Data has become the “currency” of AI, but it requires a lot of real currency to process it.

In the age of AI, data is seen as “currency.” That’s why the demand for tools to integrate, store, and process data is a growing priority among enterprises, writes IEEE ComSoc. The amount of data generated worldwide is expected to reach 180 ZB by 2025, up from 120 ZB in 2023. And all of this will require a lot of hardware.

The average size of datasets needed to train AI models has increased from 5.9 million records in 2010 to 750 billion in 2023, according to BofA Global Research. In a BofA survey of 150 IT professionals, streaming data processing (44%) and machine learning (37%) were cited as key AI use cases. AI is also driving data migration to the cloud. Gartner estimates that 74% of data management platforms will be deployed in the cloud by 2027, up from 60% in 2023.

Data infrastructure software is the biggest expense for IT departments. Survey respondents estimate that it accounts for 35% of overall IT spending, with budgets expected to grow by 9% over the next 12 months. Public clouds are among the top three providers of such solutions. They also generate a lot of revenue. According to BofA, the infrastructure software industry (storage and lakes, unstructured data, etc.) is currently worth about $96 billion and could grow to $153 billion in 2028.

Image source: BofA Global Research

On the hardware side, BofA forecasts a rapid increase in capital expenditures at Amazon, Alphabet/Google, and Meta/Facebook, rising 43% year-over-year to $145 billion in 2024. Much of the increase will come from servers and hardware:

• Alphabet’s capital expenditures on IT assets will increase by $12 billion year-on-year to $28 billion.
• Meta✴, after a sharp increase in 2023, will not stop and will again increase spending on servers, network and other equipment by $7 billion year-on-year to $22 billion.
• Amazon’s hardware costs will increase by $8 billion year-on-year to $41 billion (due to AWS, retail segment costs will not change).

The researchers also noted that Meta✴ leads in terms of capital expenditure to revenue (% of revenue), and since 2022, the company has been steadily increasing its capital expenditure on AI (its own supercomputer, LLM, etc.). Meta✴’s expenses are comparable in volume to those of larger hyperscalers. One possible outcome could be the emergence of a new cloud solution for advertisers.