AI, ML and other data-intensive workloads are making the old definition of big data quaint. Today's Performance Intensive Computing as a Service (PICaaS) providers require networks capable of transporting data at unprecedented scale.
Integrating AI into your business is a complex undertaking. Check out this AI-focused white paper to add to your understanding of how colocation data centers can fit into your evolving multicloud and network infrastructure.When big data was a new challenge, we often said that transporting and analyzing large datasets was like “drinking from the firehose.” Fast-forward to the current technology cycle. The data generated through artificial intelligence and other performance-intensive computing workloads, content streaming and other sources has me reaching for an apt metaphor. “Drinking from Niagara Falls” comes to mind.
Here’s why. Cloud service providers are adding GPUs to their infrastructure pools, enabling you to choose the best “engine,” for your application. The data processing difference between CPU servers and GPU servers is astounding. A CPU server typically has up to 30 cores, whereas a GPU could have 40,000!
Cloud service providers (CSPs) are now also offering bare metal servers for high-density use cases. Bare metal servers provide a computing environment dedicated to a single application, single cloud or colocation client, and can be optimized specifically for machine learning, generative AI or foundational models trained on a broad set of unlabeled data that can be used for different tasks, with minimal fine-tuning. A GPU-equipped bare metal server is a serious data crunching machine.
AI has given rise to a new product: Performance Intensive Computing as a Service (PICaaS). This new infrastructure market is growing rapidly; IDC is forecasting this segment to reach $103B by 2027 or 12.5% of the total as-a-service market.1 It’s one more example of the many new services and start-ups spawned by AI evolution, which I discussed in a previous AI-focused blog: AI Models, AI Providers and Data Centers Keep Learning.
PICaaS requires networks capable of transporting data at unprecedented scale. That also was true back in the early days of big data. At that time, service providers responded by offering private networking options known as “onramps.” These direct connections were the best choice for companies that wanted to bypass the internet to improve throughput, latency and security.
The rapidly increasing number of enterprises that wanted to use public clouds caused a new problem: managing the onslaught of connections. CSPs came to CoreSite and one of our competitors for help. What’s interesting is that the solution then – CoreSite data centers serving as a hub for aggregating the enterprise networks and exchanging data with CSPs through a direct connection in our facilities – is identical to now, only with much higher bandwidth demands. GPU and bare metal service providers need onramps to clouds, where their services reside.
CoreSite has successfully enabled CSPs to deploy private, high-bandwidth network onramps for a decade and would be an ideal partner for AI service providers that are expanding their addressable market while enhancing their competitive advantage. We’ve worked to ensure we are future-ready, with data centers that can support emerging use cases without retooling the platform we’ve built. As an aside, that’s true for all the types of enterprises we are thrilled to call customers, not just AI solutions.
We also developed and evolved the Open Cloud Exchange® (OCX) – an industry leading networking platform for managing interconnections, bandwidth allocation, and automated network functions – simplifying those complex multicloud and wide area networking tasks.
The OCX automates virtual network interconnection to CSPs, between data centers and to systems integrators, cybersecurity solutions, technology solutions, managed services providers and companies that are members of the CoreSite fabric.
AI services providers can utilize the dense network connectivity available in our data centers, establish cross connects to their customers when possible, and take advantage of true direct cloud connections and the OCX to optimize usage and therefore costs for services offered in public cloud.
It’s sometimes lost that colocation data centers are an ideal environment for hosting AI services, in private clouds and dedicated servers. Conversely, one of the key findings in the 2023 State of the Data Center Report is that “92% of decision makers surveyed are more open than ever to moving critical workloads from public cloud to colocation, up from 68% in 2022.” I’m not at liberty to reveal specifically what that trend looks like in the 2024 State of the Data Center Report, but I can say you’ll see more of the same.
CoreSite now has 50G connectivity to CSPs and is certified in the NVIDIA DGX-Ready Data Center program in several locations. Our facilities also are designed to implement a range of cooling solutions, including liquid cooling, and have a long track record of meeting dense power requirements, just like what we see in bare metal GPU servers.
Steve Jobs said: “Some people say, ‘Give the customers what they want.’ But that’s not my approach. Our job is to figure out what they’re going to want before they do.”2
With all due respect to Mr. Jobs, that’s not our approach. As much as we work to anticipate what our customers will need and watch for the next disruptive technology, we work from a position of strength established by experience, keeping a pulse on the industry and listening to our customers.
Twenty-five years ago, they needed a way to bypass the internet. Today, many need a way to find value in a waterfall of data. While I can’t say I know what’s next, I will say that CoreSite and the many companies we partner with are ready to enable our customers to take advantage of opportunities when they knock.
Know More
Learn more about CoreSite's solutions for artificial intelligence and download our AI white paper to better understand how to get the most value from your IT modernization initiatives.
References