I=4
Chinese regulators deepen investigations into acquisition of Manus by Meta Platforms, raising the risk that it could unwind the sale
However, the investigations remain preliminary and the regulators may choose not to take any action.
Meta Platforms acquired Manus at the end of December for $2 billion.
I=6
Meta’s Avocado LLM has outperformed leading open-source base models even before post-training refinement, internal memo shows
Assessment
The fact that they haven’t started post-training refinement could signal that the models may not arrive in Q1 2026.
I=6
Meta’s Indiana data center to cost $10 billion and will have a capacity of 1GW
I=5
Meta signs multiyear, multigenerational strategic partnership with Nvidia for computer chips
I=7
Meta to buy AMD Chips worth $60 to $100 billion over the next 5 years
I=5
Meta’s Chief AI Officer reiterates that new models will come in the next couple of months
“We are releasing new models this year, with the first coming in the next couple of months,” Alexander Wang said (min 4:53). “We’re optimistic about the trajectory we’re on. The first models will be good and as the year goes on, I think we’re going to be pushing the frontier,” he added.
I=6
Ruoming Pang, hired 7 Months ago to lead Meta AI Infrastructure, departs to OpenAI
Pang joined Meta 7 months ago on a compensation package worth valued at more than $200 million over several years.
Assessment
Pang was an important executive in the MSL team and his departure shows how hard it is for tech companies to maintain such high-profile AI executives. Hopefully, his departure has nothing to do with the division’s culture such as issues arising due to competition for compute.
I=6
Meta signs multi-billion deal to rent AI chips from Google
Meta is also in talks with Google to buy the TPUs for its data centers as early as next year, the Information reported citing people familiar with the matter.
I=6
Meta’s Chief AI Scientist Alexander Wang said we will see incredible velocity from MSL in the coming months
Meta’s Chief AI Scientist Alexander Wang said in an interview that they are now done with MSL structuring and that over the coming months we will see incredible velocity from MSL (min 9:28).
He said they are excited about personal agents as they represent a big opportunity for AI, adding that Manus recently released agents that are working 24/7 on behalf of users (min 6:16).
I=5
Meta tests AI shopping chatbot, rivaling that OpenAI’s ChatGPT and Google’s Gemini
Meta is testing a shopping research feature in Meta AI that provides product suggestions and information such as images, website, and price.
Bloomberg News tested the feature and found that its recommendations are tailored to what Meta already know about you such as location and gender.
I=7
Meta creates a new Applied AI Engineering Organization to help MSL models get better and faster
Assessment
It seems like Meta is trying to cap the powers of MSL and that of its Chief AI Scientist Alexander Wang. Earlier on, Alexander Wang was the central AI authority but part of that authority is now shifting to CTO Andrew Bosworth.
This comes after it was reported that tensions are emerging between Meta’s core team and Meta Superintelligence Lab (MSL) over training data and compute. Former Meta’s Chief Scientist Yann LeCun also described Wang as “young and inexperienced”. Therefore, I think the creation of this parallel engineering team is intended to increase collaboration between the core team and the AI team. However, it may also signal that MSL’s execution is not as strong as reported.
I am not sure if I agree with the assessment here. Doesn’t it make sense to have support processes like procuring data and interfaces/tooling in another decision so that MSL can fully focus on the models itself?
Yes, it makes sense and is positive. I am just concerned that restructuring is ongoing yet they should be at a stage where they are executing intensively (but may be it’s normal). Such restructuring may see some members of MSL switch to the new organization or leave the company-due to constant restructuring, interrupting execution.
I=4
Meta acquires Moltbook, a viral social network for AI agents
I=8
Meta delays the launch of its latest AI model to May from this month due to performance issues, NewYork Times reported
Meta is once again delaying the launch of its latest AI model (Avocado) to May from this month after it failed to outperform those of Google, OpenAI and Anthropic on internal tests for reasoning, coding and writing, NewYork Times reported citing three company sources.
The sources said Avocado did better than Gemini 2.5 but has not performed as strongly as Gemini 3.0.
A Meta spoekesperson seemed to confirmed the report as he signaled that the release will mainly be about trajectory they are in.
“As we’ve said publicly, our next model will be good but, more importantly, show the rapid trajectory we’re on, and then we’ll steadily push the frontier over the course of the year as we continue to release new models. We’re excited for people to see what we’ve been cooking very soon.”
https://www.nytimes.com/2026/03/12/technology/meta-avocado-ai-model-delayed.html
I=6
Meta will spend $27 billion over 5 years on Nebius AI infrastracture
I=3
Meta’s FAIR recently introduced TRIBE v2 (Trimodal Brain Encoder), which aims to predict how the human brain responds to almost any sight or sound
More insights on it will be added…
I=4
Meta recently launched SAM 3.1 that introduces object multiplexing, enabling faster video processing with no accuracy trade-off
I=5
Meta forms an elite AI Research team for its recommendations division
Assessment
I think it’s a good strategy given Meta’s LLMs currently don’t play a major role in Meta’s recommendation systems.
