Meta to move its in-house Iris AI chip into production in September, targeting a jump in computing capacity

Meta is preparing to put its in-house Iris AI chip into production in September, according to a market report from Yahoo Finance that cites the chip as a central part of the company’s next phase of AI infrastructure buildout.

The same report links the chip move to a capacity goal that would effectively double Meta’s computing scale, projecting a rise to 14 gigawatts. In practical terms, gigawatts are a measure of electrical power, and AI training and inference at Meta’s scale depend on large data center and power capabilities.

Meta has long argued that AI is increasingly constrained by data center capacity, not just by model development. More chips, more memory, and more power for data centers can translate into the ability to run larger models and support more user-facing AI features, though the company has not always provided a full line-by-line breakdown of chip-by-chip throughput in public commentary.

The Iris chip is notable because it is designed internally rather than being purchased entirely from general-purpose accelerators. Custom silicon can be tailored for specific workloads such as training and inference, and it may reduce per-unit costs or improve performance efficiency compared with relying on off-the-shelf components, which matters as AI demand expands.

Meta did not provide additional detail in the materials surfaced by the Yahoo Finance report regarding Iris’s manufacturing partners, yield, or the exact mix of chip revisions that would be produced starting in September. It also did not specify whether the 14 gigawatt target reflects incremental data center power over a longer horizon or a near-term run rate tied to the Iris launch.

Industry analysts generally view the combination of custom chips and power expansion as a competitive lever. Larger computing capacity can support both the improvement of ranking and recommendation systems and the training of AI models used across products such as ads, messaging, and creator tools, but that advantage depends on execution across chip supply, data center construction, and energy access.

A related point for investors and customers is that the path from chip production to realized performance is rarely immediate. Even after chips begin production, organizations must integrate them into hardware systems, software stacks, and scheduling pipelines, which can take quarters to translate into measurable output.

What to watch next is whether Meta offers more specifics around Iris in future updates, such as capacity availability, delivery timelines to data centers, and any changes in AI compute planning. In particular, the market will be looking for confirmation that power expansion and chip onboarding are progressing at a pace consistent with the 14 gigawatt figure attributed in the report.