The party has started

AI exchanges: The evolution of AI-ready digital infrastructure

Travis Ewert, Global Head of Platform Development, Digital Realty

Earlier this year, we shared an overview and introduction to Private AI Exchange (AIPx). AIPx allows businesses, service providers, and other AI technology enablers to quickly and securely integrate into the open Digital Realty ecosystem.

We saw the need for AIPx based on trends with artificial intelligence (AI) and, more acutely, with the challenges that our customers and partners are working through daily. AIPx was introduced to integrate private network (fabric) and connectivity (dedicated) to an AI solution ecosystem (exchange) that includes providers, services, and infrastructure.

In this blog, I’ll cover updates around this emerging ecosystem, other providers ‘joining the party,’ and additional efforts we are pursuing to evolve end-to-end AI solutioning within AIPx. Specifically, this evolution relates to the network (and its three dimensions), compute infrastructure, applications, data-as-a-service, and innovative VARs, SIs, and OEMs building against these new reference architectures.

Back to the start... next-generation interconnection solutions

As we called out back in April, new technologies like generative AI are transforming the way digital infrastructure providers do business. This evolution is changing the nature of the network and related traffic flows in support of:

• Data acquisition
• Transformation
• Training
• Tuning
• Inference

It also changes how these workflows are architected across hybrid private and public infrastructure. These new architectures mirror what industry analysts like IDC are stating that AI-driven workloads are redefining hybrid cloud and multi-cloud networking.

Key drivers of this evolving network architecture and business model include:

• Hybrid AI: Hybrid AI refers to the integration of various AI techniques and models, deployed across both private and public environments. This approach enables organisations to optimise complex AI workflows by blending the scalability of public cloud resources with the security and control of private infrastructure. Infrastructure designed for hybrid AI provides flexibility to scale AI processes in public environments while ensuring sensitive data remains secure and performance-critical tasks are handled in private settings.
• AI on-ramps: Like cloud on-ramps, AI on-ramps enable enterprises to connect their private data sources and infrastructure to the ecosystem over dedicated, private connections ensuring data protection, compliance, and performance.
• Private networks and fabrics: As AI legislation and end-user privacy comes into the forefront, the role of the network at a local and global level will be critical. This is to ensure not only privacy, but also performance (i.e. latency) and scale. As a result, enterprises will also leverage principles for network-on-demand to consume network when required.

The new ecosystem of digital infrastructure for the AI era

As a result of the increasing change in network architecture, we’ve witnessed more providers ‘join the AI exchange party.’

Over the past couple months, Internet Exchange Providers, large Network Service Providers with private fabrics, Network-as-a-Service providers, and many of the largest networking OEM providers followed our lead, making their own announcements around the need and support of AI exchanges and networking technology.

In the six months since our AIPx announcement, we’ve witnessed a rise in AI-focused, GPU-as-a-Service providers, which offer on-demand access to GPUs through the cloud. Additionally, there’s also an increase in ‘AI neocloud’ providers—building GPU processing capabilities at scale.

These AI factories—AI infrastructure solutions aimed at simplifying the adoption and integration of AI for businesses—are seeking better ways to extend AI solution development into the larger ecosystems of AI providers and consumers. Innovative solutions like these are accessible with AIPx, where Digital Realty provides a global platform and open ecosystem for these solutions and the AI on-ramps to connect to the exchange.

Why are these AI factory instances landing within our footprint? We are leading the way and building data centre infrastructure that not only support high-power dense racks and liquid cooling solutions but also provide flexible, low-latency, software-defined networking capabilities. This approach is inspiring the development of some of the most interconnected and power-efficient data centres ever built, as our customers deploy some of the largest training clusters currently in production. As these AI deployments grow, they are creating new demands on network architecture, sparking what we call the "AI network effect."

New architecture considerations for the ‘AI network effect’

At Digital Realty, we’re seeing the “AI network effect” in three interconnection areas:

1. Intra-datacentre, or unique networking within a data canter
2. Intra-campus, or networking within a metro or campus of connected data centres
3. Inter-campus, or networking between geographies and across the globe

While each of these interconnection groupings involves technological differences, they all rely on common concepts and customer requirements like:

• High bandwidth capacity
• Low (and normalized) latency
• Scalability
• Security

Further, AI-controller (orchestration) smarts for realtime discovery, inventory, availability, of connected infrastructure, services, and networks within the broader AI exchange. You will also see the introduction of what we have coined as “PNaaS” (or ‘physical’ Network as a Service) where robotic automation of cross-connect connectivity plays a role for larger optical exchanges (on-demand and orchestrated as well).

All of this is evolving quickly, and as adoption continues to accelerate, AI workloads and related dataflows are shifting to inference, beyond that of localised training. We anticipate widespread inference deployments will only accelerate the demand for dramatic bandwidth increases and disaggregated network access across the referenced ecosystem of network and service providers.

What the future holds for the next-generation AI ecosystem

At Digital Realty, we’re at the forefront of AI ecosystem readiness and implementation within our global data centres. We’re building the next generation of interconnected (AI) exchange and connected fabrics. We believe that beyond networks, clouds, and services—the solutions our customers seek will require a global, neutral, and open multi-provider and multi-datacentre partner ecosystem.

ServiceFabric™ powers AIPx as our open global orchestration and interconnection platform. It has 120 partner ecosystem participants, offering 220+ services, accessed within ServiceDirectory™. Accessible AI services are available for enterprise customer use. Additionally, we also have 245+ cloud on-ramps that participate within our hybrid-AI ecosystem. AIPx supports on-demand data centre site-to-site connectivity across 525 locations for private fabric connectivity.

All of this is expanding at an increasing pace as we bring on more partners to expand both reach (destinations) and services (AI applications and infrastructure).

We’re working with industry-leading VARs, SIs, and OEMs, to develop AI solutions that leverage our densified High-Density Colocation for space, power, and cooling. This will also leverage ServiceFabric™ to support AIPx access, Hybrid-IT, and multi-cloud enablement as a broader bundled solution.

Our original Private AI Exchange (AIPx) announcement received a strong response. Increasingly, this is inspiring leading providers across multiple service offerings to collaborate with us on additional design and architecture considerations as we view this as an industry collaboration.

Stay tuned for more updates as the market and our progress with key partners continues. Welcome, everyone, to the AI exchange party.

Are you ready to take your enterprise into the future with the next-generation AI ecosystem? Get your VIP pass to the AI exchange party today. Please reach out to us to learn more about our global, neutral, and open partner ecosystem.

Not ready to take the plunge, but want to learn more? Explore how our colocation solutions meet the demanding needs of AI. Read the IDC Link commentary here.