We've Seen this Before... Introduction of a Private AI Exchange

Author: Chris Sharp, Chief Technology Officer, Digital Realty
April 24, 2024

Private AI Exchange (AIPx) Powered by ServiceFabric™

Today we’re announcing the introduction of Digital Realty’s Private AI Exchange (AIPx). AIPx allows businesses, service providers, and other technologies to be quickly and securely integrated into an open ecosystem where artificial intelligence (AI) solutioning can take place as the network, data, partner, and service enablement of private AI.

This platform is built on ServiceFabric™, Digital Realty’s orchestration platform that seamlessly interconnects workflow participants, applications, clouds, and centers of data exchange. All existing ServiceFabric™ customers and partners will have access to this exchange as it rolls out.

Why are we building this new platform? AI, and generative AI in particular, is driving rapid advancement in new technology and turning concepts that sound like science fiction into reality. AI, when combined with the most critical element, relevant, accurate, and properly processed data, will change the way businesses build, connect, and operate their infrastructure.

Previously, I had the opportunity to build one of the first cloud access solutions because both the cloud providers and their customers had a common issue – the workloads that they wanted to deploy just wouldn’t reliably work over the internet or they required the performance and security from more dedicated network connections. Now AI workloads are raising the same problem, and our customers are again looking for a solution. In fact, 50% of enterprise platforms are expected to adopt new infrastructure just for AI data infusion in the next five years.

Connecting public cloud to on-premises resources has been a standard practice in hybrid cloud since the beginning. The easy method was over the public internet, and it mostly worked. But as these mission-critical workloads continued to grow in complexity and sensitivity, a better method was needed – private cloud access.

Private access combined products from cloud providers like Amazon DirectConnect, Azure ExpressRoute, and Google Dedicated Interconnect with a software defined networking (SDN) platform, such as Digital Realty’s ServiceFabric™.

These products enable low-latency, high-performance, and secure on-demand access between cloud providers and customer and partner assets. With this technology, resources could be combined to get the best combination of scale, cost, control, protection, and on-demand consumption models – all while treating these capabilities as one logical architecture, despite running on different companies, in different locations.

The rise of hybrid AI

Now with these new AI associated workloads, we again have the challenge of linking together many different types of assets into a common solution. The lifeblood of AI is data, and lots of it.

That data also needs to be current, relevant, and properly processed for consumption (whether it is public in nature – or ‘crown jewel private’ within an enterprise). Storing, processing, and using AI for inference all requires significant and different types of resources, all of which must be connected in the highest performance, reliable, and secure method possible.

With the rise of cloud computing, this data now lives in many different locations used by enterprises daily. These cloud-as-a-service platforms each have data that can be incorporated to generate business value.

AI architecture evolution: We see where this is going

Companies are now attempting to integrate their data with AI platforms using the standard methods available, mostly the public internet. However, the same challenges for hybrid clouds can limit hybrid AI and the ability to manage public or private data, models, training, inference, and similar.

With hybrid AI, these challenges can become even more complex. While cloud assets can run in isolation depending on the workload, in order to have meaningful AI results, businesses will want to integrate all valuable datasets with their combined AI architecture.

An architecture to support hybrid AI securely integrates best-in-class resources with the business-specific data to build a flexible and dynamic solution that can easily be extended with minimal impact. While some solutions could be built entirely on cloud for AI workloads, which could mean high costs and scarce availability of specialized hardware, tomorrow’s solutions are trending to a deployment of per-workload AI factories into data centers, which become core to this AIPx framework.

Supporting these solutions requires a complete rethink of legacy infrastructure design as enterprise architecture must now become datacentric. Data silos will be required to become data lakes, and those data lakes will become the key location where companies bring AI technology to their data, rather than the other way around.

AI deployments will be rapidly ingesting the data in as near-real-time as possible, to gather business-critical insight, which is not possible with stale data. All of this will require high network capacity, reliability, low latency, and security to bring it all together.

Digital Realty AIPx: the open ecosystem platform enabling interconnected AI architecture

This is why we’re announcing the Digital Realty AIPx. AIPx is an open ecosystem enablement platform, allowing service providers, businesses, and other technology solutions to be integrated in minutes into a complete AI solution.

AIPx enables enterprise data to be at the center of an AI architecture, with the AI capabilities deployed and interconnected directly adjacent. The benefit of this architecture is to remove Data Gravity barriers that emerge as more data is generated and exchanged with multiple applications across end-to-end AI-enabled digital workflows. Additionally, the traditional private interconnection approach of cross connects, and uni-directional interconnection backhaul must evolve towards a multi-directional, dynamically orchestrated implementation.

With AIPx, businesses can rapidly create and extend AI solutions, including their own assets and those of the newest providers coming to market, either via our customer portal or programmatically using our open API framework. This enables access to all the AI services, models, data, and high-throughput, seamless network capabilities required to ensure a complete and modern AI solution. All done over a private data fabric.

We are already aligning with strategic partners for core AI infrastructure, services, data, network (optical exchange), broader, and related AI On-ramps, which focus on AI-oriented workflows, integration, and automation, with ServiceFabric™.

Building off these new capabilities, today, enterprises can bring AI factories to their private data sets on PlatformDIGITAL® globally. By utilizing our existing High-Density Colocation solutions with large capacity blocks, companies can address the needs for large contiguous data deployments adjacent to their users and cloud providers, in addition to supporting the unique physical and cooling attributes of their high-performance computing infrastructure.

Stay tuned, we will be making announcements, releasing use cases, and customer case studies as we bring them onto our AIPx platform. Look for additional information as we extend this next evolution of our Data Meeting Place to the industry and broader partner participation. Digital Realty is uniquely positioned to provide the necessary innovation, capabilities, expertise, and global platform that help both enterprises implementing AIPx along with the technology providers who serve them.