Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsNeither AI infrastructure nor AI software has inherently more durable growth. Infrastructure can turn demand for scarce computing capacity into usage revenue and customer commitments, but it must earn returns on costly, depreciating assets. Software can monetize established workflows and customer relationships, but AI features need to attract paying users and cover their delivery costs. The better test is whether each business can retain customers, price its value, and generate cash returns after the full cost of serving them.
What counts as AI infrastructure and AI software?
AI infrastructure includes the computing hardware, data-center capacity, networking, cloud services, and platforms that customers use to train or run AI systems. Revenue may come from hardware sales, rented capacity, usage charges, or longer-term customer commitments. AI software includes applications and features that use AI in a customer-facing product or workflow; revenue may come through subscriptions, per-seat prices, consumption charges, or embedded features.
As an Amazon Associate I earn from qualifying purchases.
The boundary is not clean. A provider may sell both compute and applications, and companies report segments that combine offerings from different layers. A segment’s growth rate therefore cannot automatically be treated as the growth rate of “infrastructure” or “software.” Microsoft Cloud, for example, includes both Azure and Microsoft 365 Commercial cloud; Alphabet describes cloud offerings across infrastructure, platform services, and applications; Alibaba reports cloud AI products and model services.
How the business models differ
| Question | AI infrastructure | AI software |
|---|---|---|
| How it earns revenue | Compute, cloud capacity, networking, platforms, and sometimes hardware sales; charges may be usage-based or contracted. | Subscriptions, per-seat pricing, consumption charges, embedded AI features, or application revenue. |
| What can support durability | Customer commitments, high and sustained utilization, renewals and expansions, revenue per unit, manageable customer concentration, and capacity lead times. | Paid adoption, renewals and retention, account expansion, revenue per customer, workflow integration, and pricing power. |
| Where costs can bite | Capital spending, depreciation, energy, equipment, networking, and the risk that capacity is underused. | Inference and hosting costs can pressure gross margins even when the provider owns fewer physical assets. |
| Key growth failure mode | Capacity arrives ahead of demand, customers are concentrated, equipment ages, or operating costs rise faster than monetization. | Features fail to convert into paid use, churn rises, competition weakens pricing, or AI undermines the monetization of a legacy product. |
| Useful economic test | Incremental cash returns and returns on invested capital after infrastructure costs. | Retention and expansion alongside contribution margin after compute and service costs. |
This is a practical comparison framework, not a standardized industry score: company filings do not define one common measure of “durability.” Revenue growth by itself is not enough. The relevant question is how much lasting customer value and cash generation remains after the costs required to deliver it.
What company results show—and what they do not
The reported figures below illustrate different business exposures. They cover different periods, segments, and company definitions, so they are not like-for-like measures of infrastructure versus software growth.
| Company and period | Reported result | What it says—and does not say |
|---|---|---|
| Alphabet, 2025 | $91.4 billion in capital expenditures. In its FY2025 Form 10-K, Alphabet said technical infrastructure investment was expected to increase significantly in 2026. | Shows the scale of one company’s infrastructure investment and its stated direction for the next year; it does not establish the return those investments will earn. |
| Microsoft, FY2025 | Microsoft 365 Commercial cloud revenue grew 15%. Microsoft attributed growth partly to installed-base expansion and average revenue per user. | Illustrates how an established subscription and customer base can contribute to growth. It is a Microsoft result, not a sector-wide software growth rate. |
| Microsoft, FY2025 | Azure and other cloud services revenue grew 34%. Microsoft Cloud gross margin percentage declined slightly, partly because of scaling AI infrastructure. | Shows why revenue growth and margin trends need to be read together. Microsoft Cloud is a broader reporting category, not a pure infrastructure business. |
| NVIDIA, FY2026 | Data Center revenue was $194 billion, up 68% year over year, according to NVIDIA’s SEC-filed annual report. | A company segment figure for Data Center; it is not an estimate of all AI infrastructure revenue or proof of industry-wide durability. |
| Microsoft, FY2024–FY2026 | Microsoft Cloud revenue was $137.7 billion in FY2024, $168.9 billion in FY2025, and $214.4 billion in FY2026. | The series combines cloud and software offerings. It cannot be labeled a pure infrastructure or software revenue series. |
| Alibaba, quarter reported for March 2026 | Alibaba Cloud Intelligence Group external revenue grew 40% year over year; the company said AI-related product revenue was 30% of Cloud external revenue for that quarter. | These are Alibaba-reported results and definitions. They indicate growth and mix for that business and quarter, not sector-wide profitability. |
Alphabet also warns that AI products may monetize differently from historical offerings, with possible changes to revenue mix and margin trends. It says that when developing new products and services it generally focuses first on user experience and then on monetization. That is the company’s stated approach, not evidence that monetization will ultimately match prior products.
Rank #2
Can infrastructure growth last after the buildout?
It can, if demand keeps pace with new capacity and the revenue from that capacity covers the full cost of building and operating it. Customer commitments can improve visibility, but a commitment is not the same as realized usage, profitable utilization, or a completed cash return.
Amazon CEO Andy Jassy said in the company’s 2025 shareholder letter that a substantial portion of expected AWS 2026 capital expenditure already had customer commitments. He also described short-term free-cash-flow headwinds, saying: “We are willing to make large capex investments and endure short-term FCF headwinds for the substantial medium to long-term FCF surplus.” This is management’s view of the investment trade-off, not proof that the expected surplus has been realized.
For an infrastructure provider, the key post-buildout question is not simply whether AI demand remains high. It is whether each additional unit of capacity can be sold and operated at attractive economics, given depreciation, energy, equipment, network costs, and the possibility that hardware ages before it is fully utilized. Alphabet has said costs in areas including depreciation, energy, equipment, and network capacity are expected to rise as AI requires more compute.
What makes AI software growth durable?
Software can build on an installed customer base, an existing workflow, and recurring subscription relationships. Those advantages matter only if customers adopt AI features, keep paying for them, and see enough value to renew or expand. A feature that increases usage but does not support paid adoption or retention may add costs without creating durable revenue.
Software economics also depend on the cost of inference and service delivery. Owning fewer physical assets does not make AI software cost-free: providers may pay for compute directly or through cloud suppliers. A useful measure is contribution after these costs, paired with renewal, retention, and expansion—not just the number of AI features launched or the revenue attributed to a broad cloud segment.
A practical way to judge durability
- Define the revenue being measured. Check whether the reported segment is infrastructure, software, or a mix. Note whether it includes hardware, cloud platforms, applications, or non-AI offerings.
- Trace growth to customer behavior. Look for paid adoption, usage, renewals, account expansion, and customer concentration. For infrastructure, distinguish commitments from delivered utilization and collected revenue.
- Pair revenue with the costs that produce it. For infrastructure, include capital expenditure and ongoing operating costs. For software, account for inference, hosting, and customer-support costs.
- Look for cash economics over time. Ask whether incremental investment is producing cash returns after the necessary costs, rather than assuming that fast growth guarantees attractive returns.
- Test the downside case. Consider idle or aging infrastructure, energy and capacity costs, weak AI-feature conversion, churn, and competition that limits pricing.
These checks help compare actual business economics without pretending that companies disclose a single, directly comparable durability metric.
Why the winning layer may be hard to separate
Infrastructure and software reinforce one another: applications need compute, while useful applications can create demand for more compute. The same provider may sell services at both layers, and its results may combine them. A rise in cloud revenue can therefore reflect different mixes of capacity, platforms, and software rather than a clean victory for one business model.
There is no comparable, independently defined industry-wide statistic in the available company examples that establishes which layer has more durable growth. Company results can show how a particular business is performing; they do not settle the broader comparison.
Quick Recap
How to read claims about AI growth
- Fast revenue growth: useful evidence of demand, but not by itself of profitability, retention, or cash returns.
- Customer commitments: evidence of contracted demand as described by the company, but not a guarantee of realized usage or investment returns.
- AI revenue share: informative only with the company’s definition, reporting period, and denominator.
- Broad cloud growth: not a pure proxy for AI infrastructure when the segment also includes software, platforms, or other offerings.
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.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →
