(248) 827-1010
  • Services
    • Wealth Management
    • Asset Management
    • Business Advisory
  • Who We Serve
  • Our Story
  • Resources
  • Services
    • Wealth Management
    • Asset Management
    • Business Advisory
  • Who We Serve
  • Our Story
  • Resources
  • Services
    • Wealth Management
    • Asset Management
    • Business Advisory
  • Who We Serve
  • Our Story
  • Resources
  • Services
    • Wealth Management
    • Asset Management
    • Business Advisory
  • Who We Serve
  • Our Story
  • Resources
Client Login
Contact Us
Client Login
Contact us

AI is an Infrastructure Business as well

Artificial intelligence is an infrastructure as much as it is a software. Every prompt sent to an AI model ultimately runs through a physical system of GPUs, servers, data centers, networking equipment, cooling systems, and enormous amounts of electricity. Building that system on a global scale requires an extraordinary amount of capital. 

Last month, I wrote about hyperscalers as one of several major new sources of demand for capital. That demand is getting larger. S&P Global Ratings projects combined capital expenditures by six major hyperscalers to exceed $1.3 trillion by 2027, with debt, leases, equity and other financing structures playing a growing role in supporting the buildout.[1] 

That leads to the next question: whether the revenue and cash flow generated by AI can eventually justify the amount of capital to build it. In other words, AI’s next test may not be intelligence, but return on capital.

The Free-Cash-Flow Machine Is Changing

For years, the largest technology companies have operated some of the best cash-generating businesses in the world. They could invest heavily in growth while still returning enormous amounts of capital to shareholders through dividends and share repurchases. AI is beginning to change that equation because much more of the cash those companies generate is being redirected toward physical infrastructure. 

Take Meta (formerly Facebook) as an example. It generated $31.9 billion of operating cash flow in the second quarter, while capital expenditures, including principal payments on finance leases, reached $31.1 billion. Free cash flow was just $784 million. The shift in capital allocation is equally notable. Meta did not repurchase any shares during the first six months of 2026, compared with $22.9 billion during the same period a year earlier, while issuing $24.9 billion of long-term debt.[2] 

Amazon tells a similar story from a different angle. Its trailing 12-month operating cash flow increased 33% to $161.4 billion, an impressive number by almost any standard. Yet free cash flow moved from positive $18.2 billion a year earlier to negative $7.6 billion, primarily because purchases of property and equipment increased by $66.1 billion.[3] 

These companies have not stopped generating cash. They are simply directing substantially more of it toward infrastructure. Strong earnings can coexist with increasing capital intensity, meaning a company can report excellent revenue and profit growth while the amount of cash left over after investment falls sharply. For investors, earnings alone may no longer tell the whole story. 

Contracted Demand Is Not the Same as Return

RPO has become an important number in the AI discussion because it represents revenue attached to contractual commitments that has not yet been recognized. A growing RPO balance gives investors visibility into future demand, and the balances at some of these companies are becoming enormous. 

Microsoft ended its fiscal year with $678 billion of commercial RPO, up 84% from a year earlier. Its cloud revenue reached $59.3 billion in the latest quarter, while Azure and other cloud-services revenue increased 43%. Microsoft also said nearly 90% of its annual cloud revenue came from customers outside frontier-model companies, suggesting that the demand is broader than the handful of major AI developers (OpenAI, Anthropic, etc.) generating most of the headlines.[5] 

Those are strong numbers, but RPO is not the same thing as profit or free cash flow. A contract tells us that a customer intends to buy capacity. It does not tell us what it will cost to build that capacity, what margins will ultimately look like, how much additional investment will be required, or what return the provider will earn on the capital deployed. Much of the infrastructure being built today has to be funded before the corresponding revenue arrives, so demand can be real without the economics being fully proven. 

Someone at the End of the Chain Has to Pay

This becomes clearer when we follow the AI value chain. Companies such as Nvidia provide critical hardware. Microsoft, Amazon, Google, Oracle, and others build the cloud and data-center capacity. Companies such as OpenAI and Anthropic consume enormous amounts of that compute to develop and operate AI models. But the chain does not end there. 

Ultimately, businesses and consumers have to generate enough economic value from AI to pay for everything upstream. That could come through subscriptions, token usage, enterprise software, advertising, new products, greater worker productivity, or reductions in operating costs. There are encouraging signs. OpenAI says ChatGPT has more than 900 million weekly active users, more than 50 million consumer subscribers and more than 9 million paying business users.[6] AI has clearly moved well beyond experimentation. 

The scale required from here, however, remains enormous. Bain & Company estimates that by 2030, the computing power required to meet anticipated AI demand could require approximately $2 trillion of annual revenue to support the infrastructure behind it. Even after accounting for potential AI-related savings and shifts in existing technology spending, Bain estimates an annual revenue gap of roughly $800 billion.[7] 

For context, Gartner expects worldwide IT spending across the entire technology industry to reach $6.37 trillion in 2026.[8] Those two numbers are not directly comparable, but they illustrate the magnitude of revenue the AI ecosystem may eventually need to support. The critical question is how much of that becomes genuinely new spending, rather than simply replacing existing software, cloud, labor, or technology budgets. 

What Does AI Do to the P&L?

For the enterprise customer, using more AI eventually has to show up somewhere in the financial statements. If a company increases AI spending by $100 million, what does it receive in return? Does it reduce headcount, allow the same workforce to produce more, improve margins, or create entirely new sources of revenue? Or does AI simply become another technology expense added to the existing cost base? 

That question will help determine how durable enterprise demand ultimately becomes. There is a strong bullish case. Goldman Sachs Research expects the growth of AI agents to drive a 24-fold increase in token consumption by 2030, while falling unit computing costs could improve hyperscaler economics as usage expands.[9] If AI enables companies to materially increase productivity while the cost of producing each unit of intelligence continues falling, the economics can improve from both directions. Usage rises while unit costs fall. 

In that scenario, today’s infrastructure investment may eventually look less like overbuilding, and more like companies racing to meet demand, they can barely satisfy. Microsoft already says Azure demand continues to exceed available capacity.[5] The risk is that adoption does not translate into economic value quickly enough. Enterprises may experiment extensively with AI but become more disciplined about spending if those projects fail to improve earnings, productivity, or costs. The infrastructure can be technologically impressive and still generate disappointing returns.

The Investment Boom Is Already an Economic Story

This spending is large enough that it no longer affects only technology companies. Bank of England staff estimate that U.S. investment in AI-related components grew 18% in the year through the first quarter of 2026, compared with average growth of just over 7% before the pandemic. After adjusting for the import content of AI-related goods, the Bank estimates that AI-related investment accounted for as much as one-fifth of overall U.S. GDP growth during that period.[10] 

That means the buildout is already supporting economic activity through data-center construction, computing equipment, software, power infrastructure, and related investment. If the cycle continues, the spending itself can remain an economic tailwind even before all of the downstream productivity benefits arrive. But there is another side to that equation: if the economics disappoint, businesses and consumers don’t show up with the demand, leading to companies eventually slowing the pace of investment, the adjustment would not necessarily be confined to a handful of technology stocks. It could remove a meaningful source of capital spending from the broader economy. 

The Real Test Is What the Capital Earns

I do not think the right conclusion is that AI infrastructure is obviously overbuilt. Based on currently reported demand indicators, I do not believe the available evidence supports that argument today. 

Cloud businesses are growing rapidly, capacity remains constrained, and hundreds of billions of dollars of future business has already been contracted. But revenue growth alone does not answer the question. 

Over the next several years, I will be watching how quickly AI revenue grows relative to capital expenditures, whether free cash flow begins to recover, how RPO converts into actual revenue and, ultimately, what return these companies earn on the capital they are deploying. 

There are two credible paths. In one, AI adoption accelerates, productivity improves, and declining compute costs allow revenue and margins to catch up with today’s infrastructure spending. The companies investing most aggressively could be rewarded for building capacity before the demand fully materializes. In the other, monetization develops more slowly than anticipated, capital spending remains high, free cash flow stays under pressure, and parts of the infrastructure ecosystem eventually prove overbuilt. 

We do not yet know which path will dominate, but we do know that AI is changing the financial profile of some of the largest companies in the world. Last month, my conclusion was that capital has a price again. The next question is whether the AI industry can earn enough on that capital to justify what it is spending. 

Footnotes

  1. S&P Global Ratings, “AI Infrastructure Investment To Exceed $1.3 Trillion By 2027,” August 27, 2026. S&P projects combined hyperscaler capital expenditure above $1.3 trillion by 2027 and expects debt, equity, leases and other financing structures to play a growing role in funding AI infrastructure.
    S&P Global Ratings: AI Infrastructure Investment To Exceed $1.3 Trillion By 2027
  2. Meta Platforms, Inc., Form 10-Q for the quarter ended June 30, 2026. Meta reported $31.86 billion of operating cash flow, $31.08 billion of capex including finance-lease principal payments and $784 million of free cash flow. Meta reported no share repurchases during the first six months of 2026 and $24.91 billion of net proceeds from long-term debt issuance.
    SEC: Meta Platforms Q2 2026 Form 10-Q
  3. Amazon.com, Inc., second-quarter 2026 results. Amazon reported trailing-12-month operating cash flow of $161.4 billion and free cash outflow of $7.6 billion, with the decline primarily driven by a $66.1 billion year-over-year increase in property and equipment purchases.
    SEC: Amazon Second Quarter 2026 Results
  4. Oracle Corporation, first-quarter fiscal 2027 results, September 10, 2026. Oracle reported revenue of $19.3 billion, operating cash flow of $23.1 billion, capital expenditures of $28.5 billion, and free cash flow of negative $5.4 billion. Oracle also reported roughly $20 billion of equity issuance, OCI growth of 121%, and RPO of $664 billion.
    Oracle Investor Relations: Q1 FY2027 Results
  5. Microsoft Corporation, fiscal 2026 fourth-quarter earnings. Microsoft reported $59.3 billion of Microsoft Cloud revenue, Azure growth of 43% and commercial RPO of $678 billion. Microsoft said nearly 90% of annual cloud revenue came from customers outside frontier-model companies, and that Azure demand continued to exceed available capacity.
    Microsoft Investor Relations: FY2026 Q4 Earnings Conference Call
  6. OpenAI, “Scaling AI for Everyone,” 2026. OpenAI reports more than 900 million weekly ChatGPT users, more than 50 million consumer subscribers and more than 9 million paying business users. These are company-reported figures from a private company.
    OpenAI: Scaling AI for Everyone
  7. Bain & Company, Global Technology Report, September 2025. Bain estimates approximately $2 trillion of annual revenue may ultimately be required to fund anticipated AI computing demand by 2030 and identifies an estimated $800 billion annual gap even after potential AI-related savings.
    Bain & Company: $2 Trillion in New Revenue Needed to Fund AI’s Scaling Trend
  8. Gartner, “Gartner Forecasts Worldwide IT Spending to Grow 14.2% in 2026, Totaling $6.37 Trillion,” July 27, 2026. Gartner identifies data-center systems and infrastructure-as-a-service spending as major drivers of technology spending growth.
    Gartner: Worldwide IT Spending Forecast for 2026
  9. Goldman Sachs Research, “AI Agents Forecast to Boost Tech Cash Flow as Usage Soars,” May 20, 2026. Goldman Sachs Research forecasts a 24-fold increase in token consumption by 2030 as agentic AI adoption expands.
    Goldman Sachs: AI Agents Forecast to Boost Tech Cash Flow as Usage Soars
  10. Bank of England, Monetary Policy Report, July 2026. Bank staff estimate U.S. AI-related investment grew 18% in the year through Q1 2026 and, after adjusting for import content, accounted for as much as one-fifth of overall U.S. GDP growth over that period.
    Bank of England: Monetary Policy Report, July 2026

 

Important Disclosures: The views expressed are those of the author as of the date of publication and are subject to change without notice. This material is provided for informational and educational purposes only and should not be construed as investment, legal, or tax advice. References to specific companies, securities, or industries are for illustrative purposes only and do not constitute a recommendation to buy, sell, or hold any security. Forward-looking statements and third-party forecasts are based on current assumptions and expectations and are subject to risks and uncertainties. Not all AI-related investments are likely to produce successful economic outcomes, and substantial uncertainty remains regarding long-term profitability across the industry. Actual results may differ materially from those expressed or implied. 

AI Artificial Intelligence Capital Expenditures Free Cash Flow Return on Capital Technology

Related Posts

The 5.3% Problem: What the 30-Year Treasury Is Telling Us 
September 3, 2026 Thought Leadership

The 5.3% Problem: What the 30-Year Treasury Is Telling Us 

Prev
Our Services
  • Wealth Management
  • Business Advisory
  • Asset Management
  • Who We Serve
Company
  • Our Story
  • Contact Us
  • Careers
  • CIG Video
Insights
  • Newsroom
  • Resources
  • Governance
Address
1 Town Center
Suite 1850
Southfield, MI 48076

© 2026 CIG Capital Advisors

Web Design & Development By SPARK Business Works

Securities Offered Through CIG Securities, Inc., Member FINRA/SIPC
Investment Advisory Services Available Through CIG Asset Management, Inc., 
SEC Registered Investment Adviser.
Insurance Services Provided by CIG Risk Management, Inc.
All Are Wholly Owned Subsidiaries of CIG Capital Advisors, Inc. 
“Elevate Your Success” and “Plan to Succeed” are registered trademarks of CIG Capital Advisors. All rights reserved.
Please read the Terms of Use and Disclosures.
Click here for CIG Asset Management Form CRS or  CIG Securities Form CRS.
Check the background of CIG Securities, Inc. or any of its Investment Professionals on FINRA’S BrokerCheck site at:  http://brokercheck.finra.org/