Microsoft and Meta Boost Data Center Leases as AI Computing Power Demand Soars

Microsoft’s latest data‑center lease agreements signal a decisive shift toward massive on‑premise AI workloads, according to a Bloomberg report. The company has earmarked new facilities in several regions, including a series of hyperscale sites in Africa that will host custom AI accelerators for training large language models (LLMs). By co‑locating these accelerators with its Azure cloud backbone, Microsoft aims to cut inference latency for African developers and enterprises while also reducing the bandwidth costs associated with moving petabytes of training data across continents. The Bloomberg piece notes that the African rollout is part of a broader “AI for Africa” challenge that pits Microsoft against Chinese startup DeepSeek, which has been courting the same market with its own cloud‑native AI chips.

Meta’s data‑center expansion follows a similar logic, as detailed in a report from 富途牛牛. The social‑media giant is signing long‑term leases for additional floor space in existing hyperscale campuses, with a focus on GPU‑dense clusters optimized for generative‑AI inference. Meta’s internal roadmap, which the report links to, calls for a 30 % increase in AI‑specific compute capacity by the end of 2025, a target that can only be met by securing more power‑rich real‑estate in low‑cost regions. The cumulative effect of Microsoft’s and Meta’s commitments pushes the total value of cloud‑provider data‑center leases past the $700 billion mark, a figure that underscores the scale of capital being funneled into AI‑centric infrastructure.

The financial magnitude of these leases reflects a broader industry trend: AI workloads now dominate capacity planning for the biggest cloud operators. Forbes highlighted Microsoft’s $290 million investment in South Africa’s AI and cloud ecosystem, a sum that will be used to build out both edge nodes and a regional AI research hub. Vice‑chair Brad Smith announced that the funds will support “training 3 million Africans on its AI technology this year,” a program that couples hardware provisioning with education initiatives. By embedding AI training resources directly within the continent’s data‑center fabric, Microsoft hopes to create a virtuous cycle where local talent can develop models that, in turn, drive demand for more compute capacity.

Wired’s coverage of a South‑African startup attempting to reverse brain‑drain by establishing an AI research lab adds context to the competitive environment Microsoft is entering. The article describes how the startup’s plan to attract expatriate engineers hinges on access to world‑class AI hardware, which is precisely what Microsoft’s new leases will provide. While the Wired piece does not quantify the startup’s hardware needs, it emphasizes that “high‑performance GPU clusters” are the linchpin for any serious AI research effort. This aligns with Meta’s strategy of expanding GPU‑dense clusters, suggesting that both firms are positioning their leased capacity as a platform for third‑party developers and research institutions that lack the capital to build their own hyperscale facilities.

Taken together, the data‑center lease surge illustrates a capital‑intensive pivot from traditional cloud services to AI‑first infrastructure. The $700 billion threshold, first reported by 富途牛牛, serves as a benchmark for the escalating cost of power, cooling, and real‑estate required to sustain the next generation of LLMs and multimodal models. As Microsoft and Meta lock in long‑term leases, they also secure preferential access to the power‑dense sites needed for next‑generation AI chips, which are expected to deliver orders‑of‑magnitude improvements in FLOPS per watt. The strategic implication is clear: control over physical AI compute resources is becoming as valuable as the software platforms that run on them, and the race to lock down that control is now being waged on a global, multi‑continental scale.