What are experts saying on what data platforms look like in the next 2-5 years?

Experts anticipate a fundamental shift in data platforms over the next two to five years, evolving from passive "systems of intelligence" to active "systems of action" driven by AI agents [1, 4]. Historically, only **10-20% of data insights** were ever successfully operationalized [4, 5]. The new paradigm aims to close this gap by deploying "swarms of agents" that translate real-time data insights directly into automated business outcomes, creating a form of "intent-driven engineering" . This transition is expected to cause a significant increase in data processing, with some experts forecasting a **10-20x increase** in workloads and API calls, requiring vertically integrated infrastructure to manage costs [1, 26]. These agents will not only respond to user prompts but will also be triggered by system events like server crashes or security incidents, further embedding them into core business operations .

This agent-driven future redefines the requirements for the underlying data foundation, elevating the importance of business context . While data quality remains crucial, experts argue that rich business context—the implicit logic and semantic meaning previously held by human experts—accounts for **50% of an AI agent's accuracy** . To support this, enterprises are pursuing strategies to create a unified, harmonized data layer, which is seen as a prerequisite for any successful AI implementation [13, 25]. This is being achieved through two complementary approaches: consolidating data into a single data warehouse for easier agent access and leveraging open standards like Apache Iceberg to enable cross-cloud data federation, which allows querying data in-place across providers like AWS, Azure, and Snowflake without costly data movement [14, 22]. Concurrently, the demand for training data is evolving, with a predicted need for "super STEM" data for next-generation models and an expected rise in the value of human-generated data points as synthetic data improves [2, 29].

The physical infrastructure supporting these platforms is poised for a radical transformation, with one expert predicting the entire computing stack will be "unrecognizable" within five years . This overhaul is fueled by massive capital investment, with plans to spend an additional **$3 to $4 trillion** on data center capacity in the US alone over the next five years . This spending boom is largely driven by calculations of the inference capacity required to provide every person with a ChatGPT-like product [15, 19]. While AI today often relies on a centralized, mainframe-like model, some predict a shift over the next three years toward a more distributed architecture with varied form factors . More speculative predictions suggest that data centers in space could become the most important AI infrastructure development in the next three to four years to support future off-world activities [7, 16, 21].

These technological shifts are creating new competitive dynamics and strategic tensions. On one hand, platform companies like Snowflake and Databricks aim to become the indispensable foundational layer for enterprise data and proprietary AI, creating a strong competitive moat [9, 12]. On the other hand, a countervailing force is emerging, with some experts predicting that new AI-driven tools will **dramatically lower the switching costs** of data between vendors, threatening the incumbent SaaS model . This tension between platform lock-in and AI-enabled data portability will define the market. The stakes are high, with one CEO predicting that within a few years, the world's technology infrastructure will consolidate and run on either a US or a Chinese AI stack, highlighting a significant geopolitical dimension to the platform race [3, 23].