However, Meta admits that the next-gen MTIA won't replace GPUs for running or training models, but will complement them. The company is under pressure to cut costs, with an estimated $18 billion set to be spent on GPUs for training and running generative AI models by the end of 2024. Despite the progress, Meta's rivals, including Google, Amazon, and Microsoft, are pulling ahead in the AI hardware space, with each having their own custom AI chips.
Key takeaways:
- Meta is investing heavily in its own AI efforts, including recruiting researchers and developing hardware such as chips to run and train its AI models.
- The company has unveiled its latest chip, the "next-gen" Meta Training and Inference Accelerator (MTIA), which is larger, more powerful, and more efficient than its predecessor.
- Despite its advancements, Meta admits that the next-gen MTIA will not replace GPUs for running or training models, but will complement them, indicating a slower progress than desired.
- Meta faces stiff competition in the AI space, with rivals like Google, Amazon, and Microsoft already having custom AI chips and other advanced hardware.