Zuckerberg's New AI Model Costs 75% Less Than Rivals — Here's His Pit…

Meta launched Muse Spark 1.1, a proprietary AI model priced 75% below rivals, backed by custom Iris chips co-developed with TSMC to reduce inference costs. Zuckerberg's $145 billion capex bet now hinges on this pricing edge offsetting slower-than-expected AI revenue growth and delays in the frontier Watermelon model. If you sell inference—whether embeddings, RAG, or LLM completions—Meta just undercut your entire price model by three-quarters and locked in custom silicon to hold that margin. You now compete on speed, accuracy, or a vertical they don't serve; anything else becomes a cost centre.