Meta is tired of waiting in line for expensive graphics cards. So, it is taking its artificial intelligence supply chain into its own hands. According to an internal corporate memo reviewed by Reuters, Zuckerberg’s social media empire is officially on track to start manufacturing its own custom-built data center AI processor, code-named “Iris,” this September. The custom silicon is designed specifically to optimize the massive content ranking, recommendation systems, and generative algorithms running across Facebook and Instagram.
The timing could not be more critical. Meta is locked in an aggressive multi-billion-dollar infrastructure race against rivals like OpenAI, Google, and Anthropic. Relying entirely on external hardware suppliers has become a massive bottleneck. It has costed the company both precious deployment time and billions of dollars in capital.
Sailing through testing and securing parts
The leaked documentation reveals that the custom Iris chip sailed through its initial bug-testing phase in just six weeks without turning up any major architectural problems. To bring the custom silicon to life, Meta is leaning on Broadcom for the physical chip design. To handle the actual high-volume fabrication, the company is partnering with Taiwan Semiconductor Manufacturing Company (TSMC).
But building a data center chip requires more than just silicon. This is especially true in an era where global hardware shortages and “chipflation” are driving up production costs across the entire tech sector. To safeguard its massive expansion, Meta has quietly locked down multi-year, long-term supply contracts with global hardware giants. The company secured high-performance memory chips from Samsung Electronics, flash storage from Sandisk, and advanced fiber-optic equipment from Sumitomo Electric.
A ridiculous amount of raw power
The launch of the Iris processor fits into a mind-boggling infrastructure expansion plan. Meta intends to deploy roughly seven gigawatts of computing capacity by the end of this year. It also has plans to double that number to a staggering 14 gigawatts in 2027. To put that in perspective, one gigawatt can power close to 800,000 homes.
Financing a global computing footprint of this size requires deep pockets. Meta projects its total infrastructure spending will hit up to $145 billion this year alone. The figure accounts for a massive chunk of Big Tech’s collective $700 billion artificial intelligence war chest. The custom Iris processors are meant to augment—not completely replace—the massive volumes of GPUs Meta still buys from Nvidia and AMD. However, shipping its own silicon allows the tech giant to deploy hardware on its own schedule.
The Android Headlines Take
You simply cannot become a dominant AI titan if you are entirely dependent on a competitor for your core infrastructure. This also applies to the “brain” that will run on that hardware, as demonstrated by the Gemini usage limit cap that Google recently placed on Meta. These kinds of sudden, third-party-dependent developments can massively disrupt business operations.
Meta’s modular, rapid-fire approach to custom silicon—planning to drop a new iteration roughly every six months through 2027—is really aggressive compared to traditional industry timelines. Bypassing the traditional hardware supply bottleneck and securing its own components from Samsung and Sandisk will help Meta build an independent ecosystem that can scale its software without constantly writing massive checks to external vendors.
The post Meta Is Massively Ramping Up Its In-House AI Silicon Production This September to Fight ‘Chipflation’ appeared first on Android Headlines.
