Ivo Ivanov, CEO of DE-CIX, explores how companies can better manage their AI data and clouds effectively, and achieve viable returns on their AI investments.

We may well look back on 2023-4 as the year of the AI boom. Microsoft declared a $10 billion investment in OpenAI, the number of people using AI tools around the world surpassed 250 million, and Dictionary.com declared ‘hallucinate’ its 2023 Word of the Year, owing to increased interest in generative AI tools such as ChatGPT, Bard, and Grok.

Businesses are paying attention to AI and their deployment brings countless benefits, from increased productivity and automation to data-driven decision-making and new revenue opportunities. However, despite these benefits, one critical question is emerging in boardrooms around the world: Are we really achieving our intended return on our AI investment?

Once the shine of new AI deployments wears off, those in charge have the unenviable task of ensuring that they’re getting the most out of their technology. In the race to AI success, many businesses underestimate the role their data and connectivity infrastructure play in enabling and supporting AI deployments, and that can lead to serious bottlenecks. Some may opt for a hybrid cloud environment while others may decide to build their AI applications fully in the cloud – but in both cases, latency, speed, bandwidth, and the sheer cost of the compute power needed can be limiting factors.

Businesses recognize this challenge. According to a survey by the MIT Technology Review, 95% of companies are already using AI in some form, and around half expect to deploy AI across all business functions within the next two years. However, the same survey revealed that the main challenges to successful AI implementation are data quality (49%), data infrastructure or pipelines (44%), and data integration tools (40%). Similar concerns were raised by European organisations surveyed by IDC, of which 22% list network performance and latency as their main concern when using or planning to use AI from the cloud, especially for use cases requiring real-time data.

Network performance: The elephant in the room?

Businesses are now facing something of a dilemma. AI is ready for them, but are they ready for AI? As data generation continues to surge, organizations are understandably finding it challenging to store all relevant information on their own infrastructure, instead choosing cloud-based data lakes and warehouses to stockpile both raw and structured data. This makes sense, however, these data sources are only useful from an AI perspective if they can be seamlessly integrated with AI models, many of which will already be in the cloud.

This gives businesses a problem. They need to move their data off-site, but if their data is off-site while they continue to deploy AI-based solutions, they are at the mercy of connectivity. It’s important to establish the distinction between AI training and AI inference. Training AI models – whether it’s building them from scratch or retraining them periodically – doesn’t necessarily require low latency connectivity, but it does depend on high bandwidth. To maximise the benefits of cloud infrastructure without incurring excessive data egress costs, businesses are turning to direct connectivity options provided by cloud vendors. When it comes to AI inference – where the model’s output is used in real time – the situation changes. Here, low latency becomes essential. Whether it’s customer service chatbots, marketing optimisation, or product development, speed and responsiveness are simply non-negotiable.

That means a network capable of handling both the high bandwidth demands of training and the low latency requirements of inference is also non-negotiable. For AI to truly to deliver on its promise, high-performance connectivity between data sources, cloud environments and AI models must be established. But how?

Ensuring AI readiness with interconnection

Businesses are investing heavily in AI, yet continue to connect their on-premise hardware, data warehouses, and AI cloud services through the public internet or third-party IP transit, leaving them with little control over data routes and no guarantees on performance. This not only impacts latency but also poses security risks to sensitive company data. To avoid these issues, businesses need dedicated, secure, and direct connections between their networks and the various clouds, services, and applications they rely on.

Managing these data flows effectively requires something known as network interconnection. Direct, high-performance links between corporate networks and cloud platforms, often facilitated through Cloud Exchanges with cloud routing capabilities, are crucial for establishing a responsive, interoperable multi-cloud or hybrid cloud environment.

What’s more, interconnection with external networks via an Internet Exchange – whether through peering or private network interconnects – ensures that data takes the most efficient path between endpoints, offering secure, low-latency, and resilient connectivity as a result. By extending this direct connection to AI-as-a-service networks through an AI Exchange, businesses can even outsource AI development and operations to third-party providers, supporting a multi-AI approach without sacrificing performance or security.

To approach this effectively on a regional or global scale, businesses are increasingly relying on high-performance interconnection providers – such as Internet, Cloud, and AI Exchange operators – for both connectivity solutions and strategic network design. It is these neutral operators that will ultimately enable an AI-ready world where businesses can not only deploy AI, but master it.