A Company That Lost in Court and Laid Off Staff Just Bet the House on AI
Meta has lost two major court cases and scaled back its Metaverse ambitions. Meta is also about to spend somewhere between $115 billion and $135 billion on AI in 2026. Both of those things are true at the same time, and that tension tells you almost everything you need to know about where this company’s head is at right now.
On April 8, 2026, Meta announced Muse Spark — its first new large language model since the formation of its Superintelligence Labs division, and its first real return to the LLM space after going quiet for roughly a year. If you blinked, you missed the hiatus. If you didn’t blink, you watched a company quietly reorganize its AI team and come back with a new name on the box.
What We Actually Know About Muse Spark
Not a lot, honestly. What Meta has confirmed is that Muse Spark exists, that it dropped on April 8, and that it comes out of the Superintelligence Labs structure Meta built after reshuffling its AI team. The model appears to be focused on Meta’s own ecosystem — which makes sense given the platforms it has to feed: Facebook, Instagram, WhatsApp, Threads, and whatever the Metaverse is calling itself this week.
What we don’t have yet is a thorough breakdown of benchmarks, context windows, or how it stacks up against what Google, OpenAI, and Anthropic have been shipping while Meta was on its break. That gap matters. A year in LLM development is not like a year in most industries. Models that were leading the pack in early 2025 look dated now. Meta is walking back into a space that moved fast without it.
The Money Is the Real Story
Here’s what actually demands attention: $115 to $135 billion in AI capital expenditure for a single year. That number, pulled directly from Meta’s latest earnings report, is not a rounding error. For context, that’s not a research budget — that’s infrastructure, compute, talent, and the kind of long-term bets that take years to pay off, if they pay off at all.
Meta is not tiptoeing back into AI. It is throwing its full weight behind a space it stepped away from, at a moment when it is also dealing with legal losses and a quieter Metaverse story. The pivot is sharp and it is expensive. Whether the spend is disciplined or panicked is a fair question to ask.
Why the Hiatus Happened and Why It Matters Now
Meta’s AI team went through real turbulence. The Superintelligence Labs formation was not just a rebrand — it reflected internal changes in how Meta wanted to structure its AI work. A year of relative silence from a company that had been one of the more open players in the LLM space, thanks to the Llama model family, was noticeable.
Muse Spark appears to be a different kind of product than Llama. Where Llama was open-weight and developer-facing, Muse Spark reads as more proprietary and more tightly tied to Meta’s own platforms. That is a strategic shift worth watching. Meta built a lot of goodwill in the developer community with Llama. A move toward closed, platform-specific models trades some of that goodwill for tighter control.
The Honest Take
Meta coming back to LLMs is not surprising. A company sitting on billions in compute infrastructure and hundreds of millions of daily active users was never going to stay out of the model race for long. What is worth watching is whether Muse Spark is a solid product or a flag-planting exercise designed to signal to investors that the AI spend has a face on it.
The $125 billion midpoint of that capex range is a number that needs results behind it. Meta’s platforms give it a distribution advantage that most AI labs would trade a lot to have. If Muse Spark is genuinely well-built and gets woven into the products people already use every day, that reach is a real asset.
But we have seen big AI announcements land with a thud before. The model is out. The benchmarks and real-world performance will tell the actual story. We will be watching, and we will not be polite about what we find.
đź•’ Published: