In recent years, more and more companies are outsourcing data storage to cloud services. For example, if a company hosts half of its data in the cloud, at an average price of $0.015 per gigabyte per month, its storage costs may increase by $10-50 (“Pricing examples,” 2024)each month.
For a small company with typical data generation patterns, cloud storage costs can range from hundreds to potentially over a thousand dollars per month (Fig. 1.3-5) in a few years, creating a potentially significant financial burden.
According to Forrester’s study “Enterprises Outsource Data Storage as Complexity Grows” (M. Ashare, “Enterprises outsource data storage as complexity rises,” 10 May 2024), which surveyed 214 technology infrastructure decision makers (M. Ashare, “Enterprises outsource data storage as complexity rises,” 10 May 2024), which surveyed 214 technology infrastructure decision makers, more than one-third of organizations are outsourcing storage to handle the growing volume and complexity of data operations, with nearly two-thirds of enterprises preferring a subscription-based model.
The situation is further complicated by the accelerated adoption of cloud-based technologies such as CAD (BIM), CAFM, PMIS and ERP -systems that further increase data storage and processing costs. As a result, companies are forced to look for ways to optimize costs and reduce dependence on cloud providers.
Since 2023, with the active development of large language models (LLM), approaches to data storage have started to change. More and more companies are thinking about taking back control of their data as it becomes safer and more profitable to process information on their own servers.
In this context, the trend away from cloud-based storage and processing of only the necessary data in favor of local deployment of enterprise LLM and AI -solutions comes to the fore. As Microsoft’s CEO pointed out in one of his interviews (JETSOFTPRO, “SaaS is Dead? Microsoft CEO’s Shocking Prediction Explained,” January 13, 2025), instead of relying on several separate applications or cloud-based SaaS solutions to perform different tasks, AI agents will manage processes in databases, automating the functions of different systems.
[…] the old approach to this [data processing] issue was: if you think back to how different business applications handled integration, they used connectors. Companies sold licenses for those connectors, and the business model was formed around that. SAP [ERP] is one of the classic examples: you could only access SAP data if you had the right connector. So it seems to me that something similar will emerge in the case of [AI] agent interaction […]. The approach, at least that we take, is: I think that the concept of the existence of business applications will probably collapse in the era of [AI] agents. Because if you think about it, they are essentially databases with a bunch of business logic
– Satya Nadella, Microsoft CEO, interview with BG2 channel, 2024 (BG2 Pod, “Satya Nadella | BG2 w/ Bill Gurley & Brad Gerstner,” December 12, 2024)
In this paradigm, the data-driven LLM approach goes beyond classical systems. Artificial intelligence becomes an intermediary between the user and the data (Fig. 2.2-3, Fig. 2.2-4), eliminating the need for multiple intermediary interfaces and increasing the efficiency of business processes. We will talk more about this approach to working with data in the chapter “Turning Chaos into Order and Reducing Complexity”.
While the architecture of the future is still taking shape, companies are already facing the consequences of past decisions. The massive digitalization of recent decades, accompanied by the introduction of disparate systems and uncontrolled accumulation of data, has led to a new problem – information overload.
