AI Bubble in the cycle analyzer | Market Cycles Report September 14 2026

Finance-specific summary (AI “bubble” via cyclical analysis)

The presenter argues there is an “AI bubble” observable in markets and demonstrates how to analyze it using a cyclical scanner / loop analyzer framework.

Rather than relying on major capitalization-weighted indices (S&P 500, Nasdaq), the method builds a custom, balanced composite designed to represent the AI investment theme and make it more suitable for cycle detection.

---

Instruments / tickers / sectors mentioned

No individual ticker symbols are explicitly listed in the subtitles. However, the custom basket is described as including:

• Chips / semiconductors
• AI infrastructure
• Hyperscalers
• Specialized software companies

Examples are referenced generically (not as tickers in the subtitles), such as “CoreWeave” and “Arm.”

For comparison, the presenter mentions these indices:

• S&P 500
• Nasdaq

---

Key methodology / step-by-step framework

1) Build a dataset / composite

• Create a custom AI composite because there is no dedicated AI composite index available.
• Select 24 (or 24–25) U.S. stocks whose prices are driven mainly by the AI hype/investment cycle (chips, infrastructure, hyperscalers, specialized software).

2) Balance and rebalance

• Use an “equilibrium” weighting approach to reduce capitalization bias (e.g., smaller companies can receive weight similar to large names as described).
• Rebalance daily using each constituent’s daily returns.
• Normalize the composite to a base value of 100 at inception.

3) Use “honest” historical constituents

• Include only companies that were actually available at the time (no backward “theoretical adjustment”).
• Address survivorship/listing timing (the subtitles reference that some firms weren’t traded throughout the entire sample, e.g., examples like CoreWeave / Arm).

4) Define the analysis window

• Start in 2022, covering 4 years / 1,000+ daily metrics.
• Rationale: align with the launch of ChatGPT (October 2022), which the presenter ties to the bubble’s “birth.”

5) Run cyclical analysis (daily)

• Feed the composite time series into a fund cycle analyzer / cyclic analyzer.
• Conduct spectral analysis to find dominant cycles.
• Perform walk-forward / out-of-sample testing:
  • Remove the last 6 months to test whether the cycle logic would have produced similar signals without seeing the future.

6) Use a “Cyclical Consensus Index”

• Identify multiple cycles from the spectrum (not just one).
• Rank cycles by strength and phase (mentions Bartels index and cycle “power/strength/severity”).
• Split cycles into bullish vs bearish camps based on phase and compute whether conditions are more tailwind or headwind.

7) Add higher timeframe context (weekly)

• Repeat the procedure on weekly data to check coherence within cycles.
• Combine daily cycle pressure with a dominant weekly cycle to assess the broader window (the presenter frames this without a “burst timing” claim).

---

Key findings and numbers (cycles, timing, and conclusions)

Daily composite / bubble top

• The constructed AI composite outperformed the S&P 500 starting late 2022 (as shown in the presenter’s chart comparison).
• The composite’s main peak / historical high occurred in late May to early June (2026).
• The presenter claims this timing suggests the bubble’s top had already occurred by then.

Dominant cycle (daily)

• Dominant spectral cycle length: ~190 days
  • Described as highly stable and having the second strongest power rating, with the highest amplitude and top ranking on “first rating” (per subtitles).
• Walk-forward test (reference to March 2026):
  • Even then, the dominant 190-day cycle indicated an uptrend phase lasting ~2–3 months (framed as cyclical perspective, not a guaranteed price outcome).
• Cycle variability:
  • By June 2026, the daily cycle length is described as 188 days, indicating cycles can vary around a nominal length rather than staying perfectly constant.

Consensus-based directional bias (daily)

• After the June peak period, the consensus shifts to strong headwinds in the cyclical state.
• The presenter also suggests the uptrend framework implies a year-end decline for the AI composite indicator when overlaying the cycles.

Secondary daily cycle simplification

• The presenter simplifies attention to 119-day and 80-day cycles:
  • They’re described as closely linked.
  • Point to a “double top.”
  • Their subsequent lows are said to synchronize in late November.
• A further synchronization is expected in November (framed as cyclical rhythm rather than a precise prediction).

Weekly context

• Weekly data shows a dominant cycle of 41 weeks.
• Weekly interpretation:
  • The underlying trend remains up (bubble trend), while cycles oscillate within it.
  • As of “today” (in the narrative frame), the 41-week cycle is in a downward phase toward December.
• Integration of daily + weekly:
  • The presenter claims synchronized daily and weekly lows toward year-end.
  • Concludes: over the next ~3 months, there is a strong headwind from dominant cycles (narratively anchored as beginning early 2022).

---

Explicit recommendations / cautions / risk framing

• No price forecast claim The presenter repeatedly clarifies that cycle outputs are not a trading system and not a guaranteed price forecast.

• Cycles are conditional information Cycles are framed like economic/technical context—use them as inputs, then make final trading and risk management decisions separately.

• User responsibility The presenter emphasizes that responsibility for using the analysis lies with the user.

• No guarantee of repeatability Even if cycles looked noticeable and predictable historically, there is no guarantee the pattern repeats exactly.

---

Disclosures / disclaimers found in subtitles

• “Cycles are not a trading system.”
• “Cycles give us some information about what to expect… you need to combine to make final trading or risk management decisions… never the only tool.”
• “This is not a price forecast.”
• Responsibility: “the responsibility of the one using this analysis.”

---

Presenters / sources

• Presenter: Lars Fontenen (host of “Cycles TV”)
• Sources / references mentioned: ChatGPT (launch timing referenced), and a New York conference where related work was reportedly presented.