Key Points:
- Meta Platforms plans to begin mass production of its proprietary “Iris” AI chip in September to boost computing capacity.
- The custom-designed chip, co-developed with Broadcom and fabricated by TSMC, completed internal testing in just six weeks.
- Meta targets deploying 7 gigawatts of computing power this year, aiming to double that figure to 14 gigawatts by 2027.
- The company expects to invest up to $145 billion in AI infrastructure in 2026, securing long-term supply deals with Samsung and SanDisk.
Social media conglomerate Meta Platforms is preparing to take a major step toward technological independence by transitioning its newest proprietary silicon from the laboratory to the factory floor. According to an internal company memo, the firm plans to begin manufacturing its custom-designed artificial intelligence chip, code-named Iris, in September. This development marks a pivotal milestone in a multi-generation hardware effort to eventually bypass expensive, supply-constrained third-party processors. By putting its own silicon into mass production, the company aims to dramatically lower its operational computing costs and gain a vital edge in the global artificial intelligence arms race.
The upcoming Iris chip represents the latest iteration of the Meta Training and Inference Accelerator program, which the company launched over five years ago to design custom application-specific integrated circuits. While early generations of the program struggled to deliver competitive processing speeds, the latest architecture has shown exceptional progress. Engineers completed full diagnostic and bug testing in just six weeks, finding zero major hardware anomalies. This unusually fast validation period has injected significant momentum into the hardware division, proving that the company’s long-term internal research is finally yielding reliable, production-ready silicon.
To bring the custom design to market at volume, the social media giant is leveraging elite partnerships across the global semiconductor supply chain. The firm is co-developing the hardware with U.S.-based chip designer Broadcom under an agreement that runs through 2029 and covers multiple future generations of silicon. For actual physical manufacturing, the company has secured fabrication capacity with Taiwan Semiconductor Manufacturing Company. Broadcom has confirmed that subsequent iterations of the processor line will be among the first commercial AI accelerators to utilize TSMC’s cutting-edge 2-nanometer manufacturing process, promising unmatched power efficiency.
The production timeline underscores an exceptionally aggressive hardware release strategy that challenges traditional industry schedules. The company plans to launch a new iteration of its custom processor roughly every six months through 2027. Traditionally, leading semiconductor manufacturers release upgraded chip architecture at intervals of a year or more due to the extreme complexities of cleanroom retooling. By committing to a twice-yearly release cadence, the tech giant aims to rapidly optimize its hardware for shifting software models, ensuring its server racks remain perfectly tailored to run the latest iterations of its Llama software and ad recommendation algorithms.
This rapid chip rollout is part of a massive physical buildout designed to double the company’s global data center capacity. Memos indicate that the firm plans to deploy approximately 7 gigawatts of computing infrastructure by the end of the year, with a strategic goal to double that footprint to 14 gigawatts in 2027. Running massive artificial intelligence models requires an unprecedented volume of continuous electrical power. By establishing this multi-gigawatt power pipeline, the company is securing the physical energy footprint necessary to run millions of concurrent model training sessions and serve real-time recommendations to over 3 daily active users.
However, building and powering this massive infrastructure network requires an astronomical financial commitment that has increasingly concerned Wall Street. The technology company expects its capital expenditures to reach as much as $145 billion this year, with almost the entire amount flowing directly into physical infrastructure assets. This projected spending represents nearly double the $72.2 billion recorded in 2025, demonstrating the immense scale of the company’s financial commitment. This aggressive investment is part of a broader industry-wide trend, with the world’s top four tech giants—Meta, Microsoft, Alphabet, and Amazon—poised to spend a combined $725 billion in 2026, representing a massive 77% year-on-year increase.
To protect its hardware pipeline from severe market shortages, the company has secured a series of multi-year, strategic supply agreements with global component leaders. The firm has locked in long-term supply contracts with Samsung Electronics for advanced memory chips, SanDisk for high-density flash storage, and Sumitomo Electric for specialized fiber-optic transit equipment. These agreements protect the company’s future balance sheets from sudden, erratic cost spikes on key components, securing a reliable supply of the raw physical materials necessary to support its continuous server deployments.
By forcing its engineers to transition to in-house silicon, the firm is seeking to reduce its exposure to external graphics processor suppliers. The company remains one of the largest buyers of high-end graphics chips, but every workload it can successfully transition to its own silicon dramatically reduces its long-term operational expenses. Using custom-tailored chips allows the social media giant to optimize processing environments for its specific workloads, providing superior energy efficiency and faster output times than general-purpose hardware.
Ultimately, the transition of this technology giant into a full-stack hardware developer marks a critical milestone in the balkanization of the global silicon industry. By constructing its own physical chips, building out multi-gigawatt data center grids, and training state-of-the-art open models, the firm has secured an independent, sovereign tech pipeline. The coming months will reveal how successfully the market digests these massive infrastructure investments, but the physical foundations for a custom, gigawatt-scale computing empire are now rolling off the production lines.
