Gartner projects public cloud reaching $1.48 trillion by 2029, doubling from 2025. Getting there requires physical infrastructure, energy capacity, and organizational governance that are all under real strain right now.
I was reading the Gartner April 2026 IT spending forecast last month and one number kept pulling my attention past the headline (https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026). Total worldwide IT spending is projected at $6.31 trillion in 2026, with 13.5% growth. That is a large number. But the number that actually stayed with me was further down: Gartner projects public cloud end-user spending will reach $1.48 trillion by 2029, against a 2025 base of roughly $723 billion. That is a doubling in four years. A $723 billion market doubling in four years requires things to go right on the demand side and the supply side simultaneously. I want to think through what "going right" actually requires, because I think the assumptions embedded in that trajectory are worth examining.
The demand-side story is plausible and I find it mostly convincing. The primary new driver is generative AI and the infrastructure requirements it brings. Training large language models requires enormous amounts of GPU compute that very few organizations have on-premise. Running inference at scale, serving responses to millions of users in real time, requires persistent, high-throughput, low-latency infrastructure. The preprocessing pipelines, embedding generation, retrieval-augmented generation architectures, fine-tuning workflows: all of this is cloud-native in the sense that the capital cost of building it on private infrastructure is prohibitive for most organizations. The Gartner April 2026 forecast projects AI spending at $2.5 trillion in 2026, and that number is itself partially a cloud spending story. Every organization that adds AI capabilities to its products or operations is adding cloud consumption that did not exist before. Gartner also projects that agentic AI will be embedded in 40% of enterprise applications by the end of 2026, which means the AI-driven cloud demand is not a single spike. It is a persistent baseline increase in compute requirements.
So the demand trajectory makes sense. The supply side is where I have more questions.
Building data center capacity at the scale required to serve $1.48 trillion in cloud demand is not just a capital allocation problem. It is a physical infrastructure problem. Power is the most constrained input. Modern GPU clusters designed for AI training and inference workloads consume electricity at densities that traditional data center design was not built to handle. A standard server rack might draw five to ten kilowatts. Dense GPU configurations can require fifty kilowatts per rack or more, and the upper bound keeps rising as chip generations advance. The power grid serving any given data center region has finite capacity. Adding major new capacity requires transmission infrastructure upgrades, new generation capacity, and in many cases regulatory approvals for large-scale power purchase agreements. These processes take years, not months, and they are running in parallel in every major data center region in the world simultaneously.
The geographic concentration of cloud infrastructure makes this harder. Northern Virginia is one of the largest data center markets in the world and has faced documented power availability constraints that have forced providers to look at adjacent regions for new capacity. Ireland hosts a disproportionate share of European cloud infrastructure relative to its grid capacity, and Irish regulators have been publicly concerned about whether continued data center growth is compatible with national electricity security. Singapore has had periods where data center construction permits were paused for grid capacity reasons. These are not hypothetical future risks. They are constraints that providers are managing today, and they are going to intensify as AI workload demand pushes power requirements higher.
Water is a related constraint that gets less attention. Data center cooling, particularly evaporative cooling in large-scale facilities, requires significant water consumption. In regions under water stress, parts of the US Southwest, the Middle East, parts of southern Europe, data center water use has become a point of public concern and in some cases regulatory pressure. Some providers are investing in more water-efficient cooling approaches, including air-side economization and direct liquid cooling at the rack level. But the transition takes time and the aggregate water requirement for global data center operations remains material.
I want to be careful not to frame this as a catastrophist argument. The hyperscalers are not naive about these constraints. They are making very large bets on energy infrastructure, including direct investment in nuclear power restart and new renewable energy procurement, precisely because they understand that power availability is a binding constraint on their growth. Microsoft, Google, and Amazon have all made announcements in the past year about power purchase agreements and energy investments at a scale that was unusual even by the standards of capital-intensive industries. These investments suggest the constraints are being taken seriously. They do not eliminate the constraints.
The market concentration dimension of the $1.48 trillion trajectory is a different kind of concern. The large majority of public cloud revenue flows through three providers: AWS, Microsoft Azure, and Google Cloud. This means the global cloud trajectory is highly dependent on the operational execution and capital allocation decisions of a small number of firms. If any of the three encounters a significant operational setback, a major outage pattern, a regulatory action that constrains their operations in key markets, or a financial constraint that limits capital expenditure, the effect on the overall trajectory is substantial. This is not a hypothetical risk. AWS had a major outage in December 2021 that affected a significant portion of the US internet. The concentration that makes cloud efficient also makes the trajectory fragile in ways that a more distributed market would not be.
From an IS perspective, the framework that I keep returning to when I look at this is the Melville et al. (2004) IT value model, which traces the path from IT capability to business process capability to organizational performance. The cloud trajectory assumes that organizations will continue to invest in cloud because cloud investment generates business value. That assumption is reasonable in aggregate. But Brynjolfsson's productivity paradox is a reminder that the aggregate relationship between IT investment and value creation is not automatic or immediate. Investment cycles can run ahead of the organizational capability to capture value from them. If cloud spending doubles by 2029 but organizational capabilities for managing, governing, and extracting value from cloud infrastructure do not keep pace, the paradox reasserts itself: more spending, not proportionally more value.
The number that grounds this for me is from the Flexera 2025 State of the Cloud report (https://www.flexera.com/blog/cloud/cloud-computing-trends-tech-spend-pulse/): 27% of cloud spend is wasted, 84% of organizations struggle to manage their cloud costs, and budgets are overrun by an average of 17%. These are the conditions in a $723 billion cloud market. The trajectory to $1.48 trillion does not automatically come with better governance. The organizations that will be best positioned in a doubled cloud market are the ones building genuine governance capability now, not the ones assuming the governance problem will sort itself out as the market matures.
My honest research question from all of this is: when we look back at the 2025-2029 period in ten years, will we see it as the era when AI-driven cloud investment delivered the productivity gains the market assumed? Or will we see it as another installment in the Brynjolfsson paradox, large IT investment followed by a delayed and painful realization that value capture required organizational change that was not happening at the same rate as the spending? I do not know the answer. I think that uncertainty is underpriced in most of the forecasting I read.
---
claims_checked:
- "Total IT spending $6.31T in 2026, 13.5% growth (Gartner April 2026)": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "Public cloud $723B in 2025, $1.48T by 2029 (Gartner April 2026)": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "AI spending $2.5T in 2026 (Gartner April 2026)": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "Agentic AI in 40% of enterprise apps by end 2026 (Gartner April 2026)": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "Cloud waste 27%, 84% struggle, 17% over budget (Flexera 2025)": "https://www.flexera.com/blog/cloud/cloud-computing-trends-tech-spend-pulse/"
claims_unverified:
- "5-10 kW per standard rack, 50+ kW for GPU clusters: widely reported engineering figures, approximate"
- "Northern Virginia, Ireland, Singapore data center power constraints: widely reported in technology press, not linked to specific sources"
- "Microsoft, Google, Amazon nuclear and renewable energy announcements: widely reported, not linked to specific press releases"
- "AWS December 2021 outage details: widely documented, not linked to a specific incident report"
sources_used:
- "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "https://www.flexera.com/blog/cloud/cloud-computing-trends-tech-spend-pulse/"
word_count: 1150
About the author
Share
More notes
Related notes