Neither AI infrastructure nor AI software has inherently more durable growth. Infrastructure providers can benefit from scarce compute, customer commitments and rising usage, but must finance capacity and cover depreciation, energy and equipment costs. Software companies can build on subscriptions and established workflows, but growth lasts only if AI features attract paying users, improve retention and earn more than they cost to deliver. The better test is whether growth produces durable customer value and attractive cash returns after the full cost of providing it.
What makes AI growth durable?
Durable growth is more than a fast-growing revenue line. It means customers continue to pay, renew or expand their use, while the company can serve them profitably and reinvest without returns deteriorating. For AI businesses, that means examining both demand and the costs required to meet it.
As an Amazon Associate I earn from qualifying purchases.
Infrastructure and software are also not cleanly separated categories. A cloud provider may sell computing capacity, developer platforms and finished applications; a software company may rely on rented cloud infrastructure and incur substantial inference costs. Large providers can operate across several layers, so a reported segment’s growth rate is not automatically evidence for one business model over the other.
How do the two models make money, and what should readers compare?
| Measure | AI infrastructure | AI software |
|---|---|---|
| Revenue mechanism | Compute, cloud capacity, networking, platforms and sometimes hardware; revenue may be usage-based or contracted. | Subscriptions, per-seat charges, consumption pricing, embedded features or application revenue. |
| Evidence of durable demand | Customer commitments, utilization, renewals and expansion, revenue per unit, customer concentration and capacity lead times. | Paid adoption, renewal and retention, account expansion, revenue per customer, workflow integration and pricing power. |
| Costs that can weaken growth quality | Capital spending, depreciation, energy, equipment, network costs and the risk that capacity is underused. | Inference and hosting costs, which can pressure gross margins even where direct asset ownership is lower. |
| Key growth risk | Capacity arrives ahead of demand, customers are concentrated, hardware ages, or costs outpace monetization. | Features do not convert to paid use, churn rises, competition weakens pricing, or AI undermines monetization of a legacy product. |
| Useful outcome measure | Incremental cash returns and returns on invested capital after the full cost of infrastructure. | Retention and expansion alongside contribution margin after compute and service costs. |
This is a practical comparison framework, not a standardized durability score defined by company filings. Revenue growth alone is insufficient: a business needs to show that customers value what it sells and that serving them can generate worthwhile returns.
What can company results tell us—and what can’t they?
The reported figures below illustrate different parts of the AI economy, but they cover different periods and segment definitions. They are company-reported results, not directly comparable measures of industry-wide infrastructure or software growth.
| Company and period | Reported figure | What it does—and does not—show |
|---|---|---|
| Microsoft, FY2025 | Azure and other cloud services revenue grew 34%. | Microsoft Cloud gross margin percentage declined slightly, partly due to scaling AI infrastructure. The revenue increase therefore does not by itself demonstrate improving margins or cash returns. |
| Microsoft, FY2025 | Microsoft 365 Commercial cloud revenue grew 15%. | Microsoft attributed growth partly to installed-base expansion and average revenue per user. This is one company’s example of growth tied to subscriptions and an established customer base, not a sector-wide software growth rate. |
| Microsoft, FY2024–FY2026 | Microsoft Cloud revenue was $137.7 billion in FY2024, $168.9 billion in FY2025 and $214.4 billion in FY2026. | Microsoft Cloud combines cloud and software offerings. This series cannot be treated as a pure infrastructure or pure software measure. |
| Alphabet, 2025 and 2026 outlook | Alphabet reported $91.4 billion in capital expenditures in 2025 and expected technical infrastructure investment to increase significantly in 2026. | Alphabet also said costs including depreciation, energy, equipment and network capacity were expected to rise as AI requires more compute. Spending and an investment outlook indicate the scale of the buildout, not whether the resulting assets will earn attractive returns. |
| NVIDIA, FY2026 | Data Center revenue was $194 billion, up 68% year over year. | This is NVIDIA’s reported segment figure, not an estimate of all AI infrastructure revenue and not directly comparable with cloud-provider revenue. |
| Alibaba Group, reported March 2026 quarter | Cloud Intelligence Group external revenue grew 40% year over year; AI-related product revenue was 30% of Cloud external revenue for that quarter. | These are Alibaba’s reported results and definitions. Rapid AI product and model-service growth does not establish sector-wide profitability. |
Why infrastructure growth can last—and where it can fail
Infrastructure providers can capture demand from customers that need computing capacity without building and operating all of it themselves. Usage-based sales can rise as workloads expand, while customer commitments may give providers greater visibility into demand. Those signals matter, but they do not remove the timing and return risks of a large buildout: capacity must be installed before it can serve workloads, and its economics depend on eventual use and the costs of keeping it available.
Rank #2
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, while also describing short-term free-cash-flow headwinds. He characterized the investment as a willingness to accept near-term cash pressure for a “substantial medium to long-term FCF surplus.” That is management’s view of expected demand and returns, not evidence that the surplus has already been realized.
Free tools Windows power users keep installed
One-click scans. No signup required.
Alphabet’s 2025 capital expenditures and 2026 investment outlook illustrate why revenue and cost need to be read together. More compute can enable more sales, but depreciation, energy, equipment and network capacity also weigh on the economics. A durable infrastructure case therefore depends on whether capacity is utilized and customer revenue covers the full cost of building and operating it—not simply on whether demand is growing.
Why software growth can last—and where it can fail
Software businesses may have an advantage when AI can be added to products customers already use, sold through existing subscriptions or embedded in familiar workflows. Microsoft says Microsoft 365 Commercial cloud growth depends partly on installed-base expansion and average revenue per user; its 15% FY2025 growth is an example of those drivers in a specific business, not proof that AI software as a whole will grow at that rate.
But an AI feature is not durable growth just because it launches or attracts trial use. It must translate into paying adoption, renewal or expansion, and the additional revenue needs to exceed inference, hosting and support costs. Companies also face the possibility that AI features fail to justify higher prices, intensify competition or weaken the economics of existing products.
AI may also change how software is monetized. Alphabet’s FY2025 Form 10-K says, “When developing new products and services we generally focus first on user experience and then on monetization.” Alphabet warns that AI products may monetize differently from historical offerings and that revenue mix and margin trends may change. Its statement describes an approach and an uncertainty; it does not establish that current AI offerings will achieve a particular level of profitability.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Can infrastructure companies sustain growth after the buildout?
That depends on what demand looks like when newly installed capacity is available and how much it costs to keep that capacity productive. Commitments and rising usage can support a durable case, but investors and readers should distinguish planned or committed demand from delivered revenue, and revenue from cash returns after capital costs. Hardware replacement cycles, customer concentration, utilization and energy costs all affect the outcome.
The same caution applies when evaluating company growth rates across layers. Microsoft Cloud 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. These overlapping portfolios mean a segment’s growth is not a clean proxy for either infrastructure or software.
How to judge which model is stronger in a specific company
Rather than choosing a universal winner, assess the mechanism that is supposed to turn AI demand into lasting cash generation:
- Identify what the company actually sells. Separate infrastructure, platforms and applications where disclosures allow; do not relabel a mixed segment as one layer.
- Test demand quality. Look for paid adoption, repeat usage, renewals and expansion. For infrastructure, distinguish customer commitments from realized usage and revenue.
- Follow the full cost of growth. For infrastructure, include investment, depreciation, energy, equipment and network costs. For software, include inference, hosting and service costs.
- Check whether economics improve as revenue grows. A rising sales figure alongside margin pressure may reflect a necessary investment phase, but it is not yet proof of attractive long-term returns.
- Consider the risks specific to the model. Underused capacity and concentrated customers matter for infrastructure; weak conversion to paid use, churn and pricing pressure matter for software.
- Judge cash returns, not a narrative. The useful question is whether incremental investment can earn worthwhile returns over time, not whether the company belongs to the infrastructure or software label.
So, which business model has more durable growth?
There is no established industry-wide comparison in these examples that proves infrastructure or software has the more durable growth profile. Infrastructure can have strong demand visibility and scale, but its capital and operating costs make utilization and cash returns decisive. Software can build on subscriptions and established workflows, but paid adoption, retention, pricing and delivery costs determine whether AI features improve the business. Because companies often sell across both layers, the most reliable judgment is company-specific: follow customer value, recurring revenue quality, incremental margins and cash returns on investment.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteQuick 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.
