Why Buy the Llama When You Can Get the Model For Free?

Microsoft, Google, and Amazon all have a fairly straightforward pitch for their CapEx spend: AI will increase the top-line sales of the company and eventually the bottom-line profit. But with the other big captain of AI CapEx, Meta, the math has always been decidedly more murky. That's because they neither sell their AI models nor sell cloud services that would directly take advantage of said models.

But Meta is always quick to note how the open-sourcing of Llama will have "halo" benefits for the company, just as they've seen with open sourcing their data center projects and things such as PyTorch. And, sure. Undoubtedly, such a move should help to drive down costs while at the same time, helping Meta competitively if Llama gains widespread adoption by others. And there's a bonus side effect that it would hurt their competitors offerings by creating a massive amount of pricing pressure, as long as Meta's models are on par with their rivals (and they're close, if not quite there, despite Mark Zuckerberg's assurances).

On paper, that all makes sense. Again, especially because Meta has run this same general playbook before. But there's also a big difference on that piece of paper...