The five segments of NVIDIA’s data center business
NVIDIA (NVDA) is eyeing AI (artificial intelligence) to boost its future growth. The company is at the forefront of the AI technology and is exploring new applications wherein AI could make a significant difference.
Over the past three years, NVIDIA has expanded its data center business from just HPC (high-performance computing) to five different sub-segments, which we’ll examine below.
NVIDIA’s Tesla GPUs (graphics processing units) are being used by 15% of the world’s top 500 supercomputers. However, this market is no longer limited to supercomputers and has now expanded to enterprises. HPC is one of the fastest-growing segments in the IT (information technology) industry, as more companies are using HPC to process data and to get insight.
As we shift to AI, more and more data centers are using NVIDIA GPUs for DL (deep learning). In DL, a computer goes through huge chunks of data in the form of images, texts, and sounds to perform image or voice recognition. In DL, computing speeds can make a difference as huge as building a $20-million or a $200-million data center.
After a DNN (deep neural network) is created, it has to be put on hyperscale datacenters to answer the billions of queries consumers post on the Internet every day. Until now, NVIDIA has had no presence in this space. This means that all its inferencing happened on CPUs (central processing units)—mostly Intel’s CPUs.
But in fiscal 3Q18, NVIDIA launched its TensorRT 3 inference acceleration platform, which uses Tensor Core GPU instruction set architecture to accelerate networks by a factor of 100. The company ensures that the platform supports all architectures and all sizes of networks to enable customers to scale out their hyperscale datacenters to support more data traffic in a cost-effective manner.
GPU cloud service
It usually costs a company huge sums to get all three capabilities of HPC, training, and inferencing. NVIDIA brings these three capabilities to the public cloud, where a consumer can rent the GPU on an hourly basis to perform her AI tasks. Cloud companies such as Google (GOOG) and Amazon.com (AMZN) use NVIDIA’s GPUs to provide GPU-as-a-service.
NVIDIA supports the cloud service through a cloud registry called NGC (NVIDIA GPU Cloud), which puts software stacks in containers and optimizes it for all cloud services and frameworks.
NVIDIA is expanding its GPU computing into different verticals, such as healthcare, construction, and automotive. For instance, NVIDIA is working with GE Healthcare (GE) to make its image-sensing devices smarter.
NVIDIA’s five sub-segments show that it’s now expanding the reach of its AI technology, and many analysts believe that AI is a far bigger opportunity than what NVIDIA claims. Next, we’ll discuss how NVIDIA is measuring its data center TAM (total addressable market).