Investment in global data centres is set to reach an impressive $1.6 trillion by 2030, highlighting a significant shift in the technology landscape. A recent report from Omdia predicts that major tech firms will allocate over $600 billion towards artificial intelligence capital expenditure in 2026 alone, signalling a robust commitment to advancing AI infrastructure.
This surge in funding indicates that the AI Factory market has surpassed a critical threshold, evolving into a model that demands substantial capital, navigates geopolitical influences, and tackles intricate engineering challenges. Omdia defines the AI Factory as a specialised facility focused on producing intelligence, with tokens serving as the primary output.
The transformation of data centres is profound; they are no longer merely support systems for businesses but are evolving into digital manufacturing hubs that generate high-value intelligence. These centres are adopting a four-layer architecture encompassing physical and energy infrastructure, network fabrics and hardware, virtualisation orchestration, and a Model as a Service ecosystem.
The current market landscape is diverse, featuring four distinct solution paradigms, including public cloud hyperscalers, compute-native specialists, private foundation providers, and regional infrastructure operators. A recent survey of over 200 businesses highlighted critical challenges in the industry, such as lengthy time-to-market, digital sovereignty, talent shortages, and systemic engineering complexity.
As the industry pivots away from compute hoarding amid the Zombie GPU effect—where expensive processors remain idle during I/O wait times—evaluation benchmarks are being redefined. Emphasis is now placed on Time-to-First-Token and vector retrieval speed, with studies revealing a 12-times improvement in indexing speed and up to a 75% reduction in compute costs.
Hyperscalers are adeptly balancing agility with sovereignty by offering fully integrated physical units while allowing for hardware and software decoupling. Rack power density has seen a remarkable increase, soaring from 10-15 kW in 2024 to as high as 250 kW by June 12, 2026, as workloads transition from proof-of-concept to production-grade deployments.
Companies like Nebius and Sensetime are redefining their business models, evolving from basic hardware leasing to a Model as a Service framework. Sensetime, in particular, is adopting an integrated strategy that encompasses infrastructure, software, and energy management, thereby enhancing control over computing resources.
Value capture is increasingly concentrated among vertical integrators and domain operators, who leverage long-term data governance and legacy system integration. Inspur Cloud is actively pursuing a strategy that blends heavy-asset infrastructure with specialised operations, expediting the journey towards full AI industrialisation.
New regulatory frameworks, such as the EU AI Act, are compelling organisations to maintain sensitive information within isolated facilities. This shift has elevated regional operators like G42 from traditional infrastructure landlords to crucial physical gatekeepers of national data.
According to Raymond Zhan, Senior Principal Analyst for Cloud and AI at Omdia, “Future competition will no longer be defined by model parameters or GPU counts, but by a comprehensive contest of energy, liquid cooling, chips, autonomous software stacks, sovereign compliance, and long-term capital endurance.” He further notes that for enterprise clients, the landscape of AI factory providers is not one-size-fits-all; choices must be tailored to align with actual business scale and the balance between steady-state and innovative workloads.
Omdia anticipates that the years 2026 and 2027 will be pivotal for the expansion of data centres, with regional and industrial operations expected to emerge as the most reliable growth sectors over the next five years.
