Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteThere is no defensible single ranking of AI infrastructure spending across cloud providers, chipmakers and model companies. The available company disclosures measure different things: cloud companies report large capital-expenditure programs that mix AI with ordinary cloud and corporate infrastructure; chipmakers report sales to those buyers rather than the cost of the data centers they use; and model companies may rent capacity, sign commitments or build facilities under accounting treatments that are not disclosed on a common basis. The most comparable current figures are the cloud providers’ own outlooks, and even those use different scopes.
What “AI infrastructure spending” actually includes
AI infrastructure is a stack, not a single product. A provider may spend on land, buildings, electrical systems, cooling, servers, storage, network equipment, general-purpose CPUs, AI accelerators and software that manages the fleet. Some of that capacity serves model training and inference; the same equipment can also run databases, search, productivity applications and other cloud workloads.
As an Amazon Associate I earn from qualifying purchases.
That is why a company’s total capex should not automatically be relabeled “AI capex.” Alphabet describes technical-infrastructure spending as including servers and network equipment, data-center land, construction and improvements. Meta says its property-and-equipment purchases primarily funded servers, data centers and network infrastructure while supporting both AI work and its core business. Microsoft discusses AI capacity alongside other assets and priorities. Amazon’s approximately $200 billion figure is the company-wide capex expectation, even though its shareholder letter focuses heavily on AWS capacity.
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Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →The categories also sit at different points in the supply chain:
- Cloud platforms finance, own or lease facilities and equipment, then sell computing capacity and services.
- Chipmakers design and sell accelerators, CPUs, networking silicon and related components to cloud operators, server makers and other customers.
- Model companies train and serve models using owned equipment, rented cloud capacity or contractual capacity reservations. Their payments can be operating expenses, capitalized assets, leases or commitments, depending on the arrangement.
Adding all three categories together can double-count one physical purchase. A cloud provider’s accelerator purchase can appear as the chipmaker’s revenue, while a model company may later pay the cloud provider to use that same accelerator.
The clearest current figures: major cloud providers
The table below keeps each number’s period and scope attached to it. Alphabet’s figure is a reported 2025 result; Microsoft, Meta and Amazon figures are 2026 management outlooks disclosed in 2026. None is an AI-only total.
| Company | Period and status | Disclosed figure | What the company says it covers | Important qualification |
|---|---|---|---|---|
| Alphabet | 2025 actual; 2026 outlook | $91.4 billion capex in 2025; a significant 2026 increase expected | Technical infrastructure such as servers, network equipment, data-center land, construction and improvements | The 2026 filing passage gives no comparable dollar forecast |
| Microsoft | Calendar year 2026 forecast | Approximately $190 billion capex, including about $25 billion attributed to higher component prices | Company-wide capital expenditure supporting cloud and other priorities | Not identified as AI-only; the $25 billion is a price effect within the outlook |
| Meta | 2026 forecast | Approximately $115 billion to $135 billion capex | Infrastructure supporting AI efforts and the core business | The later Form 10-K is the current guidance; accounting labels should not be mixed with an earlier 2025 outlook |
| Amazon | 2026 management forecast | Approximately $200 billion capex | Total company capex, with substantial discussion of AWS capacity | A shareholder-letter forecast, not a separately reported AI or AWS-only amount |
Sources: Alphabet’s 2025 Form 10-K, Microsoft FY2026 Q3 earnings materials, Meta’s 2025 Form 10-K and Amazon CEO Andy Jassy’s 2025 shareholder letter.
Alphabet: a reported base year and a qualitative 2026 step-up
What is disclosed
Alphabet reported $91.4 billion of capex in 2025. Its 2025 Form 10-K says that in 2026 it expects to “significantly increase” investment in technical infrastructure relative to 2025, including servers, network equipment and data centers. The cited disclosure does not provide a 2026 dollar figure, so it cannot be placed in a precise numerical ranking with the other outlooks.
Why the scope matters
Alphabet’s definition is broader than accelerator purchases. Data-center land, construction and improvements can be required before a server is installed, and network equipment is essential to both AI clusters and conventional cloud services. The company also describes a mix of custom tensor-processing units (TPUs) and GPU-based infrastructure. A TPU purchase is specialized compute, while the surrounding facility, power and network systems are shared assets.
Operating consequences
Alphabet expects infrastructure costs to rise as AI services require more compute. The company specifically points to depreciation, energy, equipment and network capacity. Those expenses can increase before a new service reaches scale, so a capex increase should not be read as an immediate increase in profit.
Microsoft: approximately $190 billion for calendar 2026
What is disclosed
Microsoft’s FY2026 Q3 conference-call materials forecast roughly $190 billion of calendar-year 2026 capex. The materials say approximately $25 billion of that reflects higher component prices. This is a calendar-year outlook, not a 2025 actual and not an AI-only budget.
Free tools Windows power users keep installed
One-click scans. No signup required.
Rank #2
Capacity and the hardware fleet
Chief financial officer Amy Hood said, “Broad and growing customer demand continues to exceed supply.” The same passage says Microsoft expects GPU, CPU and storage capacity constraints to continue through 2026. That is management’s description of demand and availability, not an independent verification of future returns.
The fleet uses NVIDIA and AMD hardware alongside Microsoft’s first-party networking silicon, CPUs and accelerators. This illustrates why “AI infrastructure” cannot be reduced to one accelerator line item: a cluster depends on host CPUs, high-speed networking, storage, power delivery and cooling as well as the training or inference chips.
How to interpret the $25 billion component effect
The component-price statement is a qualification on the forecast. It does not mean Microsoft will spend $25 billion on additional AI capability, nor that all of the $190 billion represents more physical capacity. Higher prices can raise the dollar value of the same type of equipment. The full outlook also covers assets and priorities outside AI.
Meta: a $115 billion–$135 billion 2026 range
Current guidance
Meta’s 2025 Form 10-K anticipates approximately $115 billion to $135 billion of 2026 capex. Meta says the program supports its AI efforts and its core business. Its filing describes 2025 property-and-equipment purchases as primarily funding servers, data centers and network infrastructure.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Do not mix the old and new accounting views
Meta’s Q2 2025 results release had projected 2025 capex of $66 billion to $72 billion, including principal payments on finance leases. The later 2025 Form 10-K reports 2025 property-and-equipment purchases and supplies the current 2026 outlook. Those labels and periods are not interchangeable. A finance-lease principal payment can appear in a cash-capex presentation even when the underlying equipment was obtained through a lease rather than a cash purchase.
Costs beyond the initial build
Meta identifies accelerating depreciation and higher operating costs as infrastructure expands. Depreciation spreads the cost of servers and facilities over their useful lives; energy, maintenance and network expenses continue after the initial capex is recorded. The 2026 range therefore represents a multi-year cost commitment, not just a one-time check for GPUs.
Amazon: approximately $200 billion, with AWS commitments and a cash-flow lag
The headline figure and its scope
Amazon CEO Andy Jassy wrote that the company expected approximately $200 billion of capex in 2026. The number is Amazon’s total-company expectation. The letter discusses AWS capacity extensively, but it does not turn the entire $200 billion into an AWS-only or AI-only figure.
Jassy wrote, “We’re not investing approximately $200 billion in capex in 2026 on a hunch.” That sentence is management’s explanation of the plan, not independent evidence that the spending will earn a particular return.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteCommitments and monetization timing
The letter says a substantial portion of planned AWS capacity already has customer commitments and that much of the 2026 AWS capex is expected to monetize in 2027–2028. A customer commitment is evidence of contracted or anticipated demand, but it does not guarantee utilization, margins or renewal after the commitment period.
Why free cash flow can weaken first
Amazon explicitly describes near-term free-cash-flow pressure while new capacity is built before revenue catches up. The sequence is economically important: cash leaves the business when land, construction and equipment are purchased; revenue may arrive only after the facility is commissioned and customer workloads are deployed. Depreciation and energy costs then continue during operation.
Why chipmaker spending cannot be ranked beside cloud capex
Chipmakers occupy a different financial position. Their reported revenue reflects sales of accelerators, CPUs, networking components and systems to customers in the supply chain. It is not the cost of all data centers that use those products, and it is not a measure of the customers’ future cloud revenue.
The disclosures in this comparison do not provide a standardized dollar series for chipmaker AI infrastructure spending. Microsoft’s description of NVIDIA and AMD products in its fleet demonstrates supplier participation, while Alphabet’s discussion of TPUs and GPUs demonstrates that cloud operators may also deploy custom silicon. Neither point supplies a comparable chipmaker capex total.
Supplier revenue versus infrastructure ownership
A cloud operator may own the building and servers while a chipmaker owns the design and manufacturing relationship. The chipmaker’s sale can be recognized when hardware is delivered; the cloud operator recognizes depreciation and service revenue over time. Comparing those two amounts without aligning the transaction date, accounting basis and asset life would compare a supplier’s sale with an operator’s long-lived asset.
Custom silicon changes the boundary
First-party accelerators and networking silicon blur the simple “cloud buys chips” picture. A provider can design or commission a chip, pay manufacturing and packaging costs, and deploy it in a fleet. The resulting expense may be embedded in the provider’s equipment capex rather than reported as an external chipmaker purchase. That is another reason a supplier-versus-cloud ranking is not established by the available figures.
Why model-company spending is even harder to standardize
Model companies can obtain compute in several ways: purchase servers, lease equipment, reserve capacity from a cloud provider or pay for usage as an operating expense. A company may also combine these methods across training, fine-tuning and inference. Public disclosures do not present a harmonized dollar comparison for model-company infrastructure spending alongside the cloud figures above.
Owned capacity
When a model company buys servers or builds a facility, the asset and related depreciation may appear on its balance sheet. The purchase can include general compute, storage, networking and accelerators, just as it does for a cloud provider.
Rented or reserved capacity
When the company rents cloud GPUs or signs a capacity reservation, the cash commitment may be recorded over the contract term. The cloud provider, meanwhile, may have already capitalized the physical equipment. Treating both the reservation and the provider’s capex as separate additions to global infrastructure would count one capacity commitment twice.
Commitments are not realized expenditure
A contractual commitment can indicate expected future payments without proving that all equipment has been delivered or all capacity has been consumed. Any comparison should label commitments separately from cash paid and from assets placed in service.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.The accounting details that change an apparently simple ranking
Actuals versus forecasts
Alphabet’s $91.4 billion is a reported 2025 result. Microsoft’s approximately $190 billion, Meta’s $115 billion–$135 billion and Amazon’s approximately $200 billion are 2026 outlooks. Mixing a completed year with forecasts can make a company appear smaller or larger simply because the periods differ.
Fiscal years versus calendar years
Microsoft’s figure is explicitly for calendar 2026, while the other disclosures are tied to company reporting and shareholder communications. A fiscal-year forecast can cover months that do not line up with another company’s calendar year. The reporting period belongs next to every number.
Recommended Free Tools
Leases and component inflation
Finance-lease principal payments may be included in one capex presentation and excluded from another. Microsoft’s approximately $25 billion component-price effect shows a second issue: inflation can increase spending without a proportional increase in installed capacity. A comparison should state whether a figure reflects cash purchases, lease payments, price changes or a combination.
Depreciation and useful life
Capex is paid upfront, while depreciation is recognized over time. Two companies can spend similar amounts but report different current-period operating costs if their equipment enters service at different dates or uses different useful-life assumptions. Energy and network expenses also continue after the purchase.
A practical comparison framework
Use the following sequence when evaluating a new spending announcement:
- Identify the period. Mark the fiscal or calendar year and whether the number is an actual, a budget, a forecast or a contractual commitment.
- Read the scope sentence. Determine whether the figure is total capex, technical infrastructure, an AI-specific line, AWS capacity or a broader company program.
- Separate the infrastructure layer. Ask how much relates to facilities and power, networking, general servers, accelerators, storage or custom silicon. Do not infer a layer split when the company has not supplied one.
- Determine who finances and owns the asset. A cloud provider may own a server that a model company rents; a chipmaker may supply the accelerator without owning the data center.
- Check accounting treatment. Look for finance leases, capitalized construction, component-price effects and whether payments are included in capex or operating expense.
- Label demand evidence. Customer commitments, cloud growth and statements that demand exceeds supply are indicators from management, not proof of long-term returns.
- Map the cash-flow timing. Record when cash is spent, when equipment is placed in service, when depreciation begins and when management expects monetization.
- Test for double counting. Do not add a cloud provider’s equipment purchase to a chipmaker’s corresponding sale or to a model company’s reservation for the same capacity.
What the current evidence does—and does not—prove
What it establishes
- Large cloud providers are planning unusually large 2026 investment programs.
- The programs combine shared infrastructure with specialized AI compute.
- Management teams cite strong demand, customer commitments or capacity constraints.
- Higher depreciation, energy, networking and other operating costs will follow the build-out.
- Cash flow can be pressured before new capacity generates revenue.
What remains unresolved
- There is no harmonized dollar ranking of cloud providers, chipmakers and model companies in the cited disclosures.
- The available figures do not isolate a consistent AI-only amount for each cloud provider.
- Alphabet’s 2026 statement is qualitative rather than a comparable dollar forecast.
- Management demand statements do not independently establish the eventual return on invested capital.
- Adding figures across supply-chain positions would risk counting the same infrastructure more than once.
How to read a future “AI spending” headline
Start with the company’s original filing, earnings material or shareholder letter and locate the definition of the number. If a headline calls total capex “AI spending,” check whether the company itself used that term. If it did not, retain the company’s wording—capex, technical infrastructure, AWS capacity or another stated scope.
Next, compare like with like: actual with actual, forecast with forecast, calendar year with calendar year, and cash purchases with cash purchases. Then ask whether the number includes leases, construction, networking, general-purpose servers or price inflation. A smaller number with a narrow AI definition may represent more specialized compute than a larger company-wide capex total.
Finally, follow the money through the supply chain. A cloud provider’s accelerator order can be a chipmaker’s revenue; a model company’s reserved capacity can be a cloud provider’s future service revenue. The economically meaningful question is not which category has the largest headline number, but which layer is being financed, who owns it, how it will be used and when it is expected to earn back its cost.
Bottom line
Cloud providers are the only category here with directly comparable, current headline figures, and even those are not equivalent: Microsoft forecasts about $190 billion, Meta $115 billion–$135 billion and Amazon about $200 billion for 2026, while Alphabet reports $91.4 billion for 2025 and only a qualitative 2026 increase. Chipmaker sales and model-company compute commitments cannot be placed beside those amounts without changing the accounting basis and risking double counting. The responsible comparison is a layered one that keeps scope, ownership, timing, demand evidence and cash-flow consequences attached to every number.
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.
