“Scaling compute” is AI industry jargon for spending more and more money on data centers and the chips (typically GPUs made by Nvidia) inside those data centers. So called hyperscalers like OpenAI, Alphabet, Amazon, Meta, Microsoft, xAI and Anthropic aim to spend nearly a trillion dollars on compute this year, more than a dozen times the cost of the Manhattan project. China is the next biggest global spender and the EU are accelerating rapidly. In this talk, Gary will argue that this is a bad bet, unlikely to pay off, filled with numerous downsides for society, and that it may well lead to a recession.
—