Follow the AI Cash Flow - Where Hyperscaler Spending Goes Next

The first dollars buy compute. The next dollars build everything required to keep that compute running.

Everyone is watching hyperscaler capital spending.

Amazon expects to invest approximately $200 billion in 2026. Alphabet raised its 2026 capital-expenditure guidance to between $195 billion and $205 billion. Meta expects to spend between $130 billion and $145 billion. Microsoft invested $41 billion in its most recent quarter alone.

Those numbers are enormous.

But the number is not the most interesting part.

The direction is.

A hyperscaler does not write one check labeled “AI infrastructure.” The money travels through a supply chain that begins with processors and quickly expands into servers, memory, networking, power equipment, cooling systems, data centers, fiber routes, construction firms, utilities, landowners and infrastructure investors.

The first dollars buy intelligence.

The next dollars build the physical systems that allow it to operate.

That is why the best way to understand the AI infrastructure market may be to follow the cash.

The Short Answer

Where does hyperscaler AI spending go?

Hyperscaler capital spending typically moves through three layers:

1. Inside the Rack: GPUs, CPUs, memory, storage and networking equipment.

2. Inside the Data Center: electrical systems, cooling, backup power, servers, construction and security.

3. Outside the Fence: power generation, transmission, substations, fiber routes, land, water systems, workforce and infrastructure financing.

Companies such as Amazon, Microsoft, Alphabet, Meta and Oracle may initiate the spending, but much of the economic value moves outward to the companies building and operating the infrastructure beneath them.

This is the Follow the AI Cash Flow framework.

What Is a Hyperscaler?

A hyperscaler is a company that operates cloud and computing infrastructure at enormous scale.

Amazon Web Services, Microsoft Azure, Google Cloud and Oracle Cloud are traditional hyperscale cloud providers. Meta also operates hyperscale infrastructure to support its platforms, AI models and expanding computing requirements.

These companies build or lease large networks of data centers, purchase enormous volumes of processors and networking equipment, and contract for power, fiber and construction capacity across multiple regions.

Their capital budgets influence far more than the technology sector.

They can shape utility investment, equipment manufacturing, commercial real estate, construction activity, regional workforce demand and the value of infrastructure-ready land.

The Spending Signal

The latest hyperscaler spending plans show that AI infrastructure has entered an industrial-scale investment cycle.

Amazon: Approximately $200 Billion

Amazon expects to invest approximately $200 billion across the company in 2026, with AI, chips, robotics and other infrastructure opportunities among the major drivers.

The physical impact is already visible in its cash flow. Amazon reported that purchases of property and equipment increased by $66.1 billion over the trailing twelve months, primarily reflecting artificial-intelligence investment. Amazon

Alphabet: $195 Billion to $205 Billion

Alphabet raised its 2026 capital-expenditure guidance to between $195 billion and $205 billion, citing accelerated capacity delivery to meet demand.

Alphabet describes its capital investment as including servers, network equipment, data-center land and building construction. Alphabet Q2 2026 Earnings Alphabet Capital Expenditures

Meta: $130 Billion to $145 Billion

Meta expects 2026 capital expenditures, including principal payments on finance leases, to reach between $130 billion and $145 billion.

The company spent more than $31 billion during the second quarter alone as it expanded the computing infrastructure supporting its models, advertising systems and future AI products. Meta

Microsoft: $41 Billion in One Quarter

Microsoft reported $41 billion in capital expenditures during the quarter ending June 30, 2026.

Approximately two-thirds went toward shorter-lived assets, primarily GPUs and CPUs. That detail matters because it shows how much of the current spending wave is still concentrated in computing equipment—and how much supporting infrastructure will be required to make those processors productive. Microsoft

Oracle: $50 Billion

Oracle expected capital expenditures of approximately $50 billion for its 2026 fiscal year as it expanded Oracle Cloud Infrastructure.

The company is also using debt and equity financing to build additional capacity for customers including Meta, NVIDIA, OpenAI, xAI and other major AI companies. Oracle

These figures cover different reporting periods and do not all represent AI spending exclusively.

But the broader signal is unmistakable:

The hyperscalers are no longer purchasing technology alone.

They are financing a new industrial system.

Why the CapEx Headline Misses the Bigger Story

Wall Street sees a hyperscaler announce a larger capital budget and immediately asks whether the company is buying more GPUs.

Usually, it is.

But a processor is only useful when it is installed inside a functioning system.

It needs memory to feed it data, networking to connect it to other processors, cooling to control the heat, electrical equipment to deliver power and a physical building capable of supporting the entire environment.

That building needs a utility connection, fiber access, land, permits, construction workers and financing.

Each new layer creates another purchase order.

The result is a multiplier effect in which one dollar of compute spending creates demand across the broader AI infrastructure supply chain.

Layer One: Inside the Rack

The first destination for hyperscaler AI spending is the computing stack.

This Includes:

• GPUs and AI accelerators

• CPUs

• High-bandwidth memory

• Data-storage systems

• Servers and rack systems

• Custom silicon

• Advanced semiconductor packaging

NVIDIA remains the most visible beneficiary, but the computing layer extends well beyond one company.

AMD supplies accelerators and processors. Broadcom supports custom silicon and networking. TSMC manufactures many of the world’s most advanced chips. Micron, Samsung and SK Hynix supply critical memory. Companies such as Supermicro, Dell and Hewlett Packard Enterprise assemble processors, memory and networking into deployable systems.

Advanced packaging is becoming particularly important because the performance of an AI accelerator increasingly depends on how processors, memory and interconnects are integrated.

The chip may receive the headline.

The surrounding system receives the purchase orders.

Layer Two: Networking and Optical Infrastructure

AI processors do not operate independently.

Large training systems may involve thousands of accelerators exchanging information continuously. When the network slows down, expensive processors can sit idle waiting for data.

That makes high-speed networking part of the compute system rather than an accessory to it.

Hyperscaler Spending Flows Into:

• Ethernet and InfiniBand networking

• Switches and routers

• Optical transceivers

• Data-center interconnection

• Fiber-optic cable

• Network-management software
  

This puts companies such as Broadcom, Arista Networks, Cisco, Ciena, Nokia, Corning, Coherent and Lumentum deeper into the AI infrastructure story.

The demand does not stop inside the building.

Meta signed a multiyear agreement valued at up to $6 billion with Corning for optical fiber, cable and connectivity products supporting its US data-center expansion. Amazon followed with its own multiyear, multibillion-dollar Corning agreement. Corning and Meta Amazon and Corning

Everyone is watching the silicon.

The hyperscalers are also buying the glass.

Layer Three: Power Inside the Data Center

Once the processors arrive, the power bill begins.

Electricity must move from the utility connection through substations, transformers, switchgear, uninterruptible power supplies, busways and power-distribution units before reaching the computing equipment.

The higher the rack density, the more complicated that system becomes.

This Sends Hyperscaler Spending Toward Companies Providing:

• Transformers

• Switchgear

• Circuit breakers

• Busway systems

• Backup generators

• Battery systems

• Uninterruptible power supplies

• Power-management software

• Modular electrical systems

Companies including Eaton, Schneider Electric, Vertiv, ABB, Siemens Energy, GE Vernova and Caterpillar sit directly in this flow of capital.

These may not look like artificial-intelligence companies.

That is exactly the point.

They manufacture the equipment that turns electricity into usable computing capacity.

Layer Four: Cooling and Water Systems

Every watt entering a processor eventually becomes heat.

That heat must be removed continuously.

Traditional data centers relied heavily on air cooling. High-density AI systems are accelerating the adoption of liquid cooling, direct-to-chip systems, coolant-distribution units and larger central cooling plants.

Hyperscaler Spending Is Therefore Moving Toward:

• Liquid-cooling systems

• Chillers and heat exchangers

• Coolant-distribution units

• Pumps and piping

• Air-handling equipment

• Water-treatment systems

• Thermal-management software

• Heat-reuse technology

Vertiv offers a good example of how AI spending travels beyond the processor. The company reported that second-quarter 2026 sales increased 24% from the previous year and raised its full-year sales guidance to approximately $14 billion as demand for critical digital infrastructure continued growing. Vertiv

Companies such as Trane Technologies, Carrier, Johnson Controls and XNRGY also become increasingly relevant as data-center operators look for ways to manage higher heat loads efficiently.

Every new GPU arrives with two invoices.

One for the compute.

Another for keeping it alive.

Layer Five: Buildings, Land and Construction

Before a data center becomes a technology asset, it is a construction project.

The spending moves through architects, engineers, general contractors, electrical contractors, mechanical contractors, steel suppliers and equipment installers.

It Also Reaches:

• Landowners

• Real estate developers

• Data-center operators

• Engineering firms

• Construction companies

• Prefabricated building manufacturers

• Logistics providers

• Skilled trades

Companies such as DPR Construction, Turner Construction, AECOM, Jacobs, Burns & McDonnell, Black & Veatch, EMCOR, Comfort Systems USA and Quanta Services help turn hyperscaler plans into physical facilities.

Data-center developers and operators such as QTS, Vantage, STACK Infrastructure, Aligned Data Centers, Digital Realty and Equinix can also capture spending when hyperscalers choose to lease capacity rather than build and own every facility directly.

The distinction between a technology budget and a real estate budget is disappearing.

AI infrastructure is both.

Layer Six: Power Outside the Fence

The data center may be finished, but the electricity still has to come from somewhere.

This is where the AI cash flow begins moving beyond the campus.

Utilities May Need to Build:

• New generation

• Transmission lines

• Substations

• Distribution infrastructure

• Battery-storage systems

• Natural-gas infrastructure

• Grid-control systems

• Microgrids

• Behind-the-meter power

Quanta Services estimates that artificial intelligence could drive approximately 30% of its addressable market from 2026 through 2030, reflecting demand for transmission, substations and other electrical infrastructure. Quanta Services

The companies positioned along this layer include utilities, independent power producers, nuclear operators, renewable-energy developers, pipeline companies and grid-equipment manufacturers.

Constellation Energy, Vistra, NextEra Energy, NRG Energy, Duke Energy, Dominion Energy and Southern Company are part of the power conversation. So are Bloom Energy, GE Vernova, Siemens Energy, Hitachi Energy, Quanta Services and the regional utilities responsible for connecting projects.

The first AI bottleneck is power.

That means power is also where the next wave of capital must go.

Layer Seven: Fiber and Regional Connectivity

An AI campus without scalable connectivity is an expensive island.

Hyperscalers need high-capacity fiber routes connecting data centers to cloud regions, storage systems, enterprise customers and end users.

That Spending Can Reach:

• Long-haul fiber providers

• Metro-fiber networks

• Data-center interconnection companies

• Internet exchanges

• Subsea cable systems

• Cloud on-ramps

• Edge infrastructure

Lumen, Zayo, Crown Castle, Ciena, Corning, Equinix, Digital Realty and CoreSite control different pieces of this connectivity layer.

Fiber investment can also serve as an early geographic signal.

New routes, interconnection expansions and long-term fiber agreements may reveal where hyperscalers expect future computing demand before the market sees a completed data-center campus.

Power tells us where AI can operate.

Fiber helps tell us where the money may be going next.

Layer Eight: Infrastructure Capital

Hyperscalers have large balance sheets.

Even they are beginning to share the cost.

Meta and BlackRock recently created a venture to develop and own a one-gigawatt data-center campus in El Paso, Texas. The project represents more than $10 billion in investment, with Meta as the initial sole occupant and BlackRock bringing infrastructure and private-financing capacity. Meta and BlackRock

This structure provides a preview of where the market may be heading.

As AI Campuses Become Larger, Hyperscalers May Increasingly Use:

• Infrastructure joint ventures

• Private credit

• Sale-leaseback agreements

• Project finance

• Real estate investment trusts

• Utility partnerships

• Long-term power contracts

• Equipment financing

That brings BlackRock, Brookfield, Apollo, KKR, Blackstone, Blue Owl and other large infrastructure investors into the center of the AI buildout.

The hyperscaler may be the customer.

The infrastructure fund may own the asset.

How AI Spending Reaches the Eight Pillars

The Follow the AI Cash Flow framework connects directly to the eight pillars of the Infrastructure of Intelligence™.

Power

Capital moves into generation, transmission, substations, transformers, backup power and energy storage.

Connectivity

Spending reaches fiber manufacturers, carriers, network-equipment suppliers and interconnection providers.

Land

Developers compete for sites with realistic access to power, fiber, water and transportation.

Water

Investment flows into efficient cooling systems, reclaimed-water infrastructure, treatment and monitoring.

Workforce

Construction, electrical, semiconductor, utility and network projects increase demand for engineers, technicians and skilled trades.

Policy

Permitting, utility regulation, tax policy and community agreements influence where spending can move fastest.

Capital

Private credit, infrastructure funds, joint ventures and long-term contracts help finance increasingly large campuses.

Sustainability

Hyperscalers fund renewable energy, nuclear power, storage, efficiency improvements and new cooling technologies to manage the environmental impact of growth.

The money may begin on a technology company’s balance sheet.

It ultimately reaches the entire infrastructure system.

Company Watch: Vertiv

Everyone is watching who manufactures the processor.

I’m watching who keeps it cool and powered.

Vertiv sits at the intersection of two of the most important AI infrastructure constraints: electricity and heat.

The company supplies power-management systems, uninterruptible power, thermal-management equipment, liquid-cooling technology and modular infrastructure used inside data centers.

Its recent sales growth is not simply a technology story. It is a physical-capacity story.

As rack densities rise, every new computing deployment requires more sophisticated power and cooling. That gives Vertiv exposure to the infrastructure spending that follows after the processor has already been ordered.

The broader watchlist includes Eaton, Schneider Electric, Quanta Services, GE Vernova, Corning, Ciena, Trane Technologies and Comfort Systems USA.

This is not a prediction of stock performance.

It is a map of where AI infrastructure demand is appearing.

What the Market Is Missing

The market still tends to divide companies into two categories:

AI companies and everyone else.

That framework is becoming less useful.

A transformer manufacturer may have more direct exposure to the AI buildout than a software company using AI in its marketing. A construction contractor may benefit from hyperscaler spending without ever appearing in an AI index. A utility may become the deciding factor in a multibillion-dollar computing project.

The companies capturing the next wave of AI spending may not describe themselves as AI companies.

They may describe themselves as electrical-equipment manufacturers, engineering firms, cooling specialists, fiber providers, utilities or infrastructure investors.

Follow the purchase order, not the label.

What Leaders Should Start Asking

Investors, utilities, developers and economic-development leaders should look beyond the hyperscaler’s headline capital budget.

You Should Ask:

• How much spending is going toward short-lived computing equipment?

• How much is funding long-lived infrastructure?

• Which physical constraint will require the next purchase order?

• Which suppliers have available manufacturing capacity?

• Where are utilities receiving the largest credible load requests?

• Which infrastructure companies have contracted backlog rather than speculative exposure?

• Where are hyperscalers leasing capacity instead of owning it?

• Which regions can deliver power, fiber and permits on the required timeline?

• Who is financing the infrastructure outside the hyperscaler’s balance sheet?

• Which communities are preparing land and workforce before demand arrives?

The CapEx announcement tells us how much a hyperscaler expects to spend.

The infrastructure questions tell us who may receive it.

The Golden Nugget

The most important AI spending may appear two balance sheets away from the hyperscaler.

Amazon may order the capacity.

A data-center developer may build the facility.

A utility may construct the substation.

An equipment manufacturer may supply the transformer.

A contractor may install it.

Following the AI cash flow means tracing that chain rather than stopping at the company that made the original announcement.

Frequently Asked Questions

Where Does Hyperscaler AI Spending Go?

Hyperscaler AI spending goes toward processors, servers, memory, networking, data-center construction, power systems, cooling, fiber connectivity, land, utilities and infrastructure financing. The spending begins with compute but spreads across the full physical AI infrastructure supply chain.

Which Companies Are the Largest Hyperscalers?

The largest hyperscale infrastructure operators include Amazon Web Services, Microsoft Azure, Google Cloud, Meta and Oracle Cloud Infrastructure. These companies operate global data-center networks and invest heavily in AI computing capacity.

Which Industries Benefit From Hyperscaler Capital Spending?

Semiconductors, electrical equipment, cooling, construction, utilities, fiber networks, commercial real estate, engineering and infrastructure finance can all benefit from hyperscaler capital spending.

Why Does AI Require So Much Infrastructure Investment?

AI systems use large clusters of processors that require enormous amounts of electricity, high-speed networking and advanced cooling. Building and connecting those systems requires data centers, power infrastructure, fiber routes, land, water strategies and skilled workers.

What Should Leaders Watch Next?

Leaders should monitor utility load requests, transformer orders, cooling-system demand, data-center construction, fiber agreements, power contracts and infrastructure-financing partnerships. These signals can reveal where hyperscaler spending is moving before a facility becomes operational.

The IOI Take

The AI investment story is expanding outward.

The first phase centered on the model.

The second centered on the chip.

The next phase will center on the infrastructure required to deploy those chips at scale.

That is where the eight pillars become visible.

Power, connectivity, land, water, workforce, policy, capital and sustainability are not secondary beneficiaries of the AI boom. They are the systems receiving the next round of investment.

The CapEx number tells us the size of the ambition.

The cash flow tells us where the Intelligence Economy is physically being built.

One Last Thought

Hyperscalers may be writing the largest checks in the AI economy.

They will not keep the money.

It will move through chip factories, equipment plants, construction sites, utility territories, fiber corridors and infrastructure funds. These need to be on your radar.

Some of the most important beneficiaries may sit several steps away from the model and far outside the traditional technology sector.

So yes, watch the hyperscalers.

But follow the cash.

Workforce

Policy

Capital

Sustainability