Meta announced plans to launch custom AI chips in September and will scale computing capacity to 14 gigawatts by 2027, aiming to reduce dependence on external chipmakers and lower infrastructure costs.
The company also released an upgraded AI model, Muse Spark 1.1, designed for agent-based systems and coding work.
These developments signaled to investors that Meta remains competitive in the broader AI race.
What happened
Meta will launch custom-designed AI chips in September as part of a plan to increase computing capacity to 14 gigawatts in 2027, up from a projected 7 gigawatts in 2026. The company is partnering with Broadcom and Taiwan Semiconductor Manufacturing. Meta also released Muse Spark 1.1, an upgraded AI model designed for agent-based systems and coding tasks that is faster than previous versions and better at diagnosing and fixing bugs.
Why it matters
By making its own chips, Meta aims to reduce reliance on suppliers like Nvidia and Advanced Micro Devices, lower infrastructure costs, and ease supply constraints. The new AI model positions Meta as competitive in the race to develop capable reasoning and coding systems. Together, these moves reassured investors that Meta remains at the forefront of AI development.
What to watch
Meta plans to develop upgraded versions of its chips on a six-month cadence. For context, a single gigawatt can power roughly 750,000 homes.
Ask the AI about this article →
Meta's announcement reflects a broader strategy among major cloud providers to build vertical integration in AI infrastructure. Rather than relying solely on established chipmakers facing supply constraints and pricing pressure, Meta is partnering with manufacturers like Broadcom and Taiwan Semiconductor Manufacturing to design chips optimized for its own workloads. This approach mirrors efforts by other hyperscalers (large cloud providers) to gain more control over their computing economics and reduce dependency on external vendors.
The aggressive scaling plan—doubling computing capacity from 7 gigawatts in 2026 to 14 gigawatts in 2027—signals Meta's commitment to supporting both its existing services and advanced AI model development. The six-month upgrade cadence for chip versions is notably rapid, reflecting the pace at which AI capabilities are advancing. Simultaneously, the release of Muse Spark 1.1 demonstrates that Meta is making progress on AI software that can coordinate multiple reasoning agents and handle complex coding tasks, capabilities that are valuable both for internal infrastructure and for potential external licensing or partnership opportunities.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
