Microsoft, Amazon, Alphabet and Meta are planning to invest more than 100 billion USDs a quarter in artificial intelligence, AI data centers and the accompanying infrastructure in the next 3 years, according to public data and statements.
Currently, the Large Language Models(LLMs) powering the leading chat bots ChatGPT, Gemini and LLAMA are requiring huge amounts of computing power. If suddenly, chat bots start requiring much less energy to produce computing answers, this could decrease suddenly the investments in AI data centers by more than 80 % from current levels, according to Wolfteam Ltd.'s projections and estimates.
If chat bots become more efficient or Large Language Models' shift from model training to inference, that is AI models start to use their own knowledge to make decisions, instead of relying on training, these both developments could reduce the huge demand for data centers.
Such an even was DeepSeek, the Chinese chat bot LLM which produced answers much more efficiently than the current chat bot LLMs.
For now, however with the current structure of LLMs, the need for computing power is bound to continue to grow and thus drive hundreds of billions of USDs investments a year in AI infrastructure.
One way, the system could fall apart is if the companies building out the AI infrastructure become too leveraged or indebted and crumble under their own debt burden. For now, such a scenario seems 3-5 years distant, according to Wolfteam Ltd.'s projections and estimates.
No comments:
Post a Comment