Meta to expand custom AI chips, balancing Nvidia, AMD deals

Meta expects to expand its use of custom chips to train AI models, Susan Li, Meta’s chief financial officer, said at a Morgan Stanley technology conference.

Li said Meta has already rolled out custom silicon at scale for ranking and recommendations workloads.

Meta is a major operator of data centers for AI and recently made large deals with Nvidia Corp. and Advanced Micro Devices Inc. for chips and equipment.

Li said the company is buying different chip types for specific tasks and that “custom silicon is a big part of that.”

Her remarks suggest Meta is balancing large supplier agreements with continued in-house processor development.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Meta’s AI chip strategy looks like a reset

Meta reportedly stopped work on its second-generation in-house AI accelerator, codenamed Olympus, after technical and manufacturing problems 1.
The change could push Meta toward outside suppliers for its most advanced AI needs. Meta has not confirmed the reported move 2.
Meta has reportedly signed multi-billion dollar deals for large volumes of AI chips from Advanced Micro Devices Inc. (AMD). It is also expected to deploy Google tensor processing units (TPUs), Google-designed AI accelerator chips, as part of a multi-vendor AI infrastructure approach 2.
At a Morgan Stanley technology conference, meta chief financial officer Susan Li said the company still builds custom chips for content ranking and recommendations. Meta already runs those chips at scale.
Li said Meta buys different chip types for specific jobs. She added that “custom silicon is a big part of that,” which points to continued in-house processor work alongside big supplier purchases.

The cost of AI strengthens chipmakers

Pulling back from Olympus could keep Meta dependent on Nvidia and AMD, which bolsters their leverage with a huge customer 2.
Building competitive accelerators takes sustained spending and long lead times. Established chip firms also benefit from mature software ecosystems and manufacturing partnerships 2.
Meta still relies on vendors, and the price tag is high. Meta guided for 2026 capital expenditures of $115 billion to $135 billion 1.
That spending ties Meta’s finances to how well its AI products perform. It also adds exposure to chip pricing and supply bottlenecks 1.