Meta is on track to begin production of its latest AI chips in September as part of an effort to reduce GPU costs amid ongoing component shortages, Reuters reported, citing an internal memo. The chips, developed under Meta’s Meta Training and Inference Accelerator program with Broadcom and manufactured by TSMC, are intended to support training for Meta’s ranking algorithms, broader AI workloads and inference. Meta expects capital expenditures between $125 billion and $145 billion this year, much of it tied to AI infrastructure, and plans to deploy 7 gigawatts of compute this year, doubling that figure next year.
Separately, Meta’s new Muse Image generator has drawn scrutiny over a feature allowing users to create AI-generated content using photos from public Instagram accounts without notifying the account owner. Only private accounts and users under 18 are automatically excluded, raising concerns among privacy advocates about consent and potential misuse. Meta has provided an opt-out setting through Instagram’s sharing controls.
Meta also launched Muse Spark 1.1, a multimodal AI model built for agentic coding and workflow automation, positioning it as a competitor to offerings from OpenAI and Anthropic. Meta said the model performs strongly in agentic tasks and tool use, pricing it slightly above Anthropic’s Claude Haiku 4.5 and OpenAI’s GPT-5.6 Luna. CEO Mark Zuckerberg posted about the release on X, his first post on the platform in three years, describing Spark as a strong, low-cost agentic model with additional releases expected soon. The announcements arrive during a highly competitive week for AI model releases across the industry.
