Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
To estimate what a technology company is really spending on AI, do not look for a single “AI capex” line. Reconcile investing cash flow with the property-and-equipment note, map servers, networking, storage and data-center construction to AI activity, then add assets still under construction, leases and other commitments. Finally, compare that spending with capacity, customer demand, pricing, depreciation and liquidity.
This method separates cash already spent from management’s forward plans and prevents an all-infrastructure figure from being mistaken for an AI-only number. The examples below use disclosures from Alphabet, Microsoft and Tesla to show how to perform the analysis without inventing precision the filings do not provide.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Financial Accounting | $74.17 | Buy on Amazon |
| 2 |
|
Loose Leaf for Financial and Managerial Accounting | $100.75 | Buy on Amazon |
| 3 |
|
Accounting for Beginners (All-in-One): Everything You Need to Learn Financial & Managerial... | $22.97 | Buy on Amazon |
| 4 |
|
ISE Financial Accounting | $94.00 | Buy on Amazon |
| 5 |
|
Loose Leaf for Financial Accounting | $93.61 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
What does “AI capital spending” actually include?
AI capital spending is usually a subset of a company’s broader technical-infrastructure investment. Most issuers do not report a standalone line for GPUs or AI data centers. Instead, AI-related investment is embedded in servers, network equipment, storage, data-center land, buildings, construction and related infrastructure.
That accounting presentation matters because a company can truthfully say it is investing heavily in AI while reporting only one combined property-and-equipment figure. Your task is to identify which disclosed categories support AI workloads, estimate the portion that can be supported by the evidence, and label the remainder as broader infrastructure rather than silently calling all of it AI.
#1 Best Overall
Categories that commonly carry AI investment
| Filing category | What it can represent | How confidently it can be called AI spending |
|---|---|---|
| Servers and computing equipment | Accelerators, CPUs, memory, racks and general-purpose compute used for training, inference and other services | High when management explicitly links purchases to AI; otherwise mixed |
| Network equipment | High-speed interconnects, switches and routers that connect distributed compute and storage | Often AI-enabling, but also used by non-AI cloud and internet services |
| Storage | Training data, model checkpoints, application data and ordinary cloud storage | Mixed unless the filing identifies an AI workload |
| Data-center land and construction | Buildings, power, cooling and site work that house future capacity | Infrastructure for many workloads; allocate to AI only when the company supplies a basis |
| Office facilities and other property | Corporate buildings and nontechnical assets | Generally not AI infrastructure unless specifically described otherwise |
Alphabet’s disclosures illustrate the issue. Its filings describe capital investments as mainly servers and network equipment, data-center land and construction, and office facilities. The company’s 2025 Form 10-K says it continues to scale technical infrastructure, particularly for AI, but the reported total is still a broad capital-investment number rather than an AI-only subtotal.
Where in a 10-K or 10-Q do I find historical AI-related spending?
Use several sections together. No single table reliably captures cash paid, equipment purchased but not yet operating, leased capacity and future construction obligations.
PC 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 & 11Crashes, 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 minute- Start with investing cash flow. Find purchases of property and equipment in the consolidated statement of cash flows. This is the best starting point for cash actually paid during the period, but it may include every business line and may differ from additions recorded in the property-and-equipment note.
- Reconcile the property-and-equipment note. Compare purchases or additions, disposals, depreciation and other movements. The reconciliation explains why cash capex and the balance-sheet increase are not identical.
- Read the asset-class breakdown. Look for servers, network equipment, buildings, land, construction in progress or similar categories. A separate “technical infrastructure” label is useful, but it still may contain AI and non-AI workloads.
- Check management’s discussion and analysis. MD&A often explains why spending changed, which capacity is being added and whether demand is exceeding supply.
- Search commitments and contingencies. Construction contracts, purchase commitments and other obligations can represent economic commitments that have not yet appeared as current-period capex.
- Read lease disclosures and debt notes. Data-center leases can provide capacity alongside owned property and equipment. Debt, cash and available liquidity determine whether the planned build-out is financeable.
- Use the earnings call for forward guidance. Treat a forecast as a plan, not as cash already spent, and record the period, currency and assumptions attached to it.
Alphabet’s reported capex provides a useful baseline
Alphabet reported total capital expenditures of $52.5 billion in 2024 and $91.4 billion in 2025 in its 2026 Form 10-K. The increase was $38.9 billion, or approximately 74.1% year over year.
| Company and period | Reported capital spending | What the figure covers | What it does not establish |
|---|---|---|---|
| Alphabet, 2024 | $52.5 billion | Broad capital investments, mainly servers, network equipment, data-center land and construction, and office facilities | It is not an AI-only subtotal |
| Alphabet, 2025 | $91.4 billion | Same broad disclosure framework; management highlighted scaling technical infrastructure for AI | The filing does not allocate the total precisely between AI and other workloads |
Use these numbers as historical cash-spending anchors. To estimate AI exposure, you must continue into the technical-infrastructure descriptions, assets-not-in-service disclosures and management commentary rather than dividing the total by an unsupported percentage.
How do I rebuild the AI scope from broad infrastructure categories?
Build an evidence ladder instead of a single guess. Put each item into a category, record the exact wording management uses, and assign a confidence level.
Step 1: Copy the company’s own asset labels
Record terms such as servers, network equipment, storage, data-center land, buildings, construction in progress and technical infrastructure. Do not replace them with your preferred labels; the original wording lets you reconcile your worksheet to the filing later.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Rank #2
Step 2: Link each label to an operating explanation
Alphabet’s 10-Q describes technical infrastructure as servers and network equipment for computing, storage and networking, together with data-center land and building construction. That language establishes what the assets are, but not that every dollar supports AI. Look for a second statement tying the assets to AI products, model training, inference demand or enterprise customers.
Step 3: Separate attributable, shared and unknown amounts
- Attributable: spending explicitly identified as AI infrastructure or a named AI facility.
- Shared: equipment that supports AI and other cloud, search, advertising or software workloads.
- Unknown: broad property, construction or equipment totals with no workload allocation.
Report an attributable subtotal and a broader technical-infrastructure total side by side. A range is more honest than false precision: a low case can include only explicitly identified AI assets, while a high case can include shared technical infrastructure when management says AI is the principal growth driver. Label the assumptions and do not present the high case as a reported figure.
Why do assets not yet in service change the timing of AI spending?
Cash can leave the company long before a data center produces useful capacity. Assets under construction or not yet in service are primarily technical infrastructure in the disclosures reviewed here. Data-center projects can take months to years from land purchase and construction through equipment installation and deployment.
The accounting timeline
- Commitment: the company signs a purchase, construction or lease agreement.
- Cash payment: deposits and progress payments appear in cash flow or remain as contractual obligations.
- Construction and installation: costs accumulate in assets under construction or a similar balance-sheet category.
- Ready for intended use: the facility or equipment enters service.
- Depreciation: depreciation begins when the asset is ready for its intended use, not necessarily when the first customer revenue arrives.
This lag creates two common traps. A large capex year may precede the capacity and revenue it was intended to support, and a later depreciation increase may reflect assets completed after an earlier spending surge. Track additions, assets not yet in service, placed-in-service timing and depreciation expense as separate lines in your model.
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 minuteQuestions to ask about construction in progress
- How much of the balance is expected to enter service in the next quarter or year?
- Are projects concentrated in a few regions, power markets or suppliers?
- Does management report permitting, electricity, equipment or labor constraints?
- Will the completed sites support AI, general cloud demand or both?
How should I read forward AI-capex guidance from an earnings call?
Forward guidance belongs in a separate column from historical cash spending. Record the date of the call, the period covered, whether the number is gross capex or a narrower category, and any cost assumptions management includes.
Microsoft’s official FY2026 third-quarter earnings call illustrates the distinction. CFO Amy Hood said: “We expect to invest roughly $190 billion in capital expenditures which includes approximately $25 billion from the impact of higher component pricing.” The statement is guidance for calendar 2026, not a report of cash already paid. The company also said it expected to remain constrained through 2026 even as GPU, CPU and storage capacity came online faster.
How to interpret a large forecast
- Separate volume from price. In Microsoft’s statement, approximately $25 billion reflects higher component pricing. A higher dollar total therefore does not automatically mean an equivalent increase in physical capacity.
- Check the supply constraint. If management still expects to be constrained, spending may be aimed at relieving bottlenecks rather than creating immediately excess capacity.
- Compare with depreciation and delivery dates. Guidance can cover equipment ordered now but installed later, so the income-statement effect may arrive over several periods.
- Look for revisions. Compare the newest outlook with the previous quarter’s range and explain whether changes reflect demand, pricing, delivery timing or a change in scope.
How do leases and other commitments extend the economic cost beyond capex?
Owned property and equipment is only one way to secure AI capacity. Alphabet reports future data-center lease payments and explains that lease arrangements provide capacity alongside purchases of property and equipment. A company can therefore expand its effective infrastructure footprint without paying the full construction cost in the current period.
Items to add to your commitment view
| Item | Where to look | Why it matters |
|---|---|---|
| Operating and finance leases | Lease commitments and lease-liability notes | Future payments can secure data-center space, power or equipment capacity |
| Purchase commitments | Commitments and contingencies | Orders for equipment or services may require cash later even if current capex is modest |
| Construction obligations | Contractual-obligation tables and MD&A | Projects can create multi-year spending needs before assets are placed in service |
| Debt maturities and new borrowing | Debt and liquidity notes | Financing capacity affects whether planned AI expansion can continue |
| Assets under construction | Property-and-equipment note | Shows capital already committed but not yet generating depreciation or capacity |
Do not add every lease payment mechanically to current-period capex. Instead, present a cash-capex measure and a broader committed-capacity measure. Explain the term, payment schedule and whether the arrangement is cancellable, because a five-year lease and a non-cancellable construction contract carry different risks.
How can I tell whether AI infrastructure spending is turning into revenue?
Capex alone cannot answer that question. Build a chain from dollars spent to usable capacity, utilization and monetization.
- Capacity: identify GPUs, CPUs, storage, networking, data-center space or other units management says are being added.
- Availability: check when assets become operational and whether supply constraints delay customer access.
- Demand: look for disclosed backlog, committed contracts, cloud growth, usage trends or statements that demand exceeds supply.
- Pricing and mix: determine whether revenue growth reflects higher prices, more usage, new products or acquisitions.
- Margin and depreciation: watch whether depreciation, power and operating costs rise as capacity enters service.
- Liquidity: verify that the spending pace is supported by operating cash flow, cash reserves or sustainable financing.
A useful worksheet keeps these observations separate. For example, strong demand with constrained capacity suggests spending is intended to unlock sales, while rising capex, weak utilization commentary and accelerating depreciation can indicate execution or overbuild risk. Neither pattern proves success by itself; it tells you which follow-up disclosures matter.
What is the right way to compare two companies’ AI spending?
Compare companies on consistent axes, not headline dollars. Align fiscal periods, currency, accounting definitions and the portion of infrastructure that is actually identified as AI-related.
| Comparison axis | Calculation or question | Interpretation |
|---|---|---|
| Capex growth | (Current-period capex − prior-period capex) ÷ prior-period capex | Shows spending acceleration, but not whether the increase is AI |
| AI-attributable scope | Explicitly identified AI assets divided by total capex | Use only when the numerator is disclosed or transparently estimated |
| Technical-infrastructure mix | Servers, networking, storage and data-center construction versus offices and other property | Indicates how much spending is directed toward operational capacity |
| Assets not yet in service | Construction-in-progress and other not-yet-operational assets | Signals future capacity and timing risk |
| Lease exposure | Future data-center and equipment payments | Captures commitments outside owned-property capex |
| Depreciation ramp | Depreciation growth relative to additions and assets placed in service | Shows when prior investments begin affecting earnings |
| Demand and capacity | Reported utilization, backlog, supply constraints and delivery timing | Tests whether spending is responding to real demand |
| Liquidity | Operating cash flow, cash, debt maturities and financing access | Measures the ability to sustain the build-out |
Alphabet and Tesla show why context is essential
Tesla’s 2025 annual report presented capital spending of $8.53 billion in 2025 and $11.34 billion in 2024, a decrease of $2.81 billion, or approximately 24.8%. Tesla described the spending as mainly for global AI and operational infrastructure and factory expansion.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Rank #4
| Company | 2024 capex | 2025 capex | Reported context | Comparability warning |
|---|---|---|---|---|
| Alphabet | $52.5 billion | $91.4 billion | Servers, networking, data-center land and construction, office facilities; management emphasized AI-related technical infrastructure | Broad total, not an AI-only figure |
| Tesla | $11.34 billion | $8.53 billion | Mainly global AI and operational infrastructure plus factory expansion | Factory expansion and automotive operations make the mix different from a cloud platform |
The comparison shows why a larger number does not automatically mean a more aggressive AI strategy. Alphabet operates a very different infrastructure model, while Tesla’s total includes factory expansion and operational assets. Compare each company’s workload mix, revenue model, asset life and balance-sheet capacity before drawing a conclusion.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Which calculations make an earnings-report reading more rigorous?
Year-over-year capex growth
Use (current capex − prior capex) / prior capex. State whether the inputs are annual or quarterly and whether the company changed fiscal-year length. For Alphabet, the annual figures above produce approximately 74.1% growth; that is growth in reported total capex, not proof that AI spending rose by the same percentage.
Capex intensity
Calculate capex / revenue for the same period, then inspect the trend. A rising ratio can indicate an investment phase, but it can also reflect a temporary revenue decline, acquisitions or a change in accounting scope.
Cash-capex versus committed-capacity range
Present current-period cash capex separately from lease payments, purchase commitments and construction obligations due in future periods. Avoid adding unlike items into one “AI spend” number; show the timing and contractual nature of each component.
Recommended Free Tools
Depreciation conversion
Track depreciation expense against additions and assets placed in service. A lag between additions and depreciation is normal for long projects. An abrupt depreciation increase can help confirm that earlier construction has become operational, but it still does not identify how much revenue came from AI.
Liquidity runway for the build-out
Compare planned capex and contractual payments with operating cash flow, cash balances, available credit and debt maturities. This is a stress test, not a forecast: it shows how much financing flexibility remains if demand, component prices or project schedules change.
Best Value
What mistakes make AI-capex analysis unreliable?
- Calling all capex AI. Offices, factories, general cloud services and legacy infrastructure may sit in the same total.
- Using earnings-call guidance as historical spending. Forecasts describe intended investment and can change.
- Ignoring price inflation. Microsoft’s 2026 outlook specifically included approximately $25 billion from higher component pricing, so dollars and physical capacity can move differently.
- Counting commitments as cash paid. Leases and purchase obligations may be economically important but are not current-period capex.
- Overlooking assets not yet in service. Construction can consume cash for months or years before producing depreciation or customer capacity.
- Comparing fiscal periods without alignment. A calendar-year forecast, a fiscal-year annual report and a quarterly filing are not interchangeable.
- Using depreciation as a revenue proxy. Depreciation records asset consumption; it does not identify AI sales or customer profitability.
- Ignoring liquidity and financing. A large plan is less credible when debt maturities, cash needs or supplier obligations are not considered.
- Quoting an unsupported AI percentage. If the company does not disclose an allocation, publish a range with assumptions or leave the amount unallocated.
How do I build a repeatable AI-capex worksheet?
Use one row per company and one column per evidence type. Update it after every annual and quarterly filing.
| Worksheet field | Entry to capture |
|---|---|
| Reporting period | Fiscal year or quarter, filing date and currency |
| Cash capex | Purchases of property and equipment from investing cash flow |
| Property-and-equipment additions | Asset-note additions, including construction in progress |
| AI-attributable amount | Only the portion explicitly identified or transparently bounded |
| Shared technical infrastructure | Servers, networks, storage and data centers without a precise workload split |
| Assets not yet in service | Balance, expected completion and deployment timing |
| Lease and purchase commitments | Payment schedule, term and cancellability |
| Forward guidance | Amount, period, assumptions and changes from prior guidance |
| Demand and capacity evidence | Backlog, utilization, constraints, delivery timing and pricing commentary |
| Depreciation and liquidity | Depreciation trend, operating cash flow, cash, debt and available financing |
| Confidence note | What is disclosed, what is inferred and what remains unknown |
Keep a quotation or page reference for every important number in your private notes. When the next filing arrives, update the same rows rather than starting with a new headline estimate. Changes in wording, asset classification or fiscal period can then be identified before they distort the trend.
What should a reader conclude from a large AI-capex number?
A large number is evidence of commitment, not evidence of return. The strongest analysis connects reported cash spending to identifiable infrastructure, shows when that infrastructure will enter service, adds leases and other obligations, and tests whether demand, pricing, depreciation and liquidity support the plan. When a company does not disclose an AI-only amount, the accurate conclusion is a bounded estimate with explicit uncertainty rather than a fabricated total.
Frequently Asked Questions
Can I estimate AI capex when a company gives no AI-specific number?
Yes, but present an attributable subtotal and a broader infrastructure range rather than a single asserted amount. Use only categories and management statements that can be tied to AI, and mark shared servers, networks and data centers as uncertain when they support other workloads too.
How should I compare a calendar-year AI forecast with a company’s fiscal-year filing?
Convert both to a common trailing-12-month or calendar-year basis only when the underlying quarterly data are available. Otherwise, show the periods side by side and explain that the forecast and historical total cover different months.
Does a rise in depreciation prove that AI revenue is growing?
No. Depreciation starts when assets are ready for use and reflects the cost allocation of completed equipment and facilities. It can confirm that prior construction entered service, but revenue attribution requires separate demand, usage or product disclosures.
Free tools Windows power users keep installed
One-click scans. No signup required.
How often should I update an AI-capex analysis?
Refresh the worksheet each quarter for guidance, commitments, construction status and liquidity, then reconcile the full-year totals when the annual report is filed. Record changes in definitions as well as changes in dollars so a reclassification is not mistaken for a spending surge.
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.
