Meta Dives Headfirst into AI Infrastructure with $145 Billion Bet

Meta Platforms is embarking on an unprecedented $145 billion infrastructure bet in 2026, primarily focused on artificial intelligence (AI). This staggering commitment exceeds even Big Tech standards and is a significant step towards driving the next trillion-dollar phase by 2027.

The company plans to deploy 7 gigawatts of AI computing capacity this year, which will double to 14 GW by the end of 2027. At the heart of Meta's infrastructure push is its custom-built Iris chip, a fourth-generation MTIA (Meta Training and Inference Accelerator) that will be manufactured in September 2026.

Meta's move into silicon design marks its entry into a growing club of hyperscalers designing their own chips. This includes Google's TPUs and Amazon's Trainium, both of which are currently industry leaders in AI acceleration. Meta's Iris chip is expected to support the company's future compute rental business, with estimates suggesting it could generate up to $22 billion annually by 2027.

However, the road ahead is not without risks. The broader industry is pouring money into AI at an unprecedented pace, which could lead to overcapacity and put a strain on Meta's financial commitments. With total AI-related spending projected to surpass $1 trillion, Meta will need to navigate this complex landscape carefully.

Despite these challenges, Meta's massive existing compute footprint and ability to price competitively give it an advantage in the market. The company has already signaled its openness to external partnerships and is building developer-facing infrastructure that could support a cloud-adjacent revenue opportunity.