Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Amazon ranks first on the comparable 2025 estimate, followed by Microsoft, Google, and Meta—but that is a ranking of total capital expenditure, not AI-only spending. The ACCC’s December 2025 comparison puts 2025 capex at approximately $125 billion for Amazon, $115 billion for Microsoft, $92 billion for Google, and $72 billion for Meta. The fourth-quarter figures in that table are estimates, and the totals include data centers, servers, networking and other assets used beyond AI.
A defensible comparison therefore needs four adjustments: align calendar and fiscal periods, identify what each company counts as capex, account for lease treatment, and separate management’s AI-attributed spending from total infrastructure investment. Meta’s own 2025 filing reports $72.22 billion of capex, including principal payments on finance leases, and guides to $115 billion–$135 billion in 2026 for AI efforts and the core business. Those figures show Meta accelerating, but they do not establish an AI-only ranking.
What is the current spending ranking?
On the ACCC’s comparable 2025 estimate, Amazon has the largest total capex, followed by Microsoft, Google and Meta. Because the comparison includes estimated fourth-quarter data and measures all capital investment, it should be read as an infrastructure-scale indicator rather than a precise measure of AI budgets.
| Company | 2025 total capex estimate | How to interpret it |
|---|---|---|
| Amazon | $125 billion | ACCC estimate; total capex, not AI-only |
| Microsoft | $115 billion | ACCC estimate; total capex, not AI-only |
| Google (Alphabet) | $92 billion | ACCC estimate; total capex, not AI-only |
| Meta | $72 billion | ACCC estimate; total capex, not AI-only |
The ACCC finalized this point-in-time snapshot in December 2025. Its 2025 table estimates the final quarter, so the values are not equivalent to four independently reported annual totals. Stanford’s AI Index also publishes estimated hyperscaler capex for 2025–2026, but those estimates are useful context, not substitutes for company filings.
#1 Best Overall
What do the companies’ own filings report?
Company filings provide the most authoritative figures, but they do not use one common definition or reporting calendar. Meta and Google report calendar-year results in the cited filings; Microsoft’s annual report covers a fiscal year and discusses AI-related research and infrastructure costs in operating expenses.
Meta: $72.22 billion in 2025 and a much larger 2026 plan
Meta Platforms’ 2025 Form 10-K reports $72.22 billion in capital expenditures. The amount includes principal payments on finance leases, a treatment that matters when comparing it with companies that present purchases of property and equipment without the same lease-inclusive headline.
Meta says it anticipates $115 billion–$135 billion in 2026 “to support our AI efforts and core business.” This is forward guidance, not a completed-year result, and it covers both AI and non-AI requirements. Relative to the 2025 reported amount, the low end would be about 59% higher and the high end about 87% higher if Meta achieves the range; those percentages are simple comparisons, not a forecast of cash spending by quarter.
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 minuteGoogle: $91.4 billion in 2025 after $52.5 billion in 2024
Alphabet’s 2025 Form 10-K reports $91.4 billion in capital expenditures, up from $52.5 billion in 2024. That is an increase of $38.9 billion, or about 74% year over year. Alphabet describes the spending as purchases of property and equipment and says its 2026 investment in technical infrastructure will increase significantly.
Google’s disclosure is therefore a reported annual total followed by qualitative 2026 guidance. It does not provide a single, audited AI-only capex line that can be placed directly beside Meta’s 2026 range.
Microsoft: separate its fiscal-year capex from AI-related R&D
Microsoft’s 2026 annual report uses a different reporting period and says its research-and-development expense includes “AI training and other infrastructure costs.” Fiscal-2026 R&D increased 9% year over year. R&D is an operating expense, not capital expenditure, so that disclosure must be shown separately from the ACCC’s capex ranking.
Counting Microsoft’s AI-related R&D as though it were additional capex would mix two accounting categories. A complete Microsoft analysis needs the company’s fiscal-year capital-investment figure for the chosen period and then a separate line for the AI costs Microsoft places in R&D.
Amazon: the largest total in the ACCC comparison
Amazon’s $125 billion figure is the highest value in the ACCC’s 2025 estimate. The comparison establishes its total-capex position, but the cited material does not provide a uniform Amazon AI-only allocation. Amazon’s total includes infrastructure supporting several businesses, so the number should not be presented as a dedicated AI budget.
Why are these numbers not directly interchangeable?
Four differences can change the apparent order even when the underlying infrastructure activity is similar.
1. Calendar years and fiscal years cover different months
Meta and Google’s cited figures are for the 2025 calendar year. Microsoft’s annual report is for fiscal 2026, which does not cover the same twelve-month window as calendar 2025. A comparison should label every figure with its exact year-end date and avoid placing fiscal-2026 R&D beside calendar-2025 capex without a prominent qualification.
The ACCC improves comparability by putting the companies into one 2025 estimate, but its fourth-quarter values are modeled rather than final reported results. Once all companies publish complete annual filings, replace estimates with actuals and preserve the original period labels.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
2. Capex is broader than AI infrastructure
Capital expenditure generally covers long-lived assets such as data-center construction, servers, networking equipment, storage and facilities. Those assets can support AI training and inference, but they can also run search, advertising, cloud, social, retail and other workloads. The ACCC explicitly warns that its capex figures are not exclusively AI spending.
Rank #3
A total-capex ranking answers “who is investing most in durable infrastructure?” It does not answer “who bought the most AI accelerators?” or “who spends the most on model training?” Those require company-specific allocations that are not disclosed consistently.
3. Lease treatment changes the headline total
Meta’s $72.22 billion includes principal payments on finance leases. Google’s cited filing describes purchases of property and equipment, while the ACCC harmonizes reported figures for its cross-company table. Lease-inclusive and purchase-only presentations can differ even when a company controls similar capacity.
When building your own spreadsheet, create separate columns for cash purchases of property and equipment, finance-lease principal and any other lease-related additions disclosed by the company. Do not silently add or remove lease amounts; state the treatment in the column heading and notes.
4. Management attribution is uneven
Meta explicitly frames its 2026 $115 billion–$135 billion outlook as supporting “AI efforts and core business,” so the range is intentionally broader than AI. Microsoft identifies AI training and other infrastructure costs within R&D, while the cited Google and Amazon figures do not establish an equivalent AI-only amount. These disclosures cannot be converted into a reliable four-company AI ranking by applying one percentage to every company.
How should you build a fair comparison?
Use the following sequence whenever you update the figures or evaluate a new claim about AI spending.
- Choose one period. Prefer a completed calendar year for all companies, or clearly label a mixed-period view. If one company reports on a fiscal year, show the fiscal year-end date rather than renaming it as a calendar year.
- Start with reported total capex. Record the exact amount from each annual report before looking at AI commentary. Keep the ACCC estimate in a separate column so an estimate cannot be mistaken for a filed actual.
- Document the accounting scope. Note whether the figure is purchases of property and equipment, a broader capital-expenditure definition, or a total that includes finance-lease principal.
- Separate operating expenses. Put Microsoft’s AI-related R&D in an R&D column. Do not add it to capex unless a source explicitly identifies a capitalized amount.
- Mark actuals, estimates and guidance. Meta’s $72.22 billion is a reported 2025 result; Meta’s $115 billion–$135 billion is 2026 guidance; the ACCC’s fourth-quarter values are estimates. These labels belong next to the numbers.
- Only then discuss AI allocation. Use a company’s own attribution when it exists, and write “AI share not disclosed” when it does not. Never infer an AI-only total from the presence of GPUs or data centers alone.
- Add strategy and monetization evidence separately. Infrastructure scale indicates capacity and commitment. It does not by itself demonstrate model quality, user adoption, revenue, margins or return on invested capital.
Is Meta catching up with the hyperscalers?
Meta is closing the gap in planned infrastructure investment, but the evidence supports a cautious answer: its 2026 guidance is large enough to approach or exceed the 2025 totals of the other companies, yet it is not an AI-only commitment and it is not directly comparable with a completed year.
Rank #4
Meta’s 2025 capex of $72.22 billion was below the ACCC’s 2025 estimates for Amazon, Microsoft and Google. Its 2026 range of $115 billion–$135 billion is approximately $42.78 billion to $62.78 billion above its 2025 result. The range also spans the ACCC’s 2025 estimates for Microsoft ($115 billion) and Google ($92 billion), and reaches into Amazon’s $125 billion estimate. That arithmetic shows the scale of the planned step-up; it does not predict where the companies will finish in the same 2026 period.
There are three reasons not to call this a definitive catch-up:
- Different time bases: Meta’s range is forward-looking for 2026, whereas the ACCC figures are 2025 estimates.
- Mixed purpose: Meta says the range supports AI efforts and the core business, not AI alone.
- Different accounting: Meta’s headline includes finance-lease principal, while other disclosures may emphasize purchases of property and equipment.
The defensible conclusion is that Meta is planning a major acceleration and could become comparable in total infrastructure outlay. Whether it catches up in AI capacity, model development or commercial returns requires additional company-specific operating data.
What does higher capex tell you about AI infrastructure?
Higher capex usually signals that a company is adding long-lived capacity, but the same dollar can support different technical strategies. A company may spend on new data centers, networking to connect accelerators, storage for training datasets, general cloud capacity or replacement equipment. Total capex therefore measures the scale of the build-out, not the exact mix of hardware and software work.
Infrastructure scale
Use total capex to compare the financial size of each company’s build-out after normalizing period and accounting treatment. The ACCC ranking is useful for this high-level question because it applies one table to all four companies, while its estimate caveat remains visible.
Model and product strategy
AI infrastructure can support internally developed models, consumer features, enterprise services or rented capacity for customers. The cited filings do not provide a standardized allocation showing how much of each company’s total is assigned to those uses. Treat strategy claims as a separate qualitative layer rather than converting them into dollars.
Monetization and returns
Capex alone cannot show whether spending is earning an adequate return. A serious assessment should separately examine revenue attribution, utilization, depreciation, operating margins and demand for the resulting services. None of those measures can be substituted with the four-company capex ranking.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What are the most common comparison mistakes?
- Calling total capex “AI spending.” This overstates what the filings establish.
- Mixing guidance with actuals. A 2026 range should not be ranked as though it were a completed 2025 result.
- Ignoring the fourth-quarter estimate. The ACCC’s 2025 table includes estimated Q4 values.
- Combining R&D and capex. Microsoft’s AI training and infrastructure costs are described within R&D, an operating-expense category.
- Hiding lease treatment. Meta’s inclusion of finance-lease principal affects comparability and must be disclosed.
- Ranking companies by an invented AI percentage. The cited filings do not supply a common allocation basis.
- Using a single year to infer a trend. Investment can rise or fall with construction schedules, equipment deliveries and accounting timing.
What should a published comparison table include?
A useful table lets readers see the number, period, accounting basis and evidence status without opening several filings. At minimum, include these fields:
- Company and reporting entity.
- Period end and calendar or fiscal designation.
- Total capex in the company’s reported currency.
- Whether the amount is an actual, an estimate or forward guidance.
- Lease treatment, including whether finance-lease principal is included.
- Any explicit management attribution to AI.
- A separate operating-expense line for AI-related R&D.
- The source document and publication date.
For the figures covered here, the table should show Meta’s reported 2025 $72.22 billion and 2026 $115 billion–$135 billion guidance; Google’s 2025 $91.4 billion and 2024 $52.5 billion; Microsoft’s fiscal-2026 R&D increase of 9% with AI training and other infrastructure included in R&D; and the ACCC’s estimated 2025 totals for all four companies. Keeping those entries distinct prevents a visually neat table from implying a precision the disclosures do not support.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →How often should the comparison be updated?
Update it when a company releases a new annual report, materially changes its capital-spending outlook or reports a quarter that replaces an estimate. Forward guidance can change, and an estimate such as the ACCC’s Q4 2025 value should be replaced by the final filing when available.
Keep prior versions in an archive rather than overwriting them. That preserves the difference between what management planned, what regulators estimated and what the company ultimately reported—an essential distinction when evaluating whether an announced AI build-out actually occurred.
Frequently Asked Questions
Why can’t AI-related R&D be added to a company’s capex total?
R&D is recorded as an operating expense when incurred, while capex is generally capitalized and depreciated over the useful life of an asset. Adding the two produces a blended metric that is neither the company’s reported capex nor its reported R&D.
Does Meta’s lease-inclusive figure mean it paid more cash than the other companies?
Not necessarily. Including finance-lease principal changes the presentation of capital investment, but it does not by itself reveal the economic cost, timing or financing structure of another company’s facilities and equipment. Compare the underlying lease and purchase disclosures before drawing a cash-spending conclusion.
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 & 11When will the 2025 comparison become more reliable?
It becomes more reliable after all four companies publish complete annual results for the same period and the estimated fourth-quarter values are replaced with reported figures. Even then, the AI-only portions may remain undisclosed or defined differently.
Can capex ranking predict which company will have the best AI products?
No. Product performance depends on models, software, data, talent, distribution and user adoption as well as hardware capacity. Capex is an input measure; it is not a product-quality or return-on-investment score.
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.
