Meta Platforms plans to manufacture its own data-center AI chip, code-named Iris, starting in September, according to a Reuters report based on an internal memo. Designed with assistance from Broadcom and built by Taiwan Semiconductor Manufacturing (TSMC), the chip aims to increase computing power and reduce reliance on external hardware providers.
Meta’s Iris Chip and the Shift Toward Custom Silicon
The Iris chip is part of a four-generation family of in-house silicon. According to the internal memo reviewed by Reuters, the program is moving faster than typical industry cycles. Testing for Iris took approximately six weeks and revealed no major issues.
Meta intends to launch a new chip roughly every six months through 2027. This cadence is significantly more aggressive than the standard industry norm of one new chip per year. While Meta continues to purchase graphics processing units (GPUs) from Nvidia and Advanced Micro Devices (AMD), Iris is designed to augment these systems rather than replace them entirely.
Financial Stakes: Balancing AI Spend with Revenue Growth
Meta’s pivot to custom silicon comes as the company aggressively scales its AI infrastructure. For the 2026 fiscal year, Meta raised its capital expenditure forecast to a range of $125 billion to $145 billion, up from the previous $115 billion to $135 billion range. In the first quarter alone, capital expenditures reached $19.8 billion.
This spending is funded by strong performance in Meta’s core business. First-quarter revenue rose 33% year over year to $56.3 billion. This represents an acceleration from 24% growth in the fourth quarter of 2025. The company also reported a 41% operating margin for the period.
CEO Mark Zuckerberg described the period as a “milestone quarter” in the company’s first-quarter earnings release, citing momentum across apps and the debut of the first model from Meta Superintelligence Labs.
Market Impact on Nvidia, Broadcom, and TSMC
The development of Iris creates a divergent impact on the semiconductor supply chain. While Nvidia’s chips remain the backbone of Meta’s current operations, the move toward in-house silicon introduces potential pricing pressure on the GPU giant over time.
Conversely, the program increases the strategic importance of Broadcom and TSMC. Broadcom provides the design assistance, while TSMC handles the actual fabrication. This shift allows Meta to potentially extract more computing power from every dollar spent on infrastructure.
Valuation and Future Outlook
Despite the massive infrastructure spend, Meta’s valuation remains tied to its ability to convert compute into engagement. Trading at approximately 24 times earnings and 19 times forward earnings, the stock reflects a balance between its high growth rate and the risks associated with its AI budget.
The success of the Iris program depends on whether these custom chips deliver the expected savings and efficiency. If Meta can successfully lower the cost of its build-out, the company’s heavy investments may translate into earnings growth faster than the market currently anticipates.
Comparing Meta’s AI Infrastructure Strategy
| Metric | Previous/Industry Standard | Meta’s New Target |
|---|---|---|
| Chip Release Cycle | ~1 year | ~6 months (through 2027) |
| 2026 CapEx Forecast | $115B – $135B | $125B – $145B |
| Compute Capacity | Current baseline | 14 gigawatts by 2027 |
Frequently Asked Questions
Iris is a custom-designed AI chip for data centers being developed by Meta Platforms with Broadcom and manufactured by TSMC. It is intended to augment Nvidia and AMD GPUs.
According to a Reuters report, Meta plans to begin manufacturing the Iris chip in September.
Meta has raised its 2026 capital expenditure forecast to between $125 billion and $145 billion.
What do you think about Meta’s move toward custom silicon? Will it successfully reduce their dependence on Nvidia? Let us know in the comments below or subscribe to our newsletter for more deep dives into AI hardware trends.
Worth a look