Developing and maintaining advanced artificial intelligence systems can be financially challenging. Likewise, spending on the much-needed AI infrastructure is not expected to slow soon. That’s because big tech companies are expected to spend more than $500 billion by early next decade as the focus shifts to running AI models rather than training them.
High-end technology, such as GPUs and TPUs, are necessary for building and running AI models and training big AI models. These parts are costly as they cost thousands of dollars and require frequent maintenance and upgrading. Operational costs are further increased by the processing and storage capacity needed to handle large datasets for model training.
In addition to hardware, companies must contend with personnel expenses since hiring and keeping specialized AI talent such as researchers, engineers, and data scientists comes with extremely competitive pay that is frequently greater than that of other IT industries.
Hyperscale companies or the big tech giants in the US are on course to spend $371 billion in building out data centers and other computing resources in 2025. According to a new study by Bloomberg Intelligence, the amount is expected to increase by over $525 billion by 2032. The surge starkly contrasts initial concerns of a shift of focus into developing cost-effective AI models following DeepSeek revelations.
“Capital spending growth for AI training could be much slower than our prior expectations,” Mandeep Singh, an analyst with Bloomberg Intelligence, wrote in the report. But the immense amount of attention on DeepSeek, he wrote, will likely push tech firms to “increase investments” in inference, making it the fastest-growing segment in the generative AI market.
The introduction of the DeepSeek models raised concerns about the necessity of funding AI infrastructure, but it also increased interest in reasoning models, which demand higher inference costs. Bloomberg analysis predicts that by 2032, training-related expenditures will account for only 14% of hyperscalers’ AI budgets, down from over 40% this year. On the other hand, about half of all AI spending that year may come from inference-driven initiatives.
Source: https://finance.yahoo.com/
