Cloud providers are forecasting the largest disclosed infrastructure outlays: approximately $190 billion for Microsoft and $200 billion for Amazon in 2026, while Meta expects $115 billion to $135 billion. Alphabet reported $91.4 billion of 2025 capital expenditures and said its 2026 technical-infrastructure investment will rise significantly. Those figures are not an apples-to-apples measure of AI spending, however. They cover different periods, assets and accounting treatments, and the available disclosures do not support a standardized dollar ranking across cloud platforms, chipmakers and model companies.
The useful comparison is therefore structural: who finances and owns capacity, what equipment the money buys, how demand is evidenced, and when the spending may turn into revenue or cash flow.
As an Amazon Associate I earn from qualifying purchases.
What this comparison can—and cannot—measure
“AI infrastructure spending” is used to describe several different transactions. A cloud platform may build a data center, buy servers and accelerators, install networking equipment, and rent capacity to customers. A chipmaker may design and manufacture an accelerator that a cloud platform purchases. A model company may buy cloud capacity, sign a long-term compute commitment, lease equipment or build its own facilities.
Those amounts sit at different points in the supply chain. Adding a cloud provider’s capex to a chipmaker’s sales can count the same accelerator twice: once when the cloud company buys it and again when the chip supplier records revenue. A model company’s compute commitment may represent future payments for capacity already included in a provider’s investment plan. No source cited here supplies a harmonized measure that makes all three categories directly rankable.
#1 Best Overall
The figures below are consequently best read as company-specific disclosures, not as a league table of “who spends the most on AI.” Every number retains the company’s own period, scope and accounting description.
Disclosed spending at a glance
| Company | Period and status | Disclosed amount | What the figure covers |
|---|---|---|---|
| Alphabet | 2025 actual; 2026 outlook is qualitative | $91.4 billion of 2025 capex | Technical infrastructure, including servers, network equipment, data-center land, construction and improvements. Alphabet expects a significant 2026 increase but gives no comparable dollar forecast in the cited filing. |
| Microsoft | Calendar-year 2026 forecast in FY2026 Q3 materials | Approximately $190 billion, including about $25 billion attributed to higher component prices | Company-wide capital expenditures, not an AI-only total. Microsoft describes a fleet using GPUs, CPUs, storage, networking silicon and other infrastructure. |
| Meta | 2026 forecast in its 2025 Form 10-K | Approximately $115 billion to $135 billion | Capex supporting AI efforts and the core business. The filing says 2025 property-and-equipment purchases primarily funded servers, data centers and network infrastructure. |
| Amazon | 2026 management forecast in the 2025 shareholder letter | Approximately $200 billion | Total company capex expectation. The letter discusses AWS capacity, customer commitments and expected monetization, but does not label the entire amount AI spending. |
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.
Why capex is not the same as AI-only investment
Alphabet: technical infrastructure with several uses
Alphabet’s reported $91.4 billion of 2025 capex is a realized, company-reported figure. Its filing defines technical-infrastructure investment broadly: servers and network equipment, data-center land, construction and improvements. Those assets can support AI services, search, advertising, cloud workloads and other products. Alphabet says it expects to “significantly increase, relative to 2025” its technical-infrastructure investment in 2026, but the cited filing does not provide a comparable dollar forecast.
Recommended Free Tools
The filing also describes both custom Tensor Processing Units and infrastructure using TPUs and GPUs. That matters because a custom accelerator purchase, a general-purpose server and a data-center building have different useful lives, supply constraints and economic roles even when they all appear under technical infrastructure.
Microsoft: a forecast affected by component prices
Microsoft’s approximately $190 billion figure is a calendar-year 2026 outlook in its FY2026 Q3 conference-call materials, not a 2025 actual and not an AI-only line item. Management says approximately $25 billion of the amount reflects higher component prices. A higher bill for the same class of equipment can raise capex without representing an equivalent increase in compute capacity.
Microsoft describes a fleet that combines NVIDIA and AMD accelerators with first-party networking silicon, CPUs and other components. The disclosed total therefore includes the facilities, power, networking, storage and compute needed to operate a broad cloud platform, not only the accelerators used for model training.
Meta: a range supporting AI and the core business
Meta’s 2025 Form 10-K gives a 2026 capex outlook of approximately $115 billion to $135 billion. The company explicitly says the spending supports its AI efforts as well as its core business. Its filing says 2025 purchases of property and equipment primarily funded servers, data centers and network infrastructure.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
Meta’s earlier 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 is the more current guidance for 2026 and uses its own property-and-equipment and capex labels; the earlier range should not be substituted for the newer outlook.
Amazon: a total-capex forecast with AWS commitments behind it
Amazon CEO Andy Jassy’s 2025 shareholder letter says the company expects approximately $200 billion of capex in 2026. That is a management forecast for Amazon’s total capital spending. The letter links a substantial portion of planned AWS capacity to existing customer commitments and says much of the 2026 AWS capex is expected to monetize in 2027–2028.
Jassy wrote, “We’re not investing approximately $200 billion in capex in 2026 on a hunch.” That is management’s explanation of the plan, not independent proof that the spending will earn a particular return. The timing language is important: capacity can be paid for and placed in service before the associated customer revenue arrives.
The infrastructure layers being financed
| Layer | Typical items | Economic question | Why it is not interchangeable |
|---|---|---|---|
| Facilities and power | Land, data-center shells, electrical systems, cooling and construction | Who finances and owns the site, and when can it be energized? | Long-lived assets can be underused if power or permits delay deployment. |
| Servers and general compute | CPUs, memory, storage and server chassis | How much ordinary cloud and enterprise demand will use the equipment? | These assets serve many workloads, not only model training. |
| Accelerators | GPUs, TPUs and other AI-focused processors | Can the owner secure supply and keep utilization high? | They can be supply-constrained, expensive and subject to rapid performance changes. |
| Networking | Switches, interconnects, optical links and first-party networking silicon | Can data move among processors fast enough for large models? | Network capacity can become the bottleneck even when compute is available. |
| Software and operations | Orchestration, reliability, security and energy-management systems | Can the equipment be operated efficiently at scale? | Some costs are expensed rather than capitalized, so capex alone misses them. |
A reported capex total blends several of these layers. A sound comparison keeps the physical layer and the owner’s accounting treatment visible instead of treating “AI infrastructure” as one uniform purchase.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Cloud providers: the clearest disclosed capital commitment
Cloud platforms finance infrastructure before customers consume it. They may own the land and buildings, purchase servers and accelerators, and recover the investment through multi-year cloud contracts or usage-based services. Their capex figures are therefore the most visible part of this comparison, but visibility does not make them AI-specific.
Microsoft’s capacity constraint and supply signal
Microsoft’s investor-relations materials state: “Broad and growing customer demand continues to exceed supply.” The same passage says Microsoft expects GPU, CPU and storage capacity constraints through 2026. This is evidence of current demand and a constrained supply chain, not evidence that every dollar of the projected $190 billion will earn an attractive long-term return.
The approximately $25 billion component-price effect also shows why nominal spending and physical capacity can diverge. If prices rise, capex may increase even if the number of deployable servers grows more slowly.
Amazon’s committed AWS capacity and delayed monetization
Amazon’s letter provides a different demand indicator: a substantial portion of planned AWS capacity already has customer commitments. It also expects much of the 2026 AWS capex to be monetized in 2027–2028. A commitment reduces uncertainty about a customer relationship, but it does not eliminate construction, delivery, utilization, energy or renewal risk.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Amazon explicitly describes near-term free-cash-flow pressure while new capacity is built ahead of revenue. That is a normal consequence of a capacity cycle: cash leaves the business when equipment and facilities are purchased, while customer payments may arrive later.
Alphabet’s TPU-and-GPU mix
Alphabet’s infrastructure plan combines custom TPUs with GPU-based systems. Custom silicon can be designed around the company’s workloads, while GPUs can offer broader software compatibility and external supply. The filing does not provide a dollar split between those approaches, so no percentage allocation should be inferred from the $91.4 billion total or the qualitative 2026 increase.
Meta’s range and finance-lease disclosure
Meta presents a range rather than a single 2026 point estimate. The company’s accounting discussion also highlights finance leases in its earlier 2025 outlook. Lease principal can represent a contractual commitment to equipment or facilities without appearing in the same way as a cash purchase in every capex presentation. When comparing companies, check whether finance-lease principal is included, excluded or disclosed separately.
Chipmakers: suppliers, not equivalent capex owners
Chipmakers occupy a different position from cloud platforms. Their reported revenue generally reflects sales of accelerators, CPUs, networking products or related components to customers such as cloud providers, system builders and enterprises. That revenue is not the chipmaker’s own data-center capex, and it cannot be placed beside a cloud provider’s total capex as though both were final infrastructure purchases.
Crashes, 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 minuteWindows 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 reinstallThe available disclosures here identify Microsoft’s use of NVIDIA and AMD products and Alphabet’s use of TPUs and GPUs, but they do not provide a standardized 2026 infrastructure-spending figure for the chipmakers. They also do not establish a comparable split between a chipmaker’s research, manufacturing capacity, packaging, inventory and corporate capital expenditures.
For an investor analyzing a chip supplier, relevant questions differ from those for a cloud owner:
Rank #4
- Is demand being converted into recognized product revenue or merely customer commitments?
- Are advanced packaging, memory, networking and manufacturing partners limiting shipments?
- How quickly could a new accelerator generation change pricing or utilization?
- How much of the supplier’s own capital spending supports AI products versus other markets?
Without company-specific, consistently defined numbers, a precise chipmaker-versus-cloud-provider spending ranking would be misleading.
Model companies: compute buyers and capacity planners
Model companies usually record infrastructure differently from cloud platforms. A developer may rent training clusters, purchase inference capacity, sign a take-or-pay commitment, lease equipment or build facilities with a partner. The resulting expense can appear as cloud services, operating expense, lease obligations or property and equipment rather than as one comparable “AI capex” line.
The sources cited for this comparison do not provide a harmonized dollar series for model-company spending. Nor do they establish a ranking of model developers by owned infrastructure, contracted compute or annual cash payments. A model company can be a major source of demand without owning the data center in which its workloads run.
To evaluate a model company’s infrastructure exposure, separate:
- Owned capacity: buildings, servers and networking recorded on the company’s balance sheet.
- Contracted capacity: committed cloud or colocation services that may require future payments.
- Usage-based compute: variable charges that rise with training runs, users or inference volume.
- Partner-funded capacity: infrastructure financed by another company in exchange for a commercial commitment.
These categories answer different questions about liquidity, operating leverage and supply security. They should not be merged into the cloud providers’ capex totals.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Demand, returns and the limits of management guidance
Management commentary supplies useful signals but not a guaranteed return calculation. Microsoft reports demand exceeding supply and continuing capacity constraints. Amazon points to customer commitments and a 2027–2028 monetization window. Alphabet describes a significant increase in technical infrastructure. Meta ties its range to both AI and its established business.
None of those statements proves that demand will remain strong for the useful life of every server or facility. Customer commitments can be renegotiated; utilization can vary; power and construction delays can postpone revenue; and model efficiency improvements can reduce the compute required for a given service. A careful analysis labels these statements as company guidance and demand indicators.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
Cash flow, depreciation and operating-cost effects
Free cash flow arrives before revenue
Amazon’s shareholder letter directly describes near-term free-cash-flow pressure while capacity is built. The same timing issue applies more broadly: equipment and construction require cash before a new cluster reaches steady utilization. A large capex forecast can therefore depress current free cash flow even when management expects later revenue growth.
Depreciation follows deployment
Once servers and facilities are placed in service, depreciation increases. Meta identifies accelerating depreciation and higher operating costs as infrastructure expands. Depreciation is non-cash in the period recorded, but it represents the accounting allocation of prior spending and can lower reported operating income.
Energy and network costs rise with usage
Alphabet expects infrastructure costs—including depreciation, energy, equipment and network capacity—to rise as AI services require more compute. These are ongoing operating costs, distinct from the initial purchase of a server or data center. A project can therefore look affordable on a capex basis while creating substantial recurring power, cooling, maintenance and networking expense.
Free tools Windows power users keep installed
One-click scans. No signup required.
Component inflation changes the economics
Microsoft’s approximately $25 billion component-price effect illustrates a further risk. Higher prices can increase the cash required to reach a capacity target. Investors should distinguish a larger equipment bill caused by inflation from a larger installed base that can serve more workloads.
A practical framework for comparing the three positions
- Normalize the period. Mark each figure as a reported actual or a forecast, and note whether it uses a fiscal or calendar year. Alphabet’s $91.4 billion is a 2025 actual; Microsoft, Meta and Amazon figures are 2026 outlooks.
- Write down the scope. Record whether the company says total capex, technical infrastructure, property-and-equipment purchases, AI-specific spending or customer commitments. Do not silently relabel a broad total as AI capex.
- Identify the owner. Ask who finances the building, owns the servers and carries the depreciation. A model company’s contracted compute is not the same asset as a cloud provider’s owned cluster.
- Map the layer. Separate land and power, data-center construction, general servers, accelerators, networking and software. Supply constraints or useful lives differ at each layer.
- Check leases and price effects. Determine whether finance-lease principal is included and whether higher component prices inflated the nominal total.
- Separate demand from returns. Customer commitments and capacity shortages indicate demand, but they do not establish utilization, pricing, margins or return on invested capital.
- Follow the cash-flow lag. Compare the spending date with the expected monetization date, depreciation ramp and recurring energy and network costs.
- Do not add supply-chain totals. A provider’s purchase can be a chipmaker’s revenue and a model company’s rented capacity. Summing them can double-count one deployment.
What the disclosures mean for a personal-finance reader
For household investors or savers, the central lesson is not that one category is automatically a better investment. It is that exposure can occur through different business models. Cloud providers carry construction, power, utilization and financing risk. Chipmakers carry product-cycle, manufacturing and supply-chain risk. Model companies carry compute-cost, contract and demand risk, often without owning the underlying facilities.
When reading an earnings release or annual report, look for the exact label attached to the number, the period covered, lease treatment, expected capacity utilization and the date management expects revenue to catch up. A headline capex figure is a starting point for analysis, not a forecast of earnings or a valuation by itself.
Why a definitive cloud-versus-chipmaker-versus-model-company ranking is not available
The primary disclosures establish large and rising cloud-platform commitments, but they do not offer a single accounting basis across the entire AI supply chain. Cloud capex is a purchase of facilities and equipment; chipmaker revenue is a supplier-side sale; model-company commitments may be future service payments or leases. The same physical capacity can appear in more than one category at different stages.
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 →Clear out junk files and repair common Windows errorsFree Scan →Even Alphabet’s figures illustrate the issue. The Stanford Institute for Human-Centered AI’s 2026 AI Index Report describes more than $150 billion in 2025 capex for Google, while Alphabet’s own 2025 Form 10-K reports $91.4 billion. The materials do not reconcile the difference, which may reflect broader grouping or accounting scope. The company filing is the appropriate source for Alphabet’s reported capex, and the two figures should not be averaged or presented as directly comparable.
Bottom line
Cloud providers are committing the largest clearly disclosed sums because they finance and operate the physical infrastructure. Chipmakers and model companies remain essential parts of the AI economy, but their spending and revenue are recorded differently. The most reliable comparison keeps the accounting scope, infrastructure layer, ownership, demand evidence and cash-flow timing attached to every number—and refuses to turn unlike figures into a false ranking.
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.
