At the recent Canalys Forum EMEA held in Berlin, key discussions revolved around the mounting challenges faced by organisations as they navigate the complexities of AI workloads. As demand for advanced AI capabilities grows, coupled with escalating energy costs, businesses are compelled to rethink their infrastructure strategies, particularly in terms of where to host these workloads.
Alastair Edwards, chief analyst at Canalys, highlighted that CIOs are confronted by a significant dilemma: how to accommodate the high-power requirements associated with training and deploying AI models without succumbing to the financial pressures of rising energy bills. During his remarks at the forum, Edwards noted, “Every company is trying to figure out what model or IT architecture they need to deploy to take best advantage of the business transformation that AI promises.”
Historically, businesses have leaned towards public cloud vendors as the preferred option for training AI workloads due to their robust infrastructure capabilities. Recent estimates suggest an increase of about 30 percent in capital expenditure for AI-capable servers this year among these providers. However, as companies transition from training to deployment phases—particularly for fine-tuning and inferencing—they are finding that relying solely on public cloud options can become unsustainable from a cost standpoint. Edwards articulated this concern by stating, “The public cloud, as you start to deploy these use cases we're all focused on and start to scale that, if you're doing that in the public cloud, it becomes unsustainable from a cost perspective.”
The shift has led many organisations to reconsider their choices. According to Edwards, necessity drives companies away from building their own on-premises data centres, which they perceive as burdensome regarding energy demands and infrastructure management. “Almost no organisation these days wants to build their own on-prem datacentre,” he remarked, suggesting a growing desire for control, security, and compliance without the hefty infrastructure overhead.
Colocation and specialized hosting providers are emerging as viable alternatives, allowing companies to mitigate challenges while accessing necessary resources. There is a noticeable trend towards new business models, such as those providing GPU-as-a-service, which are designed to facilitate access to GPU capacity for AI tasks. Companies like Coreweave and Foundry are leading this space, and even Rackspace has re-entered the market with a GPU-as-a-Service offering.
Market analytics from IDC conjoin with Canalys' perceptions, projecting a notable increase in infrastructure investment aimed at AI deployment. A reported 37 percent increase in corporate spending on compute and storage hardware for AI was recorded in the first half of 2024, with forecasts suggesting that funding in this area could surpass $100 billion by 2028.
Notably, large technology firms are making substantial investments to bolster their data centre infrastructure. Microsoft has announced intentions to allocate $100 billion for data centre enhancements, while Amazon Web Services plans to invest $10.4 billion specifically in facilities in the UK. Recent data indicates that AI-enabled systems within cloud and shared environments accounted for 65 percent of server expenditures related to AI during the first half of 2024, primarily driven by hyperscalers and cloud service providers.
Despite these robust growth spans, IDC also noted regional disparities in AI infrastructure investments. The United States currently commands nearly half of global AI infrastructure spending, followed by China and the Asia-Pacific region, while EMEA accounts for just 10 percent. However, over the next five years, the Asia-Pacific region is set to experience the fastest growth, predicted to reach a compound annual growth rate (CAGR) of 20 percent.
The emphasis on AI adoption is becoming increasingly pronounced, with many stakeholders—be they hyperscalers, cloud service providers, private enterprises, or government bodies—reshaping their priorities to centre around AI initiatives. Nevertheless, potential challenges loom on the horizon, particularly regarding energy consumption prompted by AI workloads. Lidice Fernandez, IDC’s Group VP of Worldwide Enterprise Infrastructure Trackers, cautioned, “Growing concerns around energy consumption for AI infrastructure will become a factor in data centres looking for alternatives to optimize their architectures and minimize energy use.”
Recent forecasts have signalled alarm bells regarding the escalation of energy demands associated with AI data centres, predicting a growth of up to 160 percent in the coming two years, outpacing the capabilities of utility providers to meet increased electricity generation needs. While this scenario could complicate future data centre expansion plans, current investment trends are yet to show signs of retreat, even as a recent survey revealed that 98 percent of investors express concern regarding energy availability.
Source: Noah Wire Services