The thing that's really different about a transformer ASIC versus a GPU is that a GPU is a general-purpose chip. It has to be able to run any kind of workload. A transformer ASIC, by contrast, can be completely specialized for one thing, which means you can make very different design decisions.
The core insight is that transformers have a fixed architecture, and if that architecture is going to be with us for a long time, there's a massive opportunity to build a chip that does nothing but run transformers — and does it faster and cheaper than anything else. A general-purpose chip has to be flexible, and flexibility costs you in efficiency.
The thing that's really interesting about transformers is that they've kind of eaten AI. So many of the workloads that people care about are transformers. And because the transformer is a fixed algorithm, you can etch it into silicon — you can build a chip that only runs transformers, and it will be orders of magnitude faster and cheaper than a chip that can run any algorithm.
It doesn't give you a lot of confidence in like the strategy overall. They're signing these Neo cloud deals worth tens of billions of dollars. They're built, you know, spending hundreds of billions of dollars. And yeah, they can make the argument that these type of um like doing any type of NeoCloud deals themselves is just good business. It's it's just like how it's just the best way to get ROI today. It doesn't give you a lot of confidence that there's near-term products on the horizon for Meta that are going to be able to utilize that capacity themselves.
There's a lot of pressure to bring this new energy online very quickly because states are currently competing for the opportunity to attract these data centers, even though the financial benefits of them for the state has not actually yet borne fruit.
