In the world of cloud datacenters, the competition is fierce. Providers are constantly striving to outperform each other and cut costs. And one strategy they’re increasingly turning to is developing their own chips. That’s right, these companies are creating their very own silicon to power their datacenters.
The big cloud providers have already made significant investments in developing in-house chips. From general computing to networking, storage, and AI, they’re designing custom silicon for a wide range of purposes. Just take a look at AWS. Their Graviton CPUs are estimated to power one in five cloud instances on EC2. And they’re not the only ones. Google has its Tensor Processing Unit (TPU), while Chinese companies like Alibaba, Baidu, and Tencent are also working on their own custom silicon.
But it’s not just about the big players. Even smaller cloud providers are getting in on the action. They see developing their own chips as a way to diversify their supply chains and gain a competitive edge. And it’s not just the big, flashy chips that are being developed. Many of these custom chips are designed for internal use and go unnoticed by end users. They’re focused on offloading processes like storage, networking, and security from the CPU, allowing for more efficient use of resources.
When it comes to AI and machine learning, the story is much the same. Google and AWS have been building custom AI accelerators for years. But again, these chips are primarily optimized for internal workloads. However, customers can still take advantage of them, running jobs on these specialized chips.
It’s important to note that custom silicon is not meant to replace commercial chips entirely. Chips from companies like Intel, AMD, and Nvidia still have their place. Nvidia, in particular, has gained momentum with its GPUs and has become the go-to choice for large language models. Cloud providers are not only developing their own chips, but also buying massive quantities of Nvidia’s GPUs.
Looking ahead, we can expect to see more custom accelerator deployments. Factors like power and space constraints, AI demand, and geopolitical issues will continue to drive the development of cloud silicon. In the US, AI accelerators will likely be the focus, especially among the largest cloud providers. Smaller players may stick to commercial silicon from major chipmakers.
In China, the situation is different due to geopolitical pressure. Many Chinese cloud providers are developing their own custom silicon to avoid dependence on US-developed chips. US regulations could further limit their ability to do business, so they’re taking matters into their own hands.
So, while the world of cloud silicon may be largely invisible to end users, it’s a crucial part of the infrastructure that powers our online world. These custom chips are the secret ingredients that allow cloud providers to deliver high-performance, efficient, and cost-effective cloud services.