Angehefteter Tweet
Jensen mentioned during $NVDA's earnings call that compute equals revenue. This is purely to sell the narrative. Throwing compute at something doesn’t necessarily produce anything useful to the consumer, let alone profitable. AI compute has reached the point where its cost is straining the system. The exponential growth of compute for only marginally better results will reduce AI profitability, and there will even be periods where more compute leads to worse results.
I foresee AI hitting a local peak of capability in the very near future, if it hasn’t already (in my opinion, this peak was reached around GPT4). What happens then? The same as always with AI: even more compute will be needed to reach the next peak. This is bullish for Nvidia in the short term since they sell more GPUs, but each successive peak requires exponentially more compute.
Throwing compute at the problem thus becomes less sustainable unless ever more money is poured in for smaller and smaller gains. When progress inevitably slows to that point, the point of basically no to little progress, enough becomes enough and the bubble bursts. That’s the risk. Thankfully, there will be signs.
English
