Meta (NASDAQ:META | META Price Prediction) is charging businesses to use one of its AI models for the first time, undercutting rivals significantly. On July 9, 2026, Meta launched Muse Spark 1.1, a frontier model’s paid developer tier costs roughly 25% of what OpenAI and Anthropic charge, meaning developers pay about 75% less. Mark Zuckerberg’s pitch: make AI cheap enough that everyone builds on Meta’s platform.
This marks a genuine pivot. Meta championed open-source AI; Muse Spark 1.1 is proprietary and revenue-focused. As Zuckerberg told Bloomberg on July 9, “Since this isn’t an open-source model, this is really the first time we’re seriously launching an API business.”
The Pitch in Zuckerberg’s Words
Where rivals charge $5 to $10 per million input tokens and $30 to $50 per million output tokens, Meta costs roughly a quarter of that. Vals.ai found Muse Spark 1.1 runs at about one-tenth the cost of GPT-5.5, while AnalysisAI measured input costs roughly 75% below Anthropic’s Claude Opus 4.8 and output costs about 83% lower. “The pricing is going to be very aggressive and attractive,” Zuckerberg framed it as a mission: “Someone has to build these models and make sure the highest quality intelligence is available to everyone.”
Why Meta Had to Do This
Meta committed $125 to $145 billion in 2026 capex, its largest ever, with shares gaining 14.81% in the week ending July 10. In April, Zuckerberg said: “We had a milestone quarter with strong momentum across our apps and the release of our first model from Meta Superintelligence Labs.” Meta is co-developing a custom “Iris” AI chip with Broadcom, manufactured by TSMC, cutting Nvidia dependence and lowering inference costs. A leaked memo revealed plans to put Iris into production in September and double computing capacity to 14 gigawatts. Our AI infrastructure research covers second-order beneficiaries in this report on power and data-center names beyond chipmakers.
Ticker Exposure to Meta’s AI Buildout
Broadcom (NASDAQ:AVGO) is tied to Iris and posted Q2 FY2026 AI semiconductor revenue of $10.8 billion, up 143% year over year. CEO Hock Tan said: “The momentum continues and in Q3 we expect semiconductor revenue from AI to grow over 200 percent year-over-year to $16.0 billion.” Taiwan Semiconductor (NYSE:TSM) fabricates both Iris and Nvidia GPUs. Its May 2026 revenue rose 30.1% year over year. SemiAnalysis projects Meta’s total AI compute will surpass OpenAI’s and Anthropic’s by year end.
The Open Question
On July 2, he admitted AI “hasn’t really accelerated in the way we expected” internally, creating tension with this week’s bullish launch. Muse Spark 1.1 competes on price and capability while still trailing GPT-5.5 on outright performance. Meta’s true frontier model, “Watermelon,” is still in development. If it delivers, the bet gets interesting. If not, Meta may have started a price war it cannot win. Can 75%-cheaper AI earn back a $145 billion infrastructure bill?
Contact [email protected] for any questions or corrections.
Top Gaining Stocks
Top Losing Stocks