Meta’s $145B Bet Runs Through Taiwan

Meta is about to pour $145 billion into AI hardware and foreign chip makers, raising new questions about power, privacy, and who really controls America’s digital town square.

Story Snapshot

Meta plans to start making its in-house “Iris” AI chip in September to help double its computing muscle to 14 gigawatts next year.
The chip is part of a four-step Meta Training and Inference Accelerators project, built with Broadcom and Taiwan Semiconductor Manufacturing Company.
Testing on Iris lasted only six weeks with no major issues found, setting up rapid deployment across Meta’s data centers.
Meta expects to spend up to $145 billion on AI infrastructure in 2026, feeding worries about Big Tech power and an AI bubble.

Meta’s Iris Chip: A New Weapon in Big Tech’s AI Arms Race

Meta Platforms, owner of Facebook and Instagram, plans to begin manufacturing its new “Iris” artificial intelligence chip in September 2026. The chip is designed for Meta’s data centers and is part of a push to boost the company’s computing capacity to 14 gigawatts next year. An internal memo reviewed by Reuters laid out the plan and described Iris as one piece of a much larger hardware roadmap built inside Meta. Meta declined public comment, keeping most details behind closed doors.

The Iris chip sits inside a four-generation project called Meta Training and Inference Accelerators, or MTIA, that Meta says it will design in-house. This project aims to handle both training large AI models and running them for billions of user requests. The memo notes that the Iris chip has already gone through six weeks of testing without major bugs, which is unusually quick for such complex hardware. This speed matters because Meta’s earlier in-house chip efforts struggled for more than five years.

Broadcom, TSMC, and the New Silicon Power Structure

While Meta talks about “in-house” chips, the memo makes clear it is leaning heavily on foreign and corporate partners to get Iris built. Broadcom, a major chip firm, is helping design the silicon, and Taiwan Semiconductor Manufacturing Company is set to manufacture it. Taiwan Semiconductor Manufacturing Company produces many of the world’s most advanced chips, which means Meta’s new AI power depends on overseas factories and global supply chains. That reliance clashes with growing calls to bring more chip production back to the United States.

Conservative readers will recognize the risk here right away. Any shock to global shipping, energy prices, or Asian security could slow Iris production and choke Meta’s computing plans. Lawmakers and officials have already pushed companies like Samsung and SK Hynix to open more plants on American soil. If those pressures grow, Meta’s tight partnership with Taiwan Semiconductor Manufacturing Company could become a political target. That tension sits right under this AI hardware story, even if Meta avoids talking about it in public.

$145 Billion for AI Power: Costs, Control, and Possible Bubble

The memo says Meta could spend up to $145 billion on AI infrastructure this year, making it one of the biggest players in a roughly $700 billion Big Tech build-out. That money will go toward data centers, networking gear, and custom chips like Iris that can run large language models and image systems. Meta claims this in-house hardware will help cut costs and reduce its dependence on Nvidia and Advanced Micro Devices graphics chips. But it has not shared public benchmarks to prove those savings.

Some analysts doubt that these huge bets will pay off for investors or the wider economy. Bob Elliot of Unlimited Funds has warned that the overall AI spending boom could require revenue growth that is “largely impossible” to sustain. Others point to rising memory prices and a possible long-term shortage as signs of “chipflation” that could drive costs even higher. Put simply, Meta is spending enormous sums to gain more control over AI computing, but it may be building on shaky financial ground and fragile global supply lines.

What Meta’s AI Push Means for Everyday Americans

All this new hardware will not sit in a lab; it will run the systems behind Facebook, Instagram, and Meta’s growing AI tools. More computing power means Meta can process more data, build more invasive profiles, and target users with ever smarter algorithms. For Americans worried about Big Tech influence on speech, elections, and family life, doubling Meta’s computing capacity is not a neutral step. It is an expansion of the same machine that already shapes much of our online world.

From a conservative view, the core question is simple. Who decides how this new AI power is used, and who can hold Meta accountable if it crosses the line? The Trump administration is pushing for stronger American industry and tougher oversight of globalist systems, but Meta’s Iris plan shows Big Tech still moves fast and mostly in secret. Without clear rules on data use, content control, and free speech, a 14-gigawatt AI engine in Menlo Park may feel less like progress and more like another threat to the rights and values of everyday citizens.