1. Meta is investing billions in AI efforts, including developing hardware like the next-gen Meta Training and Inference Accelerator (MTIA).
2. The next-gen MTIA has improved performance compared to its predecessor, with a physically larger design, more processing cores, and higher clock speed.
3. Meta is facing pressure to cut costs and catch up to rivals like Google, Amazon, and Microsoft, who are ahead in the generative AI space with their own custom chips and hardware.
Meta is heavily investing in its AI efforts to catch up with competitors in the generative AI space. A significant portion of this investment is going towards recruiting AI researchers, as well as developing hardware, specifically chips to run and train Meta’s AI models. The company recently unveiled its latest chip, the next-gen Meta Training and Inference Accelerator (MTIA), which is a 5nm chip with more processing cores, power consumption, internal memory, and clock speed compared to its predecessor.
Although Meta claims the next-gen MTIA is currently live in 16 of its data center regions and providing up to 3x better performance, they are not using it for generative AI training workloads at the moment. Instead, it is intended to complement GPUs rather than replace them. Despite the progress, Meta is still moving slowly compared to rivals like Google, Amazon, and Microsoft, who have already made advancements in custom AI chip development.
Meta is under pressure to reduce costs, as they are expected to spend a significant amount on GPUs for training and running generative AI models. In-house hardware presents an attractive alternative to outsourcing these services. The company needs to catch up quickly if they hope to achieve independence from third-party GPUs and compete effectively with their rivals in the AI space.