AI Infrastructure Spending: How Cloud Providers, Chipmakers, and Model Companies Compare

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

Short answer: cloud providers are currently the only group in this comparison with large, directly disclosed infrastructure-spending forecasts. Microsoft expects about $190 billion of calendar-year 2026 capital expenditure, Amazon’s chief executive expects about $200 billion of total 2026 capex, Meta forecasts approximately $115 billion to $135 billion, and Alphabet reported $91.4 billion of 2025 capex while guiding to a significant 2026 increase. Those figures are not equivalent measures of “AI spending,” however. They cover different periods, assets and accounting treatments, and each company says the investment supports a mix of AI, cloud, core-business and other capacity.

There is no defensible, standardized dollar ranking here across cloud operators, chipmakers and model companies. Cloud capex pays for facilities and equipment; chipmaker revenue records sales into that supply chain; and model-company compute may appear as cloud consumption, equipment purchases, leases or contractual commitments. Adding the figures would risk counting the same server or accelerator more than once.

#	Preview	Product	Price
1		MINISFORUM MS-02 Ultra Workstation Mini PC, Intel Core Ultra 9 285HX (24C/24T, up to 5.5GHz), PCIe...	$1,659.00	Buy on Amazon
2		GMKtec EVO-X2 AI Mini PC Ryzen Al Max+ 395 Superchip 128GB LPDDR5X 2TB SSD	$3,649.99	Buy on Amazon

Why an apples-to-apples AI spending ranking does not exist

“AI infrastructure spending” can describe several layers of one transaction. A cloud operator may finance a data-center building, buy servers and networking equipment, install accelerators and then rent the resulting capacity. A chip designer or manufacturer records revenue when it sells an accelerator, CPU, networking part or related system. A model company may buy hardware, reserve cloud capacity or pay a usage bill. The same accelerator can therefore appear as a supplier sale, a cloud provider’s capital investment and a model company’s operating expense.

The public disclosures available for this comparison do not provide a harmonized measure for all three categories. The most reliable comparison is therefore a structured reading of each company’s reporting scope, rather than a league table that pretends every dollar is the same.

#1 Best Overall

MINISFORUM MS-02 Ultra Workstation Mini PC, Intel Core Ultra 9 285HX (24C/24T, up to 5.5GHz), PCIe 5.0 x16, 32GB RAM 1TB SSD,USB4 v2 80Gbps, Dual 25GbE+10GbE+2.5GbE, Wi-Fi 7, 350W PSU

• High-Performance AI Processor：The MS-02 Ultra features an Intel Core Ultra 9 285HX (24C/24T, up to 5.5 GHz, 13 TOPS NPU), delivering fast and efficient performance for AI inference, algorithm development, and media workloads. A PCIe x16 expansion slot supports desktop-class GPU upgrades for advanced model training and accelerated computing tasks. It's ideal for creators, engineers, and teams handling intensive parallel workloads.
• 4 × M.2 PCIe 4.0 + 4 × DDR5 SODIMM slots:Four DDR5 SODIMM slots support up to 256 GB of memory, while ECC helps maintain data integrity in mission-critical environments. Four PCIe 4.0 M.2 slots support up to 24 TB of storage, supporting RAID 0/1/5/10, combining high-speed performance with data protection. It allows for the creation of independent scratch disks, media libraries, and project drives, providing high-throughput for production workflows.
• PCIe & USB 4.0 v2: Up to three PCIe slots can be equipped, including a dual-slot x16 GPU. The main slot supports PCIe 5.0, meeting the needs of high-bandwidth creative and computing workloads. USB 4.0 v2 (80Gbps) supports high-bandwidth external storage and displays.
• Ultra-fast Networking: Wi-Fi 7 further enhances wireless performance with next-generation speeds and low-latency stability. Intelligent bandwidth switching optimizes throughput in different network environments, ensuring optimal performance for enterprise or local networks. Dual 25GbE ports (providing up to approximately 3.125 GB/s bandwidth, about 25 times faster than traditional 1GbE), enabling seamless large-scale file transfers and parallel computing. 10GbE and 2.5GbE ports, with support for Intel vPro technology, ensure enterprise-grade remote management and deployment flexibility.
• Server-grade thermal architecture: Utilizing a dedicated CPU/GPU airflow design, equipped with a 6-pipe dual-fan cooler, it maintains stable performance even under sustained loads, delivering up to 140W Turbo power while maintaining a 100W TDP, and operating with noise levels as low as 36 dB. An integrated 350W power supply ensures stable and reliable output for demanding computing tasks and fully loaded extended configurations.

$1,659.00

Buy on Amazon

The seven questions that make a comparison meaningful

• Period: Is the number an actual result or a forecast, and is it a fiscal or calendar year?
• Scope: Does it cover total capex, technical infrastructure, an AI-specific program or a customer’s contracted compute?
• Accounting: Are finance-lease principal payments, leased assets or component-price inflation included?
• Layer: Does the spending fund land and buildings, power, cooling, servers, networking, accelerators or custom silicon?
• Ownership: Who finances and owns the capacity, and who merely rents access?
• Demand evidence: Are there customer commitments, capacity constraints or cloud-growth indicators?
• Timing: When do depreciation, energy costs, cash outflows and revenue recognition occur?

What the major cloud providers have actually disclosed

The following figures should be read as company-reported capex or outlooks, not as a common measure of AI-only investment.

Alphabet: a reported baseline, followed by a qualitative increase

Alphabet’s 2025 Form 10-K reports $91.4 billion of capital expenditure for the year ended December 31, 2025. It says: In 2026, we expect to significantly increase, relative to 2025, our investment in our technical infrastructure, including servers and network equipment, and data centers. The filing does not provide a comparable dollar forecast for 2026 in the cited passage.

Alphabet’s definition is broader than accelerator purchases. Its technical-infrastructure description includes servers and network equipment as well as data-center land, construction and improvements. That means the $91.4 billion includes the physical platform required to operate services, not just machine-learning hardware. Alphabet also describes a mix of custom Tensor Processing Units (TPUs) and infrastructure using both TPUs and graphics processors, so “AI infrastructure” is not a single equipment line.

As this installed base expands, Alphabet expects infrastructure costs such as depreciation, energy, equipment and network capacity to rise. Those are operating consequences that follow the capital outlay and can continue even after a facility is complete. Read Alphabet’s 2025 Form 10-K.

Microsoft: the largest numerical outlook, with an explicit component-price caveat

Microsoft’s FY2026 Q3 earnings-call materials point to approximately $190 billion of calendar-year 2026 capex. The company says about $25 billion of that reflects higher component prices. This is a calendar-year outlook, not a 2025 actual and not an AI-only budget.

Microsoft says its fleet combines NVIDIA and AMD accelerators with first-party networking silicon, CPUs and accelerators. Spending therefore spans general compute, specialized AI processors and the network fabric that connects them. The company’s chief financial officer, Amy Hood, said, Broad and growing customer demand continues to exceed supply. The surrounding disclosure expects GPU, CPU and storage capacity constraints through 2026. That is management’s description of demand and supply, not independent proof that every planned dollar will earn an attractive return.

The $25 billion component-price statement matters when comparing years. A higher bill can represent more capacity, higher prices for the same components, or both. Treating the entire increase as additional computing power would overstate the physical expansion. See Microsoft’s FY2026 Q3 earnings materials.

Meta: a range that explicitly supports AI and the core business

Meta’s 2025 Form 10-K forecasts approximately $115 billion to $135 billion of 2026 capex. The filing says this investment supports its AI efforts and its core business. It also says 2025 property-and-equipment purchases primarily funded servers, data centers and network infrastructure.

The range is important: the midpoint is not a commitment, and the upper and lower ends imply a $20 billion planning spread. Meta’s earlier 2025 guidance used a different presentation that included principal payments on finance leases; the later Form 10-K is the appropriate source for the current 2026 outlook. Finance-lease treatment can make two otherwise similar infrastructure programs look different in capex tables.

Meta also warns that expanding infrastructure accelerates depreciation and raises operating costs. A capacity build therefore affects both the cash-flow statement when equipment is acquired and the income statement over the assets’ useful lives. Read Meta’s 2025 Form 10-K.

Amazon: a $200 billion company-wide plan tied to AWS demand

Amazon chief executive Andy Jassy says the company expects approximately $200 billion of total capex in 2026. He wrote, We’re not investing approximately $200 billion in capex in 2026 on a hunch. That sentence is management’s rationale, not an external validation of the investment case.

The shareholder 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. It also describes near-term free-cash-flow pressure because Amazon must build capacity before the related revenue arrives. The headline $200 billion covers Amazon as a whole, so it should not be presented as AWS or AI capex alone.

Customer commitments reduce demand uncertainty for some capacity, but they do not eliminate construction, power, financing, utilization or pricing risk. A committed reservation can also differ from recognized revenue depending on contract terms and delivery timing. Read Andy Jassy’s 2025 shareholder letter.

Why “capex” is not the same as “AI capex”

Capital expenditure is an accounting category, not a technology label. A data-center shell, electrical substation, cooling plant, fiber connection, general-purpose CPU server and AI accelerator can all be capitalized. The companies’ own descriptions show why a single total cannot be treated as an AI bill.

The physical stack paid for by a cloud operator

1. Site and building: land, construction, structural improvements and security.
2. Power and cooling: substations, backup generation, distribution, cooling systems and efficiency equipment.
3. Network: switches, optical links, routers and the interconnect required for distributed training and inference.
4. General compute: CPUs, storage and hosts that run control planes, databases and non-AI workloads.
5. Accelerated compute: GPUs, TPUs or other AI accelerators, plus high-bandwidth memory and specialized servers.
6. Software and operations: some costs may be expensed rather than capitalized, so a capex total will not capture the entire cost of operating an AI service.

Alphabet’s filing explicitly includes servers, network equipment, land and data centers. Microsoft describes a mixed fleet of third-party and first-party silicon. These disclosures support a layered interpretation, not an assumption that every dollar buys an accelerator.

Rank #2

GMKtec EVO-X2 AI Mini PC Ryzen Al Max+ 395 Superchip 128GB LPDDR5X 2TB SSD

• EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
• AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
• AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
• EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
• QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.

$3,649.99

Buy on Amazon

Where chipmakers fit—and why supplier revenue cannot be added to cloud capex

Chipmakers occupy a different position in the chain. Their reported revenue generally reflects sales of accelerators, CPUs, networking components, systems or related products to cloud operators, server manufacturers and other customers. Revenue is not the same as the chipmaker’s own capital investment, and it is not a measure of how much end users spend on complete AI infrastructure.

Three different chipmaker questions

• How much do chipmakers invest? A company’s own factories, equipment, research facilities and packaging capacity would be its capex, but that number is not supplied here in a standardized form.
• How much do they sell? Product revenue indicates demand moving through the supply chain, but it can include chips later installed in a cloud provider’s capital assets.
• How much capacity do they enable? A custom accelerator or networking component may be strategically important without appearing as a separate line in a cloud provider’s capex disclosure.

For that reason, a supplier’s sales cannot be stacked on top of Microsoft’s, Amazon’s, Meta’s or Alphabet’s capex to produce an industry total. The cloud operator’s purchase of a server may already contain the chipmaker’s product revenue. Adding both counts one economic transaction at two points.

Model companies: compute buyers, not a comparable capex category

Model companies can build their own clusters, purchase servers, lease equipment or reserve capacity from a cloud provider. They may report these arrangements as property and equipment, finance leases, cloud hosting expense, purchase commitments or a combination. A contractual commitment can represent future capacity rather than cash already spent.

What can and cannot be concluded

• Can be concluded: model development and serving require access to accelerators, storage, networking, power and data-center operations.
• Cannot be concluded from the available evidence: a comparable dollar ranking of model-company infrastructure spending against the cloud totals above.
• Cannot be assumed: that a model company’s committed cloud capacity is additional to the provider’s capex. The provider may be building that capacity specifically to satisfy the commitment.

A useful model-company comparison would need consistent disclosures for owned equipment, lease liabilities, cloud commitments, utilization, depreciation and operating compute expense. Those data are not provided here, so inserting estimates would create false precision.

How to compare the disclosures without double-counting

Use the following sequence when analyzing a new disclosure:

1. Label the period and whether the figure is actual, guidance or a range.
2. Copy the company’s own scope before assigning an AI interpretation.
3. Separate facilities, power, networking, general compute and accelerators where the filing does.
4. Check finance-lease treatment and whether the number includes principal payments or leased assets.
5. Record demand evidence separately from spending. A commitment or shortage statement is not a return-on-investment calculation.
6. Map the transaction chain so a supplier sale is not added to the customer’s equipment purchase.
7. Track the lag between spending, depreciation, operating costs and monetization.

Cash flow, depreciation and the lag to monetization

The investment cycle affects financial statements in different phases. Cash leaves when land, buildings, equipment or lease obligations are paid. Once assets are placed in service, depreciation generally spreads the cost over their useful lives. Energy, maintenance, networking and staffing add recurring operating expense. Revenue may arrive later, particularly when a new region or accelerator cluster is built ahead of demand.

Amazon explicitly describes near-term free-cash-flow pressure while capacity is constructed and expects much of the 2026 AWS investment to monetize in 2027–2028. Microsoft’s stated supply constraints indicate that demand is currently ahead of available capacity, but constraints do not establish future margins or utilization. Alphabet expects higher infrastructure costs as AI services consume more compute. Meta identifies accelerating depreciation and higher operating costs as its infrastructure expands.

These are timing observations, not forecasts of shareholder returns. A project can have committed customers and still face delays, price inflation, power limitations, utilization changes or technology obsolescence. Conversely, a general-purpose facility can support multiple workloads over a long life, making it misleading to assign its entire cost to one AI product.

How to read conflicting industry totals

A secondary AI Index report cites more than $150 billion in 2025 capex for Google. Alphabet’s own 2025 Form 10-K reports $91.4 billion. The materials available here do not reconcile the difference, which may reflect broader accounting or grouping. Do not average the numbers or present them as interchangeable. For Alphabet’s reported capex, prefer the company’s filing.

The same discipline applies to older guidance. Meta’s Q2 2025 release projected 2025 capex of $66 billion to $72 billion, including principal payments on finance leases. Its later 2025 Form 10-K provides the current 2026 outlook and different accounting labels. Comparing the earlier range directly with the later figure without explaining the change in period and treatment would produce a misleading trend.

Consult the AI Index 2026 report for the cited secondary figure, but keep its scope separate from Alphabet’s filing.

A practical decision framework for investors, analysts and technology buyers

If you need a cloud-provider comparison

• Start with reported capex and identify the period.
• Mark whether the company gives a range, a point estimate or only a qualitative outlook.
• Separate total-company spending from the portion explicitly tied to cloud or AI.
• Look for component-price effects, finance-lease treatment and customer commitments.
• Estimate the operational burden: depreciation, energy, networking and maintenance.

If you need a chip-supply assessment

• Track accelerator, CPU, networking and memory availability rather than treating all semiconductor revenue as AI capacity.
• Distinguish the chipmaker’s own factories and R&D from sales to cloud customers.
• Check whether a reported sale is incorporated into a server or system already counted in a provider’s capex.
• Consider custom silicon alongside merchant accelerators; both can support AI workloads.

If you need to evaluate a model company’s compute position

• Separate owned equipment, leases, cloud consumption and future commitments.
• Ask who bears power, maintenance, depreciation and utilization risk.
• Match reserved capacity to expected training and inference workloads.
• Do not treat a large commitment as current expenditure or guaranteed revenue.

What the current evidence supports

The defensible conclusion is directional rather than a three-way ranking. Cloud operators are committing extraordinary sums to a layered infrastructure build, with 2026 outlooks ranging from Meta’s approximately $115 billion–$135 billion to Amazon’s approximately $200 billion total capex and Microsoft’s approximately $190 billion calendar-year outlook. Alphabet’s $91.4 billion 2025 actual provides a reported baseline, while its 2026 guidance is qualitative.

Those amounts include shared facilities and capacity for AI and non-AI workloads. Chipmakers supply important components but their sales are not an additional pool of end-user infrastructure spending. Model companies may own or rent substantial compute, yet the available disclosures do not support a standardized dollar comparison. Any article or spreadsheet that simply adds cloud capex, chipmaker revenue and model-company commitments is likely to double-count the same hardware and to mix actuals with forecasts.

The most informative analysis keeps the layers separate, preserves each company’s accounting definition and follows the cash-flow path from construction and equipment purchases to depreciation, operating cost, customer usage and eventual monetization.

Quick Recap

Bestseller No. 1

MINISFORUM MS-02 Ultra Workstation Mini PC, Intel Core Ultra 9 285HX (24C/24T, up to 5.5GHz), PCIe 5.0 x16, 32GB RAM 1TB SSD,USB4 v2 80Gbps, Dual 25GbE+10GbE+2.5GbE, Wi-Fi 7, 350W PSU

$1,659.00

Bestseller No. 2

GMKtec EVO-X2 AI Mini PC Ryzen Al Max+ 395 Superchip 128GB LPDDR5X 2TB SSD

$3,649.99

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.