META Is The World's Largest VC
I liked META’s previous strategy of building open source language models. 99% of language models will be smaller models - many open source - that are perfectly appropriate for managing the task at hand. Some will be multi-faceted, others will specialize in a particular vertical, having been trained on vertical-specific data sets.
However, META has abandoned the open source approach in favor of developing frontier models. The problem is that META keeps shifting its AI team around and Zuckerberg’s vision is not clear to me. I like META’s smart glasses effort, but beyond that, I don’t see META’s path for delivering a return on its CapEx spend. META has invested $48 Billion YTD in CapEx. That number will grow next year, but META has nothing to show for it, which is why I compare META to a venture fund.
If META wishes to sell “AI” services to Enterprises, it should acquire Enterprise data rather than mess around with proprietary closed models. Acquire CoStar (CSGP) for CRE data to train its models with. Acquire Verisk (VRSK) for P&C insurance data. Acquire Bloomberg, ICE (ICE) or the LSE (LSEG) if it wants to deploy AI services across Market Data and related services and so on.
I believe all of the frontier LLM companies are putting the cart before the horse and should invest in modernizing economic infrastructure via Web3 / Blockchain technology. Once the transaction chain is digitized, AI agents and tools will then be positioned to deliver the type of automation we dream about.