Following on from xAI launching its Grok 4 mannequin, which it claims is now the best-performing AI system on all fronts, Meta has introduced the following steps in its AI growth journey.
And it’ll be robust for some other AI developer to match or higher this.
As we speak, Meta CEO Mark Zuckerberg has introduced that Meta’s trying to set up “a number of” multi-gigawatt datacenters within the U.S., starting with its “Prometheus” undertaking in Ohio, which will likely be adopted by its new “Hyperion” cluster, which is being inbuilt Louisiana.
Each of those new mega-clusters will set up main new compute energy for Meta’s AI tasks, and can price, in line with Zuckerberg, “a whole lot of billions of {dollars}” to finish.
Which is a mind-boggling sum of money, greater than the GDP of some nations.
And this, in line with Zuck, is just the start.
Meta’s Prometheus knowledge heart will likely be its first main AI knowledge undertaking, geared toward powering Meta’s new “superintelligence” push. The undertaking will incorporate the capability of some 600k NVIDIA H100 models, which is able to give Meta a significant benefit in compute.
These NVIDIA chips are key for AI growth, enabling large compute energy to drive the decision-making of AI techniques, which scan by trillions of information sources to provide you with the absolute best solutions to queries.
For comparability, each OpenAI and xAI reportedly maintain round 200k H100s every.
That alone will give Meta a transparent lead, however Meta’s additionally constructing extra knowledge processing capability on prime of that, because it appears to be like to transcend the restrictions of the most recent generative AI fashions.
Which is able to ideally, for Meta not less than, lead us nearer to the event of synthetic common intelligence (AGI), and machines that really “suppose” like people, and replicate the human mind in circuitry.
The present wave of AI fashions, whereas spectacular, should not near this, as they’ll solely replicate what they’ll discover in knowledge patterns primarily based on human inputs. True AGI goes a step past that, and will ultimately develop human growth, by exploring prospects past human pondering.
Which is able to both take us to the following stage of civilization’s growth, or mark the top instances for humanity. No one is aware of which path they’ll take us on as but.
However Meta’s decided to search out out, which is why Zuckerberg has put collectively a workforce of main AI specialists to discover the potential of AI in all new methods.
These new investments in AI infrastructure will help that effort, and can price Meta a large quantity of capital within the quest to find out what’s doable on this entrance.
Which feels barely regarding, but additionally inevitable, to some extent.
To be clear, the present generative AI instruments are nowhere near AGI, and there’s skepticism as as to if it’ll even be doable to simulate human-like pondering in machine techniques.
However Meta’s seemingly going to search out out, with Zuckerberg noting that:
“Meta Superintelligence Labs may have industry-leading ranges of compute and by far the best compute per researcher.”
It has the cash, the info, and now the employees to discover this next-level push.
Which, within the brief time period, may depart each different AI undertaking within the mud. However in the long run, you may anticipate to see quite a lot of spooky tales about AI figuring out the inefficiency of people stemming from Meta’s next-level analysis.
Meta’s first mega datacenter, “Prometheus,” is anticipated to be absolutely activated early subsequent yr, so each different AI undertaking will probably be trying to collect up funding and alternative earlier than that time.
At which stage Meta appears to be like set to change into the main AI undertaking, by an unmatchable margin.
