Meta has released what it calls its first significant large language model since hiring Alexandr Wang as chief AI officer — a hire reportedly worth $14 billion when the Scale AI deal closed. The model, named Muse Spark, arrived alongside an announcement that Meta plans to spend between $115 and $135 billion on AI capital expenditure in 2026, roughly double what the company spent the year before.
According to CNBC’s reporting, Muse Spark is built around efficiency: smaller parameter counts than comparable models, but matched capabilities at a fraction of the compute cost. That framing has become the standard competitive positioning in AI development this year. Almost every major lab is claiming it can do more with less. Whether those claims hold under independent evaluation takes longer to sort out.
Wang’s background is in data and evaluation. He co-founded Scale AI, the company responsible for labelling and quality-checking training data for most of the major AI systems currently deployed. His appointment raised expectations that Meta would push harder on data quality and systematic model evaluation. Muse Spark is the first external signal of what that direction looks like.
The capex figures are hard to ignore. $115 to $135 billion represents a commitment at a scale that goes beyond hedging. At that level of spend, Meta is signalling that it intends to compete at the frontier of AI development regardless of near-term return on investment. That’s a meaningful strategic statement from a company that, for much of the past decade, was buying or copying frontier technology rather than building it from scratch.
How Muse Spark performs on independent benchmarks, and how quickly Meta can close the gap with OpenAI and Google, will be clearer in the weeks ahead as researchers run systematic evaluations.
Sources
- i. www.cnbc.com
Commentarii · 0