Capex Supercycle: What the Fiber Overbuild Teaches Advisors
Capex Supercycle: What the Fiber Overbuild Teaches Advisors
Clients keep asking a version of one question: is today's AI spending a durable shift or a replay of 2000? What advisors can do is study how the last capex supercycle unfolded, identify which signals appeared before the turn, and build a monitoring process that doesn't depend on narrative.
What Is a Capex Supercycle?
A capex supercycle is a multi-year stretch in which an entire sector raises capital spending well ahead of realized demand. It is justified by forecasts, partly debt-financed, and reinforced by competitors who fear falling behind.
Gartner estimates that AI providers will spend nearly $1 trillion in 2026 and a further $6 trillion from 2027 to 2030, extending hyperscaler spending trends many portfolios already carry.
The Telecom Fiber Overbuild: How the Cycle Unfolded
The late-1990s telecom fiber overbuild is the closest modern template. According to the Federal Reserve Bank of San Francisco, capital spending by publicly traded telecom service firms rose from $47 billion in 1995 to a $121 billion peak in 2000, then fell to $49 billion by 2002.
The Buildout and the Dark Fiber Glut
Buildout: Telecom firms laid about $90 billion of fiber-optic cable between 1997 and 2001, per the Dallas Fed.
Efficiency shock: Wavelength division multiplexing let each strand carry far more traffic, multiplying effective capacity. Efficiency can expand demand, as the Jevons paradox suggests, but it expands supply first.
Glut: The FDIC reported in 2002 that only about 2% to 5% of U.S. fiber was actually lit. The rest was dark fiber.
Collapse: Communications equipment orders fell from roughly $13.3 billion in June 2000 to $3.6 billion by September 2001, per the same FDIC report.
Demand did not vanish: worldwide telecom revenue grew roughly 8% in 2001 (Dallas Fed). Capacity outran it.
Who Captured Returns and Who Absorbed Losses
Equipment makers: The S&P communication-equipment index fell 86% between March 2000 and September 2001, erasing $793 billion in value (Dallas Fed).
Lenders: Telecom accounted for 25.8% of corporate bond default dollar volume in 2001, more than any other industry (FDIC).
Later users: Returns shifted from builders toward whoever used the capacity after the reset.
Dot-Com Bubble Lessons for AI Infrastructure Spending
The San Francisco Fed research found little evidence of economy-wide overinvestment in the late 1990s. Expectations ran too high mainly in two places: telecom services and dot-com firms. For AI infrastructure spending today, the better question is where any excess is concentrated.
Where the Parallels Hold
Spending is justified by long-dated demand forecasts rather than current revenue.
Newer, more leveraged entrants are building alongside incumbents.
Gartner expects over 80% of neocloud providers to stop operating independently by 2030.
Where Today's Cycle Differs
Funding: Many 1990s carriers relied on debt. By September 2001, the industry carried a CCC+-equivalent rating, per the Dallas Fed. Gartner considers today's hyperscalers better positioned because of their diversified revenue, though their rising bond issuance deserves attention.
Asset life: Dark fiber kept its option value and could be lit when demand caught up. AI chips face a shorter replacement cycle, and Gartner warns many compute providers may not recoup spending before the current chip generation is replaced.
Capital Cycle Analysis: Signals Advisors Can Monitor
Capital cycle analysis, the supply-focused framework popularized by Marathon Asset Management and reviewed by the CFA Institute, starts from a simple premise: returns tend to fall when capital floods into an industry and recover when it leaves.
Capex to Revenue Ratio and Overinvestment Cycle Warnings
Capex to revenue ratio: Rising capital intensity without matching revenue growth is the core supply signal.
Credit spreads: In the last cycle, lenders absorbed a large share of the damage. Spreads on the most leveraged builders are worth tracking, particularly when credit and equity markets disagree.
Pricing and utilization: Falling rental or usage prices suggest supply is outrunning demand.
Supplier orders: Equipment orders turned sharply in 2000 and 2001.
Impairment risk: Gartner says over $3.5 trillion of AI infrastructure spending could be exposed to impairment by the end of 2030 if the current pace continues.
No single overinvestment cycle signal is decisive; a cluster moving together says far more.
Is There an AI Capex Bubble? Framing It for Clients
Treat it as a monitoring exercise, not a prediction.
Separate builders, financiers and end users; each carries different risk.
Define in advance which signals would prompt a portfolio review, the same discipline behind systematic sell signals.
Conclusion
The 1990s capex supercycle did not end because the internet was a fad. It ended because capacity, financing and expectations ran ahead of revenue, a reminder that great technology can still be an awful investment. Advisors who set clear rules before the next capex supercycle peaks keep their process ahead of the narrative.
Turn Your Capex Cycle Framework Into Rules With Surmount
Surmount Wealth lets advisors turn a thesis like this into a rules-based strategy, test it on historical data, and automate it on existing brokerage accounts.
Hypothetical illustration: the "Capex Cycle Monitor"
Illustrative concept only. Not an existing Surmount strategy, not backtested, and no performance is implied.
Supply signal: Tracks capital intensity across a user-defined basket of infrastructure builders, reviewed monthly.
Credit trigger: Trims builder exposure toward a broad market ETF or cash equivalents when a high-yield spread measure crosses a user-set threshold.
Value-chain tilt: Leans toward capacity users when builder capital intensity is above its historical range.
Limitations: thresholds are user-set; signals can lag or misfire; outcomes vary with costs, taxes and account constraints.
Why advisors explore Surmount:
Turn any documented thesis into automated rules
Prebuilt strategy library or fully custom logic
Works with existing brokerage accounts, no asset transfers
No coding required
Rules-based strategies carry model risk and do not remove market risk.
Book a demo to see how your capex cycle framework could run as rules.
For informational and educational purposes only; not investment advice. Historical patterns do not guarantee future outcomes.
FAQ: Capex Supercycle
What Defines a Capex Supercycle?
A multi-year stretch when an entire sector raises capital spending well ahead of realized demand, typically justified by forecasts and financed partly with debt.
Why Did the Telecom Fiber Overbuild Happen?
Carriers built on aggressive demand forecasts while new technology multiplied each strand's capacity. By 2002, only about 2% to 5% of U.S. fiber was lit, per the FDIC; the rest was dark fiber.
When Does a Capex Supercycle End?
Historically, when capacity, financing and expectations run ahead of revenue. The turning point is usually clear only in hindsight, which is why predefined signals matter.
How Does Capital Cycle Analysis Work?
It focuses on supply rather than demand. Returns tend to fall when capital floods into an industry and recover when it leaves.
Is AI Infrastructure Spending a Bubble?
That can't be known in advance. Advisors can track the capex to revenue ratio, credit spreads and impairment risk rather than relying on forecasts.
