Monitoring AI Equity Market Risk Through Data-Driven Intelligence

I invest my own money in AI and semiconductor stocks. I built this to measure the risk in that exposure instead of guessing at it — a multi-factor read on the equities across the AI supply chain, run the same way every trading day, published in full. I am not a financial advisor: this is research, not investment advice.

The objective is not to predict markets, but to identify evolving risk conditions before they become widely visible.

See today’s reading → Explore the methodology

Risk regime monitor

CRITICAL
Critical regime · Level 3
NORMALALERTCRITICAL
0%
P(–30% / 90d)
0%
Drawdown SOXX
0
families under stress
Last completed run · close of 2 September 2026 · updated every trading day
SCROLL
Live risk data

The market, measured

Where the numbers stand at the last completed run — the close of 2 September 2026. Every figure traces back to a primary source (Polygon, FMP, Alpha Vantage APIs), each carrying its instrument, date and origin. Nothing is quoted from memory: every value is auditable.

0
Estimated probability of a ≥30% SOXX decline within 90 days, from the cycle peak
▲ from 27% (prior run)
0
SOXX drawdown from the cycle peak (655.95), the benchmark I measure everything against
−20% threshold crossed
0
Consecutive sessions at internal alert level 3 — a level only steps down once a full recovery checklist is met
level 3 holding
0
Signals in the highest-risk state: S2 structure
elevated risk

Evolution of the probabilities

Downside risk by time horizon

SOXX vs drawdown thresholds

Closing price and the published drawdown scenarios

Two more charts: volatility & credit, drawdown vs NVDA

Volatility & credit

Market volatility and high-yield credit conditions

SOXX vs NVDA drawdown

Distance from the 52-week peak

Risk signals

The 9 indicators, current reading

Hover over each card. The LED colour reflects the current state; the bar shows the stress level. Each indicator moves along rules fixed in advance — no discretionary overrides, no case-by-case judgement.

A · Trend & priceB · ParticipationC · Volume D · VolatilityE · CreditF · Euphoria
The framework

A transparent, multi-factor framework

Sector risk is decomposed into 9 indicators across 6 signal families, each moving along rules fixed in advance and applied without discretion. From the market close to a one-line verdict, the same fourteen steps run every trading day: ingest primary data, measure the nine signals, combine the families into a regime, estimate the probabilities, archive the result. Thresholds, weights and calibration values stay proprietary.

57
Universe — AI-related companies across the AI supply chain (semiconductors, hardware, hyperscalers, software, infrastructure).
9
Indicators — each moving from low to elevated risk by rules set in advance, never case by case.
6
Signal families — price trend, market breadth, volume, volatility, credit conditions, euphoria/fragility.
3
Risk regimes — NORMAL → ALERT → CRITICAL, driven by how many signal families are under stress at once, with three alert levels (Watch / Alert / Critical).
4
Calibrated-probability scenarios — D10 (−10%), D20 (−20% in 30d), D30 (−30% in 90d), SYS (S&P −20% with AI at the epicenter).
SOXX
Sector benchmark — flagship NVDA, plus a basket of high-beta names used as early-warning “canaries”.
1
De-escalation path — risk can jump several steps up in a single day, but steps down one level at a time, and only once recovery conditions hold for a sustained stretch of sessions.
v5.20
Methodology as the single source of truth. Prudential asymmetry only on the action thresholds, never on the probabilities.
Read the methodology → Read the daily notes →
Historical memory

The mirror of 2000

The NASDAQ lost 77.9% in 31 months. At the peak the VIX was low and complacent: the same profile the system watches today.

NASDAQ 1999–2002 · the anatomy of a bubble

From the 5,048 peak (Mar 10, 2000) to the 1,114 low (Oct 9, 2002)

At the dot-com peak (Mar 2000)
Extension above the 200-day MA+54.9%
VIX~21–24 (complacent)
Outcome at 31 months−77.9%
Current profile (2 Sep 2026)
Extension above the 200-day MA+16.8%
VIX15.20 (below 20)
Analogous phasetop, not panic
Frequently asked questions

Questions, answered

The essentials on what AI Market Observatory is, how it works, and what it is not.

What is AI Market Observatory?
It is an independent quantitative research platform that monitors AI-related equity market risk every trading day. It decomposes risk into 9 indicators grouped in 6 signal families and produces a risk regime and four calibrated probability scenarios. It is for informational purposes and does not provide investment advice.
Is this investment advice?
No. AI Market Observatory publishes independent quantitative research on AI-related equity market risk. It does not provide investment advice, recommendations or guarantees of future market outcomes. Any decision remains the reader's own.
How are the probabilities computed?
A superforecaster engine starts from a historical base rate and applies adjustments justified line by line, then runs sanity checks against historical anchors such as the dot-com cycle. The adjustment weights and base rates are proprietary. Probabilities stay calibrated: prudential asymmetry lives only on the action thresholds, never on the probabilities.
What are the alert regimes?
Three regimes: NORMAL, ALERT and CRITICAL, driven by how many signal families are under stress at once. The system can jump several steps up in a single day but steps down one level at a time, and only after a full de-escalation whose conditions must hold for a sustained stretch of sessions.
What data and sources does it use?
Prices and macro come from primary sources via API (Polygon, FMP, Alpha Vantage); every value has instrument, date and source tracked. The sector benchmark is SOXX, with NVDA as flagship and a basket of high-beta names used as early-warning canaries. The specific series, proxies and endpoints behind each signal are not published.
Is the research independent?
Yes. There is no affiliation, no product to sell and no position to talk up. The process, the sources and every figure published are documented and auditable, and the rules are fixed before they are needed — only the calibration parameters stay proprietary. The framework runs every trading day, so the published read is always the current one, and conclusions follow measurable indicators rather than narrative.
Who I am
Retail investor · AI practitioner

Andrea Pilotti

I invest my own money in AI and semiconductor stocks, and I work with artificial intelligence for a living. This observatory exists because I needed to measure the risk in my own exposure instead of guessing at it — and nothing I could find did that transparently, every day, in language a non-professional can check. So I built it, and I publish it as it is. Full bio on LinkedIn →

The daily run is carried out by eleven specialised AI agents I engineered — data ingestion, quantitative analysis, probability calibration, risk management, quality assurance and editing, each owning one dimension of the process. They are not a company standing behind me: they are what my day job looks like pointed at my own portfolio. No single output is a forecast, and nothing goes out until I have reviewed and signed it off myself. See the fourteen steps of the daily run →

And what I am not. I am not a financial advisor and I am not licensed. I never say what to buy, sell or hold — I publish measurements, not instructions — and nothing here is investment advice. Every decision is yours, ideally with professional guidance.

RESEARCH FRAMEWORK · 2026

Methodology, universe, indicators and model limitations

The full documentation of how the framework reads AI-related equity market risk, kept current with the methodology it describes.

Read the methodology →
About the figures. The data on this page comes from the last completed daily run (close of 2 September 2026) and is updated at every run; it is not a real-time feed. What this is not. I am not a financial advisor and I am not licensed. This observatory produces alerts for informational purposes only — never trading recommendations, and none of it is investment advice. The goal is risk management, not returns. The probabilities are calibrated model estimates, not certainties. Every investment decision remains yours, ideally with professional guidance.