The influence of reasoning AI models from DeepSeek and OpenAI is anticipated to redirect the focus of investments in artificial intelligence (AI) while simultaneously increasing overall AI spending. The launch of DeepSeek models prompted observers to reconsider the necessity of investment in AI infrastructure. However, it also heightened attention on reasoning models, which necessitate increased spending on inference, as reported by Bloomberg on March 17. Consequently, Bloomberg Intelligence now predicts that investments in AI from major companies like Amazon, Meta, and Microsoft will grow at a faster rate than previously anticipated, with a larger portion of that spending allocated to operating AI systems post-training rather than on data centers and chips. These corporations are projected to invest $371 billion in data centers and computing resources by 2025, representing a 44% increase from expenditures in 2024, and could reach $525 billion per year by 2032, according to the report. By 2032, it is expected that nearly half of all AI spending will be focused on inference, as reasoning models help companies generate greater revenue from software. Simultaneously, the investment share for training is forecasted to decrease from 40% to 14%. DeepSeek made a significant impact on the AI landscape in late January by launching AI models that matched the performance of those from OpenAI and Google at a significantly lower cost and with considerably fewer Nvidia GPUs. Following DeepSeek’s groundbreaking model release, Meta CEO Mark Zuckerberg noted in an earnings call that the U. S.

AI sector is shifting toward AI processing, or inference, as the popularity of reasoning AI models grows. In February, OpenAI introduced what it described as its “most cost-efficient” reasoning AI model, called the o3-mini, highlighting its “small” yet “powerful and fast” capabilities, especially in science, coding, and mathematics. This model offers three levels of reasoning—low, medium, and high—for more challenging tasks. Belonging to OpenAI’s o1 series, the model can reason through various tasks, although it has longer response times compared to non-reasoning models. Nonetheless, reasoning models are capable of addressing more complex tasks and solving tougher problems.