identifying trusted providers and strengthening the relationship with these businesses is one common strategy that lets enterprises hedge risk while exploring nascent technology, as reported by nearly three-quarters of leaders in a KPMG survey published in September.
Vendors have stepped up to the plate. Microsoft updated its Copilot Studio to ease AI agent development at the end of last year. ServiceNow released a no-code, low-code AI agent builder in March. Databricks beefed up its agent-building offerings in June. Salesforce unveiled its newest platform to ease AI agent building in October.
Even OpenAI, which typically focuses more closely on AI model making, has released tools this year to ease the development of AI agents.
Enterprises have never had as many options for adopting agentic AI, according to Tim Sanders, chief innovation officer at G2.
“They’ve made it really easy to stand up an agent,” Sanders told CIO Dive. “Out of all the different ways to get agents for a company, these agent builder platforms that use no code or simple prompting are No. 1, and I would anticipate they’ll actually gain in their leads.”
