ChatGPT, foundational and LLMs are wonderful and will admit to many use cases in the future. But, they certainly don't grow out of thin air. The computational requirements for subsequent AI advances are growing exponentially, and certainly faster than the acceleration in individual CPU architecture.
Grid and MPPAs have enabled much progress. They're indispensable, really, because it's hard to justify, much less command, the computing power to be able to do the increased heavy lifting required, as the following chart hints.
We will need new architectures and paradigms to allow for continued advances. Quantum computing is still in its infancy, but it points towards the future.
In the meantime, don't dismiss the incumbent chip makers, foundries, and IP. It's still the best we got.
Commentaires