The AI infrastructure buildout shows little sign of slowing, even as technology companies and enterprises face growing questions about power availability, supply chains, hardware costs, and the economic return from artificial intelligence.

Worldwide data center capital expenditure is forecast to exceed $3 trillion by 2030, according to Dell'Oro Group's Data Center IT Capex 5-Year July 2026 Forecast Report. The research firm says the outlook has increased sharply from its January 2026 forecast, driven by higher hyperscaler spending guidance, stronger expectations for global data center power capacity, and rising commodity costs.

At the center of that spending wave are high-end accelerators used in AI-optimized servers.

Unlike conventional enterprise computing, modern AI workloads require large quantities of specialized processors for model training and inference. NVIDIA's GPUs have become the most visible example, but the broader accelerator market also includes alternatives from AMD, Google, Amazon, and other semiconductor vendors.

Dell'Oro expects these high-end accelerators to represent the largest portion of data center capital expenditure through 2030. That makes AI hardware a key determinant not only of data center construction but also of the economics of cloud computing and enterprise AI adoption.

The scale of the projected investment is significant. The largest U.S. hyperscalers alone could account for roughly half of global data center capex, according to Dell'Oro Vice President of Research Baron Fung.

That concentration highlights the growing influence of companies such as Amazon, Microsoft, Google, and Meta on the global AI infrastructure market. Their expanding AI services require enormous quantities of compute, networking, storage, and power, creating a feedback loop between AI demand and data center investment.

Yet the next stage of the buildout may look different from the first.

The initial AI infrastructure surge has largely been associated with training increasingly sophisticated models. As generative AI applications mature, inference is expected to become a much larger component of workloads. AI agents, enterprise copilots, recommendation systems, search experiences, and real-time applications can all generate sustained inference demand.

Dell'Oro expects general-purpose servers to benefit from that expansion, along with growing storage and agentic AI workloads. That suggests data center spending will increasingly extend beyond specialized GPU clusters.

The change matters for enterprise technology buyers. AI infrastructure is not simply about purchasing accelerators. Production AI systems require conventional CPUs, storage, networking, virtualization, orchestration, databases, cybersecurity, observability, and increasingly sophisticated cooling and power systems.

The result is a more heterogeneous computing environment.

Dell'Oro says accelerated and heterogeneous computing, combined with improvements in server efficiency, could help offset some of the rising costs and infrastructure requirements associated with AI. In other words, increasing AI demand does not necessarily mean every workload will run on the most expensive accelerator available.

The industry is also seeing a new customer segment emerge: AI-specialized cloud providers, or neoclouds.

Dell'Oro projects this segment, which includes AI model builders and neocloud service providers, to grow at nearly 60% compound annual growth through 2030. These providers are building infrastructure specifically around accelerated computing and AI workloads, creating an alternative to conventional hyperscale cloud services.

That growth could reshape the competitive landscape.

Companies that need AI capacity but cannot justify building large private clusters can increasingly rent specialized infrastructure from providers optimized for GPU workloads. At the same time, hyperscalers are investing heavily in their own AI infrastructure and developing custom silicon to reduce dependence on third-party accelerators.

The result is likely to be a highly competitive market spanning hyperscale clouds, neocloud providers, semiconductor companies, server manufacturers, networking vendors, and data center operators.

Power, however, may become the industry's most significant physical constraint.

AI accelerators consume substantial amounts of electricity, and high-density AI clusters can require different power and cooling architectures than conventional data center deployments. Building new capacity can also take years because of grid interconnection requirements, permitting, equipment availability, and local infrastructure constraints.

Dell'Oro's forecast specifically identifies power availability and supply-chain conditions as factors that could influence the pace of future investment.

For technology companies and enterprise buyers, the implication is straightforward: AI infrastructure decisions are becoming long-term capital allocation decisions rather than short-term IT purchases.

The question is no longer whether AI will require more compute. It is how much compute will be economically sustainable, where that capacity will be located, which workloads deserve specialized acceleration, and how quickly the industry can convert infrastructure investment into measurable business value.

Market Landscape

The AI infrastructure market is expanding across several layers simultaneously.

At the hardware level, NVIDIA, AMD, Google, Amazon, and other chipmakers are competing to supply accelerators and custom AI silicon. At the infrastructure level, hyperscalers such as Microsoft Azure, AWS, and Google Cloud are expanding data center capacity while specialized neocloud providers target customers with demanding GPU workloads.

Meanwhile, enterprise organizations are beginning to invest in AI infrastructure for private and hybrid environments. This creates demand for platforms capable of managing both traditional workloads and accelerated AI computing.

Dell'Oro's forecast suggests that high-end accelerators will remain the largest spending category through 2030, but growth in inference, agentic AI, storage, and general-purpose computing indicates that the AI data center will become increasingly heterogeneous.

Strategic Outlook

The $3 trillion capex projection illustrates both the opportunity and the risk surrounding the AI infrastructure boom.

Investment can create the capacity needed to support a new generation of AI applications, but infrastructure spending must eventually be justified by utilization and economic returns. Dell'Oro notes that enterprise investment remains constrained by uncertainty around AI returns, while hyperscalers continue to drive a disproportionate share of global spending.

The next phase of the market will therefore be defined by efficiency as much as scale. Better accelerators, custom silicon, advanced cooling, optimized networking, workload scheduling, and efficient inference could determine which AI infrastructure investments deliver sustainable returns.

For enterprise technology leaders, that means AI strategy increasingly needs to account for infrastructure economics. Model selection, cloud architecture, data governance, workload placement, and compute utilization are becoming interconnected decisions.

Top Insights

• Global data center capex could exceed $3 trillion by 2030, underscoring the extraordinary infrastructure investment required to support expanding AI workloads.
• High-end AI accelerators remain the primary spending driver, making GPU availability, pricing, efficiency, and utilization critical concerns for cloud providers and enterprises.
• AI-specialized neoclouds are expanding rapidly, with Dell'Oro forecasting nearly 60% CAGR as organizations seek flexible access to accelerated computing.
• Inference and agentic AI will broaden infrastructure demand, increasing the importance of general-purpose servers, storage, networking, and orchestration alongside accelerators.
• Power availability could constrain AI growth, forcing data center operators to prioritize energy efficiency, advanced cooling, and strategic infrastructure locations.

Get in touch with our MarTech Experts