So it began with apps, and now it’s on to AI?
Proper. We’ve taken that very same end-to-end precept and utilized it to AI. Should you’re attempting to ship worth with AI, you may both purchase a bunch of disparate components from completely different distributors and attempt to sew them collectively your self, or you may search for an built-in system the place all the things you want is already related.
What disparate components can somebody sew collectively to make a full AI stack?
An intentional AI stack wants a cohesive mixture of layers to get a job achieved: compute infrastructure, an AI mannequin, an orchestration platform and the consumer interfaces. At Google, we’ve intentionally invested in each single layer. We offer the {hardware} like Tensor Processing Units (TPUs), frontier fashions developed by Google DeepMind like the Gemini family of models, the Gemini Enterprise Agent Platform and the interfaces folks use each day, like Maps and Gmail. We’ve basically achieved the trying to find you and put all the required parts proper contained in the field.
Did we all know we needed to have a full-stack method means again when Google first began engaged on AI?
It was completely a deliberate, decades-long technique. As an example, our wager on customized TPUs is already over 10 years outdated. We acknowledged early on that there’s large worth in proudly owning our personal provide chain and uncooked infrastructure when serving up the world’s most vital web companies. Proudly owning that thread all through all the stack lets us ship a degree of service, efficiency and reliability that is very laborious to realize for those who’re on the mercy of a number of events.
On the flip aspect, does adopting a full-stack platform restrict builders indirectly?
That’s a particularly reasonable concern, however locking folks in would not align with our ethos. No firm does open supply fairly like Google; we usually give away foundational know-how and supply code that all the trade relies on.
We like to explain our AI platform as “opinionated however extensible” and “batteries included” — that means all the things it is advisable construct and run an utility is able to exit of the field. Nonetheless, if you wish to use one other firm’s AI mannequin as an alternative of Gemini, or hook up completely different software program as an alternative of Google Workspace, you may plug these proper in. We would like you to make use of our merchandise each day primarily based on the completeness of our platform, not as a result of we pressured you right into a closed selection.
