AI infrastructure may prove useful even if some companies building it deliver disappointing returns to shareholders. Microsoft, Oracle and Digital Realty offer three different ways to follow the buildout: a broad cloud platform, a fast-growing but capital-intensive cloud infrastructure business, and data-center capacity. Their growth and plans are worth watching, but they do not establish that their shares are attractively valued or likely to outperform.
What past infrastructure booms can—and cannot—tell investors
Railways, canals and telecommunications show why the usefulness of infrastructure and the returns earned by its original investors are separate questions. An Oxford Review of Economic Policy article on infrastructure investment manias notes that physical assets sometimes remained after investors lost capital. A railway can keep carrying passengers or freight even if the investors who financed its construction did poorly.
The telecom boom offers a more specific caution about overbuilding. In a Fall 2003 Economic Quarterly article, the Federal Reserve Bank of Richmond described overinvestment in long-distance fiber. The article says, “By early 2001, it became apparent that massive overinvestment had taken place in the sector, particularly in the area of long-distance fiber-optic cable.” But its authors also rejected the idea that the boom and bust could simply be explained as a bubble. The episode is a warning about investment and financial strain, not proof that every infrastructure boom follows one path.
A 2009 American Economic Review paper by Ľuboš Pástor and Pietro Veronesi found empirical support for its model of stock-price bubbles during technological revolutions in two historical samples: the U.S. railroad period from 1830 to 1861 and the Internet period from 1992 to 2005. That is a finding about a particular model and sample, not a forecast for AI stocks.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →#1 Best Overall
The Bank for International Settlements’ July 14, 2026 working paper, “The AI investment race,” models a current risk: firms competing for a few dominant positions may invest more than is economically efficient. Its calibrated model produces overinvestment of around 1.5 times the efficient level in a conservative baseline, and around three times when demand is less elastic. These are model-dependent estimates, not observed overinvestment or an official prediction of a crash. The paper’s summary puts the broader lesson this way: “Exuberance about new technologies often brings about investment booms.” It also explains that debt and financial links could increase fragility if demand falls short.
Three companies with different kinds of AI infrastructure exposure
These examples sit at different points in the infrastructure chain. Their reported growth rates and capacity figures are not directly comparable: the companies sell different things and report different segments. The figures below describe results or plans, not expected investment returns.
| Company | Where it fits | Recent evidence | What to watch alongside growth |
|---|---|---|---|
| Microsoft | Broad cloud platform offering Azure and other cloud services; not a pure-play AI cloud company. | Microsoft’s FY2026 Q4 investor release reported 43% growth in Azure and other cloud services revenue. | Cloud growth and continuing AI infrastructure investment do not by themselves establish what investors are paying for that growth or what returns the shares may deliver. |
| Oracle | Cloud infrastructure provider investing to expand capacity for large customers, including AI workloads. | Its June 10, 2026 FY2026 results reported Q4 IaaS revenue growth of 93% year over year and total cloud revenue growth of 47%. Remaining performance obligations were $638 billion. | FY2026 free cash flow was negative $23.7 billion as Oracle invested in infrastructure. Oracle expected to raise about $40 billion in FY2027 through debt and equity financing. |
| Digital Realty | Data-center landlord and capacity developer, rather than a seller of cloud compute services. | Its July 23, 2026 Q2 results presentation reported about 3 GW of in-place IT capacity and about 9 GW of future development IT capacity as of June 30, 2026. | The development figure is pipeline capacity, not all operating capacity or contracted revenue. Delivery, customer demand and infrastructure constraints matter. |
Microsoft: a large platform, not a pure AI bet
Microsoft’s reported Azure and other cloud-services growth makes it a way to watch demand for cloud infrastructure within a much broader platform business. Its earnings materials also describe continuing investment in AI infrastructure. The 43% figure is for Azure and other cloud services revenue in FY2026 Q4; it should not be read as the growth rate of a stand-alone AI business or as evidence about the stock’s valuation.
Oracle: strong cloud growth with substantial funding needs
Oracle’s FY2026 Q4 IaaS growth of 93% and total cloud revenue growth of 47% describe different measures. The company also reported $638 billion in remaining performance obligations (RPO). Much of the increase in RPO during Q3 and Q4 was tied to large AI contracts. RPO represents contracted work not yet recognized as revenue; it is not the same as revenue already earned or cash already collected.
Oracle said prepaid and customer-supplied hardware for large AI contracts totaled $75 billion, which it said reduces the capital it must raise for those data centers. That support sits alongside negative $23.7 billion in FY2026 free cash flow as the company invested in cloud infrastructure, and its expectation of raising about $40 billion in FY2027 through debt and equity financing. Investors following Oracle can track whether contracted demand turns into delivered capacity and revenue, and how the buildout is funded.
Digital Realty: operating sites and a much larger development pipeline
Digital Realty’s reported 3 GW of in-place IT capacity and 9 GW of future development IT capacity illustrate the difference between existing capacity and a pipeline. The company also reported record bookings in its 0–1 MW plus interconnection category and an increase in development capacity. Those are relevant indicators of activity, but the capacity planned for development is not automatically built, operating or leased. The figures do not establish how much future revenue will be earned from that pipeline.
How to assess the risks behind the growth
AI-related demand can benefit a range of infrastructure businesses, but growth, contracts and construction plans answer different questions. When following these companies, separate evidence of demand from evidence that projects can be financed and delivered profitably.
- Revenue versus contracted work: A reported revenue increase reflects recognized business in a period. A backlog or RPO figure points to contracted future work, but does not mean all of it has been delivered, recognized as revenue or collected in cash.
- Operating capacity versus planned capacity: Existing data-center capacity is not interchangeable with a development pipeline. New projects still need to be completed and brought into service.
- Investment versus cash generation: Capital spending can support future growth while weighing on current free cash flow. Check how a company expects to fund construction and equipment, including any debt or equity financing plans.
- Execution and customer concentration: Large AI contracts can support a buildout, but reliance on a small number of large customers can make results sensitive to their plans and ability to follow through.
- Supply and infrastructure constraints: Cisco, a networking supplier rather than a cloud operator, reported that AI infrastructure demand from hyperscaler customers accounted for approximately 6% of its FY2026 revenue, compared with less than 2% in FY2025. Cisco also cited memory constraints and costs—an example of how suppliers can benefit from demand while facing limits on equipment availability and expense.
- Project and financing risk beyond the largest platforms: Specialist provider IREN reported FY2026 AI Cloud Services revenue of $128.8 million, up from $16.4 million in FY2025, alongside a data-center pipeline exceeding 5 GW. It also reported a FY2026 net loss of $702.6 million, impairments tied mainly to converting Bitcoin-mining assets, and substantial GPU financing. Applied Digital reported long-term hyperscaler leases for AI data-center capacity, with projects expected to begin operations in 2027 and 2028, as well as a FY2026 net loss attributable to common stockholders of $249.2 million and secured debt issuance to fund development. These examples show why planned growth and leases need to be considered alongside transition, financing and delivery risks.
What “stocks to watch” should mean
Watching a stock is not the same as recommending it. The company releases and presentations cited here provide operating results, contracts, financing plans and capacity figures; they do not establish current share valuations, expected returns or whether a particular stock suits an individual investor.
Best Value
A practical watchlist can track whether demand becomes recognized revenue, whether planned capacity reaches operation, and whether companies can fund that expansion without financial strain. It can also compare the market price of each company with the cash flow and risks an investor believes are justified. No single growth percentage answers those questions, especially when one company sells cloud services, another is expanding infrastructure, and a third develops data-center capacity.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
