Infrastructure gap: 80% adopt GenAI, 17% keep pace
A10 Networks' report finds about 80% of tech firms have adopted GenAI, but only 17% say their infrastructure has kept pace, highlighting a widening gap between use and operational readiness. As a practical workaround, some practitioners suggest introducing a Model Context Protocol (MCP) server above non‑PII data layers to let analysts query multi‑system data in plain English without exposing schemas. Together the numbers and the MCP idea point to two simultaneous needs: infrastructure scale and simpler safe access patterns for business users. (x.com, )
A lot of companies already put generative artificial intelligence into real products, but their pipes still look like they were built for ordinary web traffic. In A10 Networks’ March 25, 2026 writeup of its 2025 survey, about 80% of tech firms said they had adopted generative artificial intelligence, while only 17% said their infrastructure had kept pace. (a10networks.com) That mismatch shows up in very physical places: servers, networks, storage, and load balancers. A10’s broader 2025 report says 76% of organizations are already using generative artificial intelligence, but only 19% have automated scaling for artificial intelligence workloads. (a10networks.com) Automated scaling is the part that adds more computing capacity when demand spikes, like opening extra checkout lanes when a store suddenly fills up. If only 19% have that in place, a popular chatbot or coding assistant can turn from “instant” to “wait 20 seconds” the moment usage jumps. (a10networks.com) The strain is not just speed. A10 says 49% of respondents named security as the biggest pain point, and 53% were only somewhat confident their setup could handle future artificial intelligence needs. (a10networks.com) That helps explain why adoption can outrun readiness. A team can buy a model application programming interface in a week, but rebuilding identity controls, traffic management, observability, and data governance across a company can take quarters. (gartner.com) (a10networks.com) One response is to make the infrastructure bigger. A10 says 79% of organizations plan to modernize infrastructure within 18 months, and 42% already use hybrid cloud for artificial intelligence workloads so they can mix on-premises control with cloud burst capacity. (a10networks.com) The other response is to make access simpler. Instead of asking every analyst to learn table names, field names, and five different software interfaces, some teams are putting a Model Context Protocol server in front of approved data sources. (anthropic.com) (modelcontextprotocol.io) Model Context Protocol is an open standard that lets an artificial intelligence assistant connect to outside tools and data in a consistent way. Anthropic introduced it on November 25, 2024, and the specification describes servers as the services that provide context and capabilities to the model. (anthropic.com) (modelcontextprotocol.io) In practice, that means a company can expose a narrow, safe menu instead of handing over the raw database map. An analyst can ask for “last quarter renewals by region” in plain English, while the server decides which approved system to query and which fields to hide. (anthropic.com) (modelcontextprotocol.io) The “non personally identifiable information” part is important because it limits what the assistant can touch. If the server sits above non sensitive sales, inventory, or operations data, a company can widen access for business users without exposing customer identities, payroll records, or the full schema of internal systems. (modelcontextprotocol.io) (a10networks.com) So the story is not that companies lack artificial intelligence tools. The story is that one race is about bigger infrastructure and the other is about safer shortcuts, and right now the first race is lagging while the second is becoming a practical workaround. (a10networks.com) (anthropic.com)
