Why high-performance computing has moved to the cloud | Computing
Two leading analysts’ reports, from Intersect360 Research and Hyperion Research, show the high-performance computing market has reached an inflection point. The cloud segment includes Microsoft, Amazon Web Services, and Google.
Intersect360 says high-performance cloud spending by high-performance computing customers grew by 44 percent from 2016 to 2017, to about $1.1 billion—much faster than the growth in the total high-performance computing market, which is still mostly traditional on-premises hardware clusters.
The two related reasons for the faster cloud adoption of high-performance computing are pretty clear to me.
First, most on-premises high-performance computing hardware has aged poorly, with clearly inferior capabilities compared to the newer high-performance computing hardware in the public clouds. It’s much cheaper to use a public cloud’s high-performance computing service rather than replacing the high-performance computing servers in your datacenter, so the cloud providers are the ones buying lots of this gear to keep up with the growing cloud demand.
Also, IT shops—even those who have resisted the cloud—face budget limitations, so they just can’t buy that much high-performance gear for use on-premises.
Second, the platform features that the public cloud provides are way too compelling. With the rise of serverless computing, container-based computing, and machine learning, high-performance computing systems can easily mix services as needed, with the integration of those services being a relatively simple process.
Indeed, like other reasons that companies are picking the cloud, it’s not about infrastructure anymore. It’s all about platform features. The attraction to those platform features is what grows the infrastructure. Razors need razor blades. And those razors are in the cloud, not in the datacenter.
I suspect that on-premises high-performance computing will continue to be a niche play, with the primary users being research, graphics processing, and other use cases that businesses dabble in when they need to. The ability to use high-performance computing as a service, via a public cloud, is just too compelling. The public cloud is just too cheap, too good, and too tempting.