It took less than a year for Alexandr Wang to ship something. In September 2025, Meta announced Meta Superintelligence Labs with the Scale AI founder at the helm. Now, April 2026, Muse Spark is here — and it's already running inside Meta AI and the company's new "shopping mode" feature.
What Muse Spark Actually Does
The model powers queries across Meta's family of products. That's significant because it means Meta has crossed the threshold from "we're experimenting" to "this is in production across billions of users." The model is being released under an open-source license — a move that echoes Meta's Llama strategy but with a sharper focus on utility rather than pure research.
This isn't some distant R&D prototype. This is the model your aunt is using when she asks Meta AI for recipe recommendations. The co-occurrence data is brutal: Meta AI + Muse Spark spiked with a z-score of 13.27 — that's not a test, that's a product launch.
Meta is pulling top engineers from across the company into its new Applied AI Engineering division as part of a push to "compete in the AI race."
The $21 Billion Footnote
While Muse Spark grabbed headlines, theCoreWeave announcement quietly committed Meta to an additional $21 billion in AI cloud infrastructure from 2027 to 2032. That's on top of the existing $14.2B deal running through 2031. The message is clear: compute is the moat, and Meta is digging deeper.
Combined with the Applied AI Engineering memo — where Meta is literally pulling engineers from other divisions to focus on model improvement — it's clear the company has decided it will not be a spectator in this race.
What This Means for the Field
Muse Spark signals a shift in how Big Tech trains and deploys. It's no longer enough to release a model and call it a day. Meta is embedding its model directly into revenue-generating products (shopping, ads, assistant) and open-sourcing it to build ecosystem lock-in. The Anthropic and OpenAI playbook — proprietary API access — is being challenged by an open, integrated approach.
The interesting tension: Meta is simultaneously the most open (Llama, Muse Spark open-source) and the most closed (products that keep users inside the Meta ecosystem). Whether that hybrid model works longer-term against pure-play AI companies remains the open question.