‍👉 Background: Meta, formerly Facebook is the owner of the Meta family of apps, including Facebook, Instagram, Whatsapp, and of course its little metaverse project which has cost more than $36 billion USD.

‍👉 What happened: Now, Meta has moved on to the next shiny object - AI models. On top of its large language model named Llama, Meta is looking to build its own in-house AI chip ,which would be used to train its own AI systems.

‍👉 What else: If successful, Meta would ramp up production of its AI chip for wide-scale use — and potentially become a major competitor to Nvidia. But Meta has tried to build in-house chips back in 2022, but it failed, and then Meta just ended up buying $10 billion worth of Nvidia’s chips.

‍

‍What's the key learning?

‍

‍💡When companies spend billions on a product, they have two choices: keep paying for someone else's product or build it themselves. It’s a huge upfront investment, but if successful, it can lead to major cost savings in the long run.

💡Meta expects to spend up to $65 billion USD on its AI infrastructure in 2025. And, although they failed in the past, Meta is hoping its AI chip development mirrors what Amazon did with AWS.

💡Amazon initially built AWS to support its own e-commerce operations. But then it realised that it could sell this cloud infrastructure to other businesses, and they turned it into a multi-billion-dollar empire — worth more than 65% of its annual operating income. So, if Meta’s chips perform well, it means they could cut their dependence on Nvidia and it could become a whole new part of their business.