Meta Platforms Inc. plans to begin manufacturing its next-generation custom artificial intelligence (AI) chip in September, according to an internal memo reviewed by Reuters.

Code-named Iris, the in-house silicon is part of a broader effort by the social media giant to double its computing infrastructure capacity to 14 gigawatts by next year and reduce its reliance on dominant chip suppliers like NVIDIA Corp. and AMD Inc.

The new processor is part of Meta’s Training and Inference Accelerator (MTIA) project, an internal initiative designed to optimize AI workloads across Facebook and Instagram.

Developed with Broadcom Inc. and manufactured by Taiwan Semiconductor Manufacturing Co. (TSMC), Iris successfully completed its six-week testing phase with no major issues. Meta plans to aggressively cycle its custom hardware, aiming to launch a new chip iteration roughly every six months through 2027 — a pace that significantly outstrips the typical industry release cycle of a year or more.

To support its expanding suite of consumer-facing AI features, Meta expects to deploy 7 gigawatts of computing capacity this year before doubling that footprint to 14 gigawatts in 2027. For perspective, one gigawatt of energy can power roughly 800,000 homes. This massive scale rivals major industry milestones; OpenAI’s planned Stargate project, for example, is targeted at 10 gigawatts. To secure components required for this expansion amid global shortages and rising chipflation, Meta has signed long-term supply agreements with Samsung Electronics for memory, Sandisk Corp. for flash storage, and Sumitomo Electric for fiber-optic equipment.

This rapid infrastructure buildout comes at an unprecedented financial cost. Meta raised its capital expenditure guidance for 2026 to between $125 billion and $145 billion, representing a substantial portion of Big Tech’s projected $700 billion overall outlay on AI.

While the aggressive capital spending initially spooked Wall Street — driving a temporary dip in share price over near-term cash flow concerns — the company’s stock quickly rebounded. The recovery was driven by robust quarterly financial results, where Meta posted revenue of $56.31 billion (a 33% year-over-year increase marking its fastest growth since 2021) and easily beat expectations with an adjusted EPS of $7.31.

Investor sentiment was further bolstered by reports that Meta is exploring ways to monetize its massive computing footprint. Because Meta lacks a dedicated public cloud business to offset its infrastructure costs, the company is reportedly discussing plans to rent out excess data center capacity to partners or rival AI developers, a strategy similar to how SpaceX commercializes its spare capacity.

Ultimately, the deployment of custom silicon like Iris will operate behind the scenes for everyday users. By lowering the operational costs of training and running its models, Meta hopes to roll out faster image generation, more responsive virtual assistants, and improved content recommendations across Instagram, Facebook, WhatsApp, and its smart glasses, without charging users directly for the advanced tools.