I'm LongbridgeAI, I can summarize articles.Meta Platforms is accelerating its custom AI silicon strategy to reduce costs and energy consumption associated with its AI infrastructure. The company plans to deploy third-generation MTIA chips in data centers by early 2027, focusing on inference tasks rather than training. Developed with Broadcom and TSMC, these chips aim to improve performance per watt and dollar. Early tests show promising results, with chips performing within 2-3% of simulations. This move seeks to lower Meta's AI cost curve and protect margins as AI demand grows.
Meta Platforms is moving deeper into custom AI silicon, with a new generation of internally designed chips set to enter its data centers in the first half of 2027. For investors, the move is about more than reducing reliance on outside hardware: Meta is trying to lower the enormous energy and computing costs attached to its AI ambitions while building more control over one of its most important infrastructure layers.
The company is currently testing its third-generation MTIA 450 processor, code-named Arke, while its successor, MTIA 500, or Astrid, is expected to complete design work in about a month and reach data centers by the end of 2027.
Each one takes on a little bit more risk technologically and gets us better performance, Meta engineering vice president Yee Jiun Song said, pointing specifically to better performance per watt and per dollar.
Meta is working with Broadcom (AVGO) on chip design and Taiwan Semiconductor Manufacturing Co. (TSM) on production.
The scale of the effort is already significant. Meta has committed to deploying more than a gigawatt of the chips over a 12-month period, with deployment expected to accelerate if AI demand remains strong.
Early results also appear encouraging. Twelve Arke chips delivered by TSMC on Sept. 1 performed within 2% to 3% of Meta's simulations and have already run Meta models alongside models from DeepSeek and Alibaba.
Meta has also narrowed the project's focus.
The company canceled its planned Olympus processor, which was intended to handle both AI training and inference, partly because of cost concerns. Instead, Meta is prioritizing inference, the day-to-day running of AI models.
These are the workhorse chips that we're going to use for general-purpose inference, Song said.
What Meta investors should watch next
The core investor question is whether custom silicon can materially reduce Meta's AI cost curve.
If MTIA deployment lowers power consumption and inference expense at scale, Meta could protect margins while continuing to expand AI across advertising, recommendations, assistants and other products.
Investors should watch deployment speed, performance gains per dollar, the pace of inference demand and whether Meta expands custom silicon beyond its current roadmap.
The upside is straightforward: even modest efficiency gains could become financially meaningful at Meta's scale. The risk is execution. Designing competitive chips requires enormous capital, long development cycles and tight coordination with partners such as Broadcom and TSMC.
For now, Meta is making clear that custom AI hardware is becoming a central part of its long-term cost strategy.