Meta at the AI Crossroads

A couple of days ago, NYT reported that Meta will delay releasing its new foundational AI model to May this year. While Meta never really publicly disclosed when they would likely release their new model, it was widely believed to be sometime in March. However, NYT’s reporting indicated that while the model outperformed Llama 4 as well as Gemini 2.5 (released in March 2025), it fell short of Gemini 3.0 which was released in November 2025.

Not being able to match Gemini 3.0 is certainly a blow to Meta’s ambition in staying at the frontier. While Meta spokesperson tried to downplay the reporting, I do think the reporting was incremental. If they get there by May, Meta will be six months behind the frontier models which has been sort of consensus timeline. But if they cannot match Gemini 3.0 even by May or delay it again, the gap between Meta and the frontier models will diverge from the consensus timeline. Given the Llama 4 fiasco, Meta’s ambition as a model developer was already under the scanner and this news certainly did not help.

In light of this news, some investors started to wonder whether Meta even needs a SOTA model. Interestingly, Alex Heath reported yesterday that “An interesting question that has been asked internally is whether Meta even needs to compete at the frontier of AI”. I will share my thoughts on this question as well as Meta’s potential layoffs behind the paywall.

In addition to “Daily Dose” (yes, DAILY) like this, MBI Deep Dives publishes one Deep Dive on a publicly listed company every month. You can find all the 66 Deep Dives here.