49% of enterprise decision-makers say legacy on-premises software has delayed, scaled down, paused or canceled AI projects. The finding comes from a Google Cloud and Hypothesis Group survey of 1,250 US enterprise IT and business decision-makers, and it points to a clear shift: the main barrier to enterprise AI adoption is no longer inadequate model capability, but the drag of legacy systems.
What this is
Here, “legacy software” mainly refers to systems running on companies’ own computers and servers rather than being natively deployed in the cloud. Google Cloud calls the cost of maintaining legacy systems an “innovation tax”: patching, security fixes and routine operations continuously consume IT budgets and engineering time.
The contradiction is that AI is accelerating the discovery of vulnerabilities in old code, but security teams may not have enough staff to handle the resulting increase in alerts. The survey also highlights vendor lock-in: replacing underlying technology can carry substantial migration costs. As a result, even a new tool that performs well in isolation can lose momentum after entering a production environment built around legacy systems, permission workflows and data interfaces.
Notably, Google Cloud says nearly 90% of Fortune 100 companies are already using Gemini Enterprise to build, scale or govern agents. Agents—AI software capable of planning and executing multi-step tasks—have entered the enterprise, but “using” them does not mean core processes have been modernized.
Industry view
We believe this survey offers more practical insight into enterprise AI deployment than another model benchmark. The real competition is not simply about which foundation model a company chooses, but whether it can connect AI to existing data, permissions and business systems. As model capabilities converge, system integration, data governance and operational maturity will determine project returns more directly.
There are also risks in the survey’s boundaries. The data covers only US respondents, and Google Cloud commissioned the research, creating a natural incentive to emphasize the value of cloud migration and unified platforms. Not every legacy system should be replaced; many support audit requirements, stability or specialized business logic. Hasty migration can introduce compliance disruptions, higher costs and security exposure.
A safer path is usually gradual replacement: start with measurable, lower-risk processes, then progressively connect them to core systems. Simply attaching AI to legacy systems as they are may also amplify historical inefficiencies.
Impact on regular people
For enterprise IT: The competitive focus of AI projects is shifting from demo quality toward post-deployment stability, access control and the cost of modernizing legacy systems. IT departments will place greater emphasis on interface governance and full-lifecycle maintenance.
For individual professionals: As agents take on process tasks, repetitive operations will decline, but employees will still need to cross-check across systems, handle exceptions and review results. The ability to explain business rules and validate AI output will become more important.
For the consumer market: Enterprise AI modernization usually begins in customer service, sales and back-office operations. Consumers may experience faster service or more automated processes, but when bad decisions are amplified by systems, access to appeals and human intervention will remain essential.