Broadcom: Three 1 Million XPU Clusters By 2027

Many believe AI is in a hype cycle. Nothing could be further from the truth.

Hype cycles result when the expectation upon a technology exceeds the capabilities. Another way of looking at it is the features that hit the market simply are not there.

While there is a lot of hype surrounding AI, specifically generative models, we are seeing features hitting the market. This week, X made Grok available to everyone on the platform. This not only includes text prompting but also image generation.

Of course, this is one company in a field of many that are involved. We see all the search engines incorporating AI to some degree. social media applications will soon be full of these services.

And to put a cherry on top, the rollout of AI agents is just beginning.

Hence, there is no hype. What is taking place, however, is a lot of money being tossed around. Nowhere is this greater than with compute.


Image generated by Grok

Broadcom: Three 1 Million XPU Clusters By 2027

Broadcom doesn't get the attention that NVIDIA does. That said, this is a formidable player in the field.

On the most recent earnings call, the CEO opined the expectation of there being three 1 million XPU clusters online in 2027. As a comparison, xAI is making news with its 100K cluster in Memphis.

We are looking at roughly a 10x over what is out there at this time.

So who is involved with this? While it is hard to know exactly, it is mostly the expected players.

Meta, Google, and perhaps, Amazon are involved. The first two have collaborations with Broadcom.

The expected market size is expected to be $60B-$90B.

An Acceleration of Compute

Most recent studies have the amount of AI compute doubling every 6 months. There is no reason to believe this is going to slow. In fact, as these numbers show, it is likely to continue for a number of years.

If AI is going to reach its potential, a lot more compute is required. This is not only for the training of models but inference. Jensen Huang even came out with his forecast that inference is 1 billion times larger than training.

This makes sense when we consider that inference is used each time a prompt is done. Adding in real world AI, such as autonomous driving along with personal agents, and we can see how the demand will explode.

General purpose technologies (GPT) affect the entire economy. Traditionally we were dealing with things as electricity and steam power. The impact was felt far beyond a single industry.

Generative AI is similar.

It is going to be the foundation of society. To say this is going to have the impact that electricity did is not an overstatement. We are going to see society change completely over the next couple decades.

Broadcom is just one of the companies to keep an eye on. It is at the core of what is taking place.


What Is Hive

Posted Using InLeo Alpha

Sort:  

It could be the race of this century for which big firms get to capitalize most on the future of AI.