Inside Meta's July 2026 AI Offensive: Muse Spark 1.1, the Iris Chip, and a New Cloud Business

Stock surge: Meta shares rose roughly 8–9% on July 1. CNBC reported a 9% pop, Yahoo Finance cited an 8.8% close, and other sources noted a 9.3% intraday spike—one of Meta's largest single-day gains .

Muse Image: First Image-Generation Model from Superintelligence Labs (July 7)

On July 7, Meta rolled out Muse Image, its first image-generation model from Meta Superintelligence Labs . It is integrated directly into the Meta AI chatbot and uses advanced reasoning to interpret complex prompts, blending multiple photos into shareable creations for chat, story, or feed . The launch was positioned as Meta expanding generative AI tools across its family of apps .

Muse Spark 1.1 and the First Paid Developer API (July 9)

On July 9, Meta released Muse Spark 1.1, a significant upgrade to the multimodal reasoning model first announced in April . Key details include:

• Capabilities: Designed for agentic coding, multistep reasoning, tool use, computer use, video captioning, and complex software development . Meta claims it surpasses prior products from OpenAI, Anthropic, and Google on key benchmarks .
• Benchmarks: The model is reported to achieve scores equivalent to Claude Opus 4.8 and rival GPT-5.5 on agentic evals, claiming #1 on MCP Atlas, JobBench, and Humanity's Last Exam .
• First paid API: This marks Meta's first model that it charges developers to use, via the new Meta Model API public preview . Every new API account starts with $20 in free credits .
• Context window: 1 million tokens .
• Announcement: Mark Zuckerberg returned to X (Twitter) to announce the model .
• Pricing: Meta's chief AI officer, Alexandr Wang, told CNBC the company will charge $1.25 per million input tokens and $4.25 per million output tokens—lower than the pricing for Google and OpenAI's flagship models . Multiple sources reported this aggressive pricing could spark an AI "price war" with OpenAI and Anthropic .

Iris AI Chip: September Production Plan (July 9)

On the same day, Reuters reported, citing an internal memo, that Meta plans to begin manufacturing its in-house AI chip codenamed "Iris" starting in September 2026 . Key facts:

• Part of MTIA program: Iris is the latest generation under Meta's MTIA (Meta Training and Inference Accelerator) chip family .
• Manufacturing partners: Broadcom and TSMC .
• Performance: At least one chip passed testing in about six weeks .
• Capacity goal: Meta aims to boost overall computing power to 14 gigawatts by next year .
• Purpose: To reduce dependence on and lower costs of Nvidia GPUs amid a component shortage .

Secondary stock bump: Following the Iris chip production news on July 9–10, Meta shares rose another ~5.8–7.5%, with one report noting a climb to $677.9 .

Open Questions and Controversies

Based on available sourcing, several areas of uncertainty remain:

• Cloud business execution risk: As of July 1, no launch date, pricing, or customer pipeline had been confirmed, making Meta Compute a "high-conviction strategic thesis with an unresolved execution question" rather than a near-term catalyst .
• Overspending worries: Meta's cloud push comes amid broad investor concern that tech giants are overspending on AI infrastructure without clear returns .
• Competitive pressure: Entering cloud infrastructure puts Meta in direct, capital-intensive competition with AWS, Azure, and Google Cloud—markets where it has no track record .
• Iris chip timeline: Production is planned for September, but the memo-based reporting leaves room for delays; the chip remains unproven at scale .