The Nvidia processors, he explains, are for processing large, giant language fashions (LLMs), whereas the Google TPU is used for inferencing, the following step after processing the LLM. So the 2 chips don’t compete with one another, they complement one another, in line with Gold.
Promoting and supporting processors might not be Google’s core competence, however they’ve the abilities and expertise to do that, mentioned Alvin Nguyen, senior analyst with Forrester Analysis. “They’ve had their TPUs accessible, from what I perceive, to some exterior firms already, primarily startups from ex-Googlers or Google-sponsored startups,” he mentioned.
As to the rumor of the Meta buy, the query is, what does Meta wish to do with them, mentioned Gold. “In the event that they’ve already constructed out a mannequin and so they’re working inference workloads, then Nvidia B100s and B200s are overkill,” he mentioned. “And so what are the choices there are actually? There are a selection of startups which can be attempting to do inference-based chips as properly, and Intel and AMD are transferring in that course as properly. So it truly is a perform of getting a chip that’s optimized for his or her setting, and once more, Google’s TPUs which can be optimized for a hyperscaler cloud sort of setting.”
Nguyen says it’s one factor to make their very own chips for their very own use, it’s one other factor to be promoting them that’s an infrastructure and a competency that they don’t have, and Intel and AMD and Nvidia are approach forward of them in that regard.
“Sure, they know the best way to do it for themselves, lengthy as you have been speaking about as a service or as a cloud service. For on premises, or for individuals who wish to take it for themselves, that’s a muscle reminiscence they need to develop,” he mentioned.
For that motive, he doubts that different hyperscalers with their very own customized silicon will go into the chip promoting enterprise for themselves as properly. “There’s nothing to cease them, however every of them has their very own challenges,” mentioned Nguyen. Microsoft, AWS and OpenAI all have a number of partnerships and would inevitably find yourself in competitors with any individual.
