Can the AI Investment Boom Sustain its Current Pace?

Corrado Tiralongo - Nov 17, 2026

AI investment continues at a historic pace, but sustaining returns will require much stronger customer spending. Adoption, pricing power, and cash generation remain critical tests.

The AI investment boom has enough momentum to carry into 2027. Projects are underway, and the largest cloud providers have committed to buying equipment, power and computing capacity. Sustaining the spending envisaged through 2030 will be harder. We believe revenues from customers will struggle to grow quickly enough to justify several more years of investment at that scale.

In our November 9 commentary, we described the broader market sell-off as a correction and flagged the risk of a deeper downturn in AI investment. In an illustrative Capital Economics scenario, sustaining the build-out would require roughly US$1.8 trillion in annual AI revenue from end customers by 2030. (1)

What it takes to earn a return

J.P. Morgan Asset Management puts combined capital spending by Amazon, Alphabet, Microsoft, Meta and Oracle at close to US$800 billion in 2026 and more than US$1 trillion in 2027, based on analyst consensus. These are estimates of total capital spending. Goldman Sachs Research estimates that most of the increase since 2022 has gone to AI projects. (2)

Capital Economics illustrates what several more years at that pace would imply. Its scenario assumes roughly US$1 trillion of cumulative AI-related capital spending by the end of 2026 and another US$1 trillion in each of the following four years, bringing the total to US$5 trillion by the end of 2030. With an average economic life of five years and depreciation spread evenly over that life, the infrastructure would have a remaining book value of about US$3 trillion in 2030. (1)

A 10% to 15% operating return on that capital would require US$300 billion to US$450 billion in annual operating profit. Allow for about US$1 trillion in annual depreciation and cash operating costs equal to 25% of sales, and the required annual revenue rises to roughly US$1.7 trillion to US$1.9 trillion. For scale, The Economist put identifiable AI revenue at roughly US$150 billion to US$220 billion on an annualized basis in mid-2026. Definitions and measurement dates vary, so the comparison illustrates the scale of the challenge with considerable uncertainty about the precise gap. (1)(3)

Amazon depreciates servers and networking equipment over five to six years. Alphabet generally uses six years for that equipment and seven to 40 years for datacentre and office buildings. (4) Actual economic lives depend on equipment performance and replacement costs.

Faster obsolescence would leave less time to recover the investment, raising the revenue hurdle.

Where would the revenue come from?

AI providers can earn revenue by capturing a share of the value created by higher productivity, winning spending that now goes to software and business services and charging customers for tools that reduce labour costs.

Capital Economics assumes AI raises global productivity by about 1% by 2030 and the major cloud providers capture a quarter of the additional value added. Under its assumptions, that yields roughly US$700 billion of additional annual revenue, leaving about US$1 trillion to US$1.2 trillion to come from existing software and services budgets and labour savings. The illustration therefore requires substantial economic gains, with a large share flowing to AI providers. (1)

Boston Consulting Group sees a credible path to paying for the build-out. Its model covers the broader AI ecosystem, including hyperscalers’ own use of AI, and estimates an annual requirement of about US$2.6 trillion by 2030. Of that, enterprises would cover about US$1.8 trillion, roughly half by redirecting existing technology budgets and half through additional spending. (5)

How much providers can charge will also matter. A company using AI to improve its margins may retain much of the gain. Stanford’s AI Index documents steep declines in the cost of running models at comparable performance. If those savings flow through to customers, usage could grow quickly while revenue grows more slowly. (6)

Revolut offers a practical example. In October, its AI operations lead described how the company made a lower-cost, self-hosted model the default for coding and kept commercial frontier models for selected tasks. (7) That approach could pressure model providers’ pricing, although self-hosted models still require computing infrastructure. Lower prices could also encourage enough additional usage to increase total spending. The outcome depends on how strongly customers respond to lower prices.

We see room for businesses to redirect existing budgets and commit new money to AI. But doing so at the required scale depends on changes to workflows, reliable results and a willingness to pay. Our concern is the timing. The investment plans leave little room for adoption or customer spending to disappoint.

What earlier buildouts tell us

Capital Economics reports that overall investment fell by nearly half in nominal terms after Britain’s railway boom ended in the late 1840s. After the dotcom bust, U.S. spending on information processing equipment fell 24% in dollar terms and 13% after adjusting for prices.

Some of the excess telecom capacity was eventually absorbed as demand caught up. The networks remained useful long after the initial rush to build them had passed. (8)

Some AI infrastructure could follow that path. Data centres and power connections can retain value while demand catches up. Specialized chips have a shorter window to earn back their cost. If they sit underused for several years, they may need replacing just as demand improves. That makes the timing of adoption particularly important to the returns on today’s equipment. (8)

How a slowdown would spread

Physical constraints could also slow the build-out while customer demand remains strong. Capital Economics identifies shortages of advanced chips and electricity, alongside local opposition over land, water and power use. Delays in approvals or grid connections could raise costs and postpone revenue from new capacity. (1)

With these constraints in view, Capital Economics projects slower spending growth in 2027 and a decline of about 10% in 2028. We see that as a plausible scenario. Replacement demand and commitments on projects already underway could make the adjustment more gradual. Continued investment and improvements in computing efficiency could support wider adoption even as spending falls. (1)

Anthropic’s November 10 prospectus shows how far ahead some companies have booked capacity. It expects to incur roughly US$452 billion of purchase and lease commitments by the end of 2030, mostly for computing capacity and infrastructure. Those commitments support infrastructure demand as capacity comes online. They can also put pressure on cash flow if customer spending falls short. (9)

Even before spending falls, slower investment growth can reduce its contribution to economic growth. Morgan Stanley Research estimates AI-related spending added about 0.3 percentage points to U.S. real GDP growth in 2025 after adjusting for imports. (10) Taiwan’s statistics agency and South Korean trade data show how strongly AI demand has supported hardware exports. Capital Economics estimates that reduced AI demand could lower 2027 growth relative to recent rates by about three percentage points in Taiwan and one in South Korea. (1)(11) Geographically diverse equity holdings can still depend on the same spending decisions.

The effects could also reach copper. Prices could weaken as investors reassess AI demand, followed by pressure on physical consumption if infrastructure spending falls. Capital Economics also attributes part of copper’s strength to tariff uncertainty, which has encouraged U.S. stockpiling and tightened availability elsewhere. Price movements therefore reflect several forces beyond AI demand. (12)

Within the AI investment chain, the effects would differ. A large cloud platform that trims expansion and keeps its customers could improve free cash flow, while its equipment suppliers lose orders. A heavily financed developer could face weaker demand just as refinancing becomes more expensive.

Financing will help determine how far those pressures spread. The Bank for International Settlements documents a growing reliance on long-term debt and financing structures that sit off company balance sheets. Meta’s Hyperion project shows how that can work, combining a joint venture with Blue Owl, operating leases and privately placed debt. As these arrangements spread, more lenders become exposed to the same spending cycle. (13) Capital Economics expects a spending adjustment with limited financial spillovers if the boom turns soon. Risks increase if expansion continues with more debt. Falling equity wealth could also weaken consumption. (1)

What matters for Canada Life Investment Management Ltd. Asset Allocation portfolios

Our asset allocation portfolio approach is to participate in AI’s benefits through a wider set of businesses and sources of return. We see opportunities in companies that can show how AI improves productivity, margins and cash generation without committing comparable amounts to infrastructure. Spending on software and applications that help businesses adopt AI may prove more durable than spending on new hardware.1

For Canadian investors, copper shows how resource equities can share exposure to the AI investment cycle. Our multi-factor framework helps diversify equity exposures, alongside attention to these shared sources of risk.

The boom becomes more sustainable if paying demand catches up with installed capacity and companies generate more cash after investment. More disciplined infrastructure spending could improve returns while adoption continues. Weak demand alongside tighter funding would be more damaging. We are watching customer spending, capacity utilisation and cash generation to judge which adjustment is taking shape.

Sincerely,

Corrado Tiralongo Signature

Corrado Tiralongo (he/him)
Vice President, Asset Allocation & Chief Investment Officer
Canada Life Investment Management Ltd.

Source notes:

1. Capital Economics, Vicky Redwood, How sustainable is the AI investment boom?, Global Economics Focus, November 6, 2026, pp. 2–8. The US$5 trillion investment scenario, resulting revenue hurdle, illustrative revenue breakdown and slowdown estimates come from this analysis.

2. J.P. Morgan Asset Management, Artificial Intelligence, “Financing the AI Buildout”, consensus data as of August 31, 2026. Goldman Sachs Research, Global AI Investment Is Forecast to Exceed $1 Trillion in 2026, August 7, 2026.

3. The Economist, AI revenues are growing fast, but not fast enough, July 2026, republished by Mint.

4. Amazon, 2025 Form 10-K, and Alphabet, 2025 Form 10-K, accounting policies for property and equipment.

5. Boston Consulting Group, Who Will Pay the $2.6 Trillion AI Infrastructure Bill?, October 29, 2026.

6. Stanford Institute for Human-Centered Artificial Intelligence, AI Index Report 2025, technical performance and inference costs.

7. David Abitbol, Revolut, Shipping like a startup, regulated like a bank, AI Engineer NYC, October 13, 2026.

8. Capital Economics, How sustainable is the AI investment boom?, November 6, 2026, pp. 5 and 8.

9. Anthropic, PBC, Form S-1/A, Amendment No. 2, November 10, 2026, “Contractual Obligations.”

10. Morgan Stanley Research, 2026 U.S. Economics Outlook, “AI Spending: Mostly Intermediate Goods and Imports.”

11. Taiwan Directorate General of Budget, Accounting and Statistics, GDP: Preliminary Estimate for 2026Q2 and Outlook for 2026–27. Reuters, South Korea exports nearly double in October on AI investment boom, November 1, 2026.

12. Capital Economics, Kieran Tompkins, Copper’s rally faces a tariff and AI reality check, Commodities Update, November 12, 2026, pp. 1–2.

13. Bank for International Settlements, Building AI on credit: structures and spillovers, October 27, 2026. Meta, Meta Announces Joint Venture With Funds Managed by Blue Owl Capital to Develop Hyperion Data Center, October 21, 2025.
The views expressed in this commentary are those of Canada Life Investment Management Ltd. as at the date of publication and are subject to change without notice. This commentary is presented only as a general source of information and is not intended as a solicitation to buy or sell specific investments, nor is it intended to provide tax or legal advice. Prospective investors should review the offering documents relating to any investment carefully before making an investment decision and should ask their financial security advisor for advice based on their specific circumstances.
This material may contain forward-looking information that reflects our or third-party current expectations or forecasts of future events. Forward-looking information is inherently subject to, among other things, risks, uncertainties and assumptions that could cause actual results to differ materially from those expressed herein. These risks, uncertainties and assumptions include, without limitation, general economic, political and market factors, interest and foreign exchange rates, the volatility of equity and capital markets, business competition, technological change, changes in government regulations, changes in tax laws, unexpected judicial or regulatory proceedings and catastrophic events. Please consider these and other factors carefully and not place undue reliance on forward-looking information. The forward-looking information contained herein is current only as of November 16, 2026. There should be no expectation that such information will in all circumstances be updated, supplemented or revised whether as a result of new information, changing circumstances, future events or otherwise.
Investment Planning Counsel and the Investment Planning Counsel logo are trademarks of Investment Planning Counsel Inc.