AI Superintelligence Pivot & Capex
Key Questions
What is Muse Spark 1.1 and how does it perform on benchmarks?
Muse Spark 1.1 is Meta's coding model that scores 71 on independent benchmarks at $0.26 per task, undercutting GPT-5.4's $0.89 cost while reducing hallucinations through abstention. It still trails GPT-5.5 by 9 points in agentic coding tasks.
Why did Meta scrap the Muse Image and Video tools?
Meta launched Muse Image/Video with automatic opt-in cloud service but quickly withdrew it after privacy backlash from SAG-AFTRA, CAA, and users over unauthorized use of public photos. The core Advantage+ ad integration survived the rollback.
What caused the shutdown of Instagram's AI @-mention tool?
The Instagram AI @-mention tool was shut down shortly after launch due to user backlash over privacy and consent concerns, similar to the Muse Image feature. This highlights ongoing product execution risks at Meta.
What is Meta's planned capex for AI in 2026?
Meta has outlined $125-145B in capex for 2026, including a potential $6.5B deal with Samsung for 2nm AI chips and expanded Nvidia partnerships. Revised cost estimates put AI build costs at $22B per GW, lower than Wall Street projections.
How have AI enhancements affected Meta's advertising results?
AI ad enhancements have delivered 17% higher ROAS and 32% lower CPA, completing full ad automation through Advantage+ integration. Ad revenue is projected at $240-243B as a result.
What is the status of Meta's AI agent development?
Zuck has admitted AI agent development stalled for four months, with the Llama 4 pipeline failing and a pivot to product integration. The Watermelon model is now catching up to GPT-5.5.
What subscription products has Meta launched for AI?
Meta introduced a $20/month AI glasses subscription and Fury AI glasses priced at $299, alongside reaching 1B AI users. These efforts support broader AI monetization alongside ad automation.
What analyst upgrades followed Meta's AI announcements?
BofA upgraded Meta citing frontier AI capability within nine months and a revised lower cost narrative, sending shares to a record $670. The stock reflects optimism around pricing advantages and ad revenue growth.
Meta all-in on AI via MuseSpark, Avocado, Scale AI. Spark 1.1 coding model scores 71 on independent benchmark at $0.26/task (vs GPT-5.4 at $0.89), with hallucination reduction via abstention; still trails GPT-5.5 by 9 points in agentic coding. Bosworth confirms Llama 4 pipeline failure, pivot to product integration. Muse Image/Video launched with cloud service but quickly scrapped after privacy backlash — SAG-AFTRA and CAA pushback, automatic opt-in design, creator trust erosion. However, the core Advantage+ ad integration survived, completing the ad automation loop (URL + budget = campaign). Instagram AI @-mention tool also shut down after backlash, reinforcing product execution risk. MCI halted. Zuck admitted AI agent development stalled 4 months. Watermelon model catching GPT-5.5. AI ad enhancements: 17% higher ROAS, 32% lower CPA. Meta launched $20/month AI glasses subscription, Fury AI glasses at $299. Meta expanded Nvidia partnership, reached 1B AI users. Capex $125-145B for 2026. Meta negotiating $6.5B Samsung 2nm AI chip deal. WhatsApp AI shift signals agentic economy. Full ad automation now complete via Advantage+ integration; ad revenue projections $240-243B. BofA upgrade cites frontier AI capability within 9 months. Major cost revision: AI build cost estimated at $22B/GW vs Wall Street's $45B, flipping narrative. Stock hit record $670.