On July 17, 2026, Apple surpassed NVIDIA in market capitalization, reclaiming its position as the world's most valuable publicly traded company. According to Reuters, Apple's market cap stood at $4.88 trillion at that point, versus NVIDIA's $4.86 trillion. Toni Meadows of BRI Wealth Management told Reuters that Apple can more easily tie AI to its services and device upgrade cycles, giving it a lighter burden than companies that must continue pouring massive sums into capital expenditure.
The reversal in market cap reflects a single moment of market valuation, and it would be a stretch to attribute it solely to Apple's low-CapEx strategy. Still, a look at Apple's recent earnings and the architecture behind Siri AI lends some credibility to the view that the company can embed AI into its existing businesses while keeping spending on its own facilities in check. Apple uses on-device Apple silicon together with a dedicated cloud, funnels money into R&D, and hands off a portion of manufacturing infrastructure to suppliers. The open question is whether this design can hold up once Siri AI sees wider use.
The Investment Gap Between Apple and Microsoft
In the first half of fiscal year 2026, Apple's cash used for purchases of property, plant, and equipment totaled $4.344 billion. That amounts to roughly 1.7% of the period's revenue of $254.94 billion, and it's down from $6.011 billion in the same period a year earlier.
Microsoft, by contrast, spent $31.9 billion on capital expenditures in just a single quarter—Q3 FY2026 (ended March 31, 2026)—with cash spending on property and equipment reaching $30.9 billion in that quarter alone. The company projects roughly $190 billion in capital expenditures for calendar year 2026, and in its most recent quarter, about two-thirds of that spending went toward relatively short-lived assets such as GPUs and CPUs. Microsoft's single-quarter cash outlay exceeds Apple's half-year figure by more than sevenfold.
| Company / Period | Revenue | CapEx-related disclosure | Simple ratio to revenue |
|---|---|---|---|
| Apple, H1 FY2026 | $254.94 billion | Property, plant & equipment purchases: $4.344 billion | ~1.7% |
| Microsoft, Q3 FY2026 | $82.9 billion | Capital expenditures: $31.9 billion | ~38.5% |
That said, these figures aren't measured on the same scale. Microsoft's capital expenditures include finance leases, and the two companies differ in both reporting periods and business models. Because Microsoft sells computing capacity itself through Azure, it must build data centers ahead of demand. Apple, whose core business is consumer devices and services, cannot be judged on AI performance or investment efficiency based on this ratio alone.
Even so, the difference in how quickly each company accumulates fixed assets is significant. Microsoft shoulders both the time lag between bringing infrastructure online and converting it into cloud revenue, and the burden of cycling through successive generations of GPUs and CPUs. Apple, by keeping that burden light, can instead use devices and operating systems it has already sold as a distribution channel for AI.
Devices as Part of the AI Infrastructure
Siri AI runs the next-generation Apple Foundation Models on both devices and servers within Private Cloud Compute (PCC). Requests that can be handled on-device are processed directly on the iPhone or Mac, while more complex tasks are routed to the dedicated cloud. Compared with services that funnel all inference into data centers, this architecture shares the computational load with the Apple silicon that users already own.
This division of labor links Apple's device sales directly to its AI infrastructure. Compatible devices include the iPhone 16 and later, as well as the iPhone 15 Pro, among others—meaning not all existing devices are supported. If users who want AI features upgrade from older devices, Apple both deploys new inference capability on the user's side and captures hardware revenue in the process. That said, Apple has not yet disclosed how much Siri AI has driven device upgrades.
There's also a monetization path built on the server side through existing services. Some Apple Intelligence features that rely on powerful server-side models—such as image generation—carry daily usage caps, and many iCloud+ plans come with expanded usage allowances. This isn't an announcement that Siri AI as a whole will become paid, but the mechanism for splitting free and paid tiers as inference demand grows is already in place.
AI Costs That Fall Outside Capital Expenditure
The fact that purchases of property, plant, and equipment came to only $4.344 billion doesn't mean Apple's technology investment has shrunk. R&D spending in H1 FY2026 reached $22.306 billion, up 33% year over year. Apple has cited infrastructure-related costs and personnel-related costs as drivers of the increase. R&D spending alone is more than five times the amount spent on property, plant, and equipment purchases.
On the income statement, R&D expenses count as operating expenses that reduce profit, but they don't flow into the cash spent on property, plant, and equipment purchases. Spending can rise for AI model development, OS integration, and computing infrastructure operations without ever showing up in CapEx—so long as no fixed assets are purchased. Apple's apparent lightness isn't about total investment so much as the low proportion of infrastructure it holds as its own fixed assets.
Apple is also expanding its own server capacity. It began producing advanced AI servers in Houston in 2025, and in February 2026 announced that production was outpacing plans. The Apple silicon-equipped servers assembled there support PCC through data centers located in the United States. A low-CapEx model doesn't mean cloud infrastructure is unnecessary.
Some of this investment also shows up on the supply-chain side. On July 8, Apple signed a multi-year agreement with Broadcom worth over $30 billion to produce more than 15 billion chips in the United States. In response, Broadcom is investing $1.5 billion in facilities in Fort Collins, Colorado. While the agreement isn't AI-specific, it illustrates well how Apple's capital allocation works: Apple commits to long-term purchasing, while the manufacturing facilities themselves remain assets owned by its supplier.
In this case, the factory equipment belongs to Broadcom, and the $1.5 billion doesn't count as Apple's capital expenditure. What shows up on Apple's side is the multi-year purchase commitment exceeding $30 billion—a way of securing supply capacity through long-term contracts while keeping its own CapEx light.
As of March 28, Apple held $27.691 billion in unconditional purchase obligations extending beyond one year. These break down into supplier agreements, distribution rights, and licenses for intellectual property and content, though Apple hasn't disclosed how much of this is AI-related. At the very least, one cannot conclude from Apple's capital expenditure figures alone that its future spending will remain equally light.
Can AI Be Connected to a 76.6% Gross Margin?
In H1 FY2026, Apple's Services revenue reached $60.989 billion, up 15% year over year. Its gross margin hit 76.6%, far exceeding the 39.9% margin for Products. Expanded iCloud+ usage allowances offer a concrete pathway for loading additional AI computation onto this highly profitable business.
That said, the current growth in Services is driven mainly by advertising, the App Store, and cloud services, and Apple has not broken out revenue attributable to AI. Siri AI itself only entered developer testing on June 8, with a general-audience English beta planned for later in 2026. It won't be offered in China, and in the EU, rollout to the iPhone, iPad, and Apple Watch will initially be held back.
In other words, what underlies Toni Meadows's assessment isn't proven AI revenue, but rather a design capable of monetization while keeping capital expenditure restrained. How much of the cloud load will on-device processing absorb? How much will increased PCC usage push up R&D spending and capital expenditure? Once the Siri AI beta launches, watching how Services revenue and infrastructure-related costs move in tandem will be the real test of the low-CapEx hypothesis.