According to an IDC study with Intel released on 28 October 2026, 89% of IT leaders see Agentic AI as a key priority in the next 12 to 24 months. VietnamNet confirmed the figure the same day. Yet only 56% of organizations have enough budget for Agentic AI infrastructure. The gap between ambition and resources is now the biggest bottleneck for the agentic AI wave.
How an agent works differently from a chatbot
A chatbot takes one question and gives one answer. An AI agent works differently. It retrieves a company’s own data, triggers models, calls business applications, executes code and applies security policies, repeating the cycle many times until a complex task is complete. An agent can process an order, reconcile a set of documents or prepare an analysis without being prompted at every step.
Agentic AI is therefore not an add-on feature but a new way of working. It runs continuously and touches many systems at once, so the infrastructure behind it must be strong, secure and always available.
Budget is the biggest bottleneck right now
Ambition is high, but real money has not caught up. Nearly half of organizations want to run AI agents but lack the budget to buy, upgrade or operate the infrastructure for them. The study also found that 45% of organizations identify modernizing or consolidating data center infrastructure as their top IT priority in 2026. In other words, before running AI agents at scale, many companies must first tidy up their own infrastructure, bring scattered systems together, upgrade old servers and reorganize data.
On-premises infrastructure stays central
The study shows 55% of organizations keep a balance between on-premises infrastructure and public cloud, while 38% mainly or fully prioritize on-premises infrastructure. Together, most companies still put on-premises infrastructure at the center rather than moving everything to the cloud. The reason is straightforward. An AI agent must retrieve proprietary data continuously, and every run touches internal data. Keeping infrastructure close to the data gives companies better control over who can access it and what it costs to run each month.
Flexible hardware instead of betting on one chip
On hardware, companies are pragmatic. Only 27% see the GPU as decisive for performance, 23% value the CPU, and about half say performance depends on context. No single chip wins every workload. That is why about 40% of buyers prioritize standard servers with matrix or vector acceleration or support for PCIe cards. They choose flexible servers that can take an accelerator card when needed instead of betting on a single architecture, saving budget today while keeping an upgrade path open for next year.
Sources: IDC and Intel study, released 28 October 2026; VietnamNet, 28 October 2026; amtech.vn, 28 October 2026.
