Meta’s next generation AI silicon is a 5nm chip designed to power the models that provide recommendations to those who use its social network platforms. The new MTIA inference accelerator is part of a “broader full-stack development program for custom, domain-specific silicon that addresses our unique workloads and systems,” Meta says. The next-gen MTIA more than doubles the compute and memory bandwidth of its predecessor, the 7nm MTIA v1 chip introduced in May 2023, resulting in 3x the performance, according to Meta, which says the new silicon is already live in 16 data centers.

While each chip “consumes more power — 90W versus 25W — it also boasts more internal memory (128MB versus 64MB) and runs at a higher average clock speed (1.35GHz up from 800MHz),” TechCrunch summarizes.

However, TechCrunch is critical of Meta’s timeline, alleging “Meta is moving slowly,” chiding that “while Meta’s hardware drags, rivals are pulling ahead.”

Examples of purpose built AI chips from Big Tech firms referenced by TechCrunch include the Google Cloud TPU v5p training chip, which went into general release this week, and the newly unveiled Axion, Google’s “first dedicated chip for running models.” Also, Amazon’s Trainium and Inferentia series, and Microsoft’s Azure Maia AI and Azure Cobalt 100 CPU accelerators.

Meta explains in a blog post it took less than nine months to “go from first silicon to production models” of the next-gen MTIA.

This “is shorter than the typical window between Google TPUs. But Meta has a lot of catching up to do,” writes TechCrunch, noting that Meta’s next-gen MTIA reveal comes one day after the company “confirmed that it plans an initial release of Llama 3 — the next generation of its large language model used to power generative AI assistants — within the next month.”

While MTIA stands for Meta Training Inference Accelerator, TechCrunch takes issue with the fact that Meta is not immediately using it for model training, but in data centers to rank and recommend content and serve ads to users on Meta platforms including Instagram and Facebook.

Bloomberg points out that Meta said it would “spend as much as $35 billion on infrastructure to support AI, including data centers and hardware” in 2024.

While it’s blog post describes an ambitious hardware and software build-out leveraging the new MTIA silicon, Bloomberg writes that “a significant amount of that spending will likely still flow to Nvidia, which builds the popular H100 graphics cards” costing tens of thousands of dollars each. Earlier this year, Meta CEO Mark Zuckerberg said the company planned to acquire 350,000 H100 GPU processors to drive its AI efforts.