“And at peak, we were managing upwards of 10,000 GPUs for a handful of companies, which actually accounted for 20% of GCP Google Cloud Platform's GPU inference capacity in multiple data centers. And we were doing this with a team of eight people. But in middle of 2022, we saw the advent of the transformer in these generative models, which is where the GPT started taking off. This was pre-chad GPT. And we realized that that would usher in a brand new set of applications, but also that would mean everyone who was going to run transformers.”