Meta Plans to Double Computing Power by 2027
Meta plans to double its computing power by 2027 as part of its expansion of AI infrastructure.
This move closely aligns with the production of Iris AI chips and the MTIA project, aimed at supporting AI feature upgrades on platforms like Facebook and Instagram while reducing reliance on external GPUs.
Meta's capital expenditure is expected to reach $145 billion this year, deploying 7 GW of computing infrastructure by 2026 and increasing to 14 GW by 2027, highlighting its long-term commitment to AI computing resources.
Source: Public Information
ABAB AI Insight
Meta has been continuously iterating its self-developed chips since the launch of the MTIA project. The announcement to double computing power by 2027 continues the strategy of hardware vertical integration to control AI infrastructure, similar to previous large-scale data center investments.
In terms of capital strategy, Meta is mobilizing significant capital expenditure for a dual-track layout of chips and data centers, motivated by the goal of reducing long-term costs through proprietary computing resources and enhancing control over AI model training and inference. The specific aim is to support deep integration of AI features on social platforms and create competitive barriers.
Similar to the data center expansions of giants like Google and Microsoft, Meta is currently in an accelerated phase of the AI computing arms race, with the doubling plan for 2026-2027 marking a shift from following to proactive positioning.
Essentially, this represents capital concentration: the massive investments by tech giants in AI computing infrastructure are accelerating the concentration of industry resources towards a few platforms with user scale and self-development capabilities, as computing power becomes the core production factor in the AI era, shifting pricing power from hardware suppliers to end platforms.
ABAB News · Cognitive Law
- The speed of doubling computing power determines the survival radius of enterprises in the AI era.
- Self-developed chips combined with massive capital expenditure is the true structural approach to mastering computing pricing power.
- User scale must ultimately translate into control over computing power; otherwise, one can only forever be a dressmaker for others.