Everyone debates whether AI models pass benchmarks — this page trades what actually moves the AI trade first: how much public attention ChatGPT, OpenAI and AGI are getting right now, and whether NVDA, semis and the AI tokens are still following it.
Readings as of 08 September 2026 · Updated 09 September 2026 · 11:11 UTC77,355 combined Wikipedia reads a day across ChatGPT, OpenAI, AGI and Nvidia — 10th percentile of the two-year window, -0.15σ over the last 30 days. Coverage this run: 4 of 4 articles.
No active divergence: attention's 30-day move is -0.15 standard deviations and NVDA is 4.1% under its one-year high. The warning arms when attention falls more than 0.4 standard deviations in a month while NVDA holds within 8% of its high.
The link chart loads when this section is visible. If it does not: attention reads -0.95σ (QUIET) as of 08 September 2026, NVDA's 30-day move is +14.9%, and the AI-token basket's is +24.4%.
The question this chart answers: is the crowd's curiosity still confirming the price? When the violet line rolls over while the green line holds the top of its range, the story is being sold into strength — the exact setup the warning card watches for.
Cross-correlation between daily changes in the attention index and daily returns, shifted ±20 trading days. A positive lag means attention moves first. The honest answer is printed even when it is zero — a claimed lead that is not in the data would be marketing, not research.
Lead-lag bars load with the chart library.
Attention → NVDA: strongest at lag +2 days (attention leads NVDA), correlation -0.107 across 497 paired days.
Attention → AI-token basket: strongest at lag +14 days (attention leads the basket), correlation -0.091 across 364 paired days.
The thesis is a chain: public attention → narrative momentum → the listed supply chain re-rates → the AI tokens follow with a lag. Each row is one link, measured on its own terms, dated, with its two-year percentile where one exists. A missing link is shown missing — never zero.
| Link | Reading | Percentile | 90 days | How to read it |
|---|---|---|---|---|
| ChatGPT attentionIs the crowd still looking? | 48,365/day | 5th | 7-day average Wikipedia pageviews, 2-year window | |
| AGI attentionIs the endgame story spreading? | 10,682/day | 99th | 7-day average Wikipedia pageviews, 2-year window | |
| OpenRouter catalogueIs developer supply still expanding? | 431 models (+106 in 90d) | n/a | — | listed models; rankings feed is not public, so this is supply, not usage |
| NVDA 30-dayIs the supply chain re-rating? | +14.9% | 78th | 30-day price move, percentile vs its own 2-year distribution of 30-day moves | |
| SMH 30-dayIs it broader than one stock? | +4.6% | 51th | semiconductor ETF, same 30-day reading | |
| AI-token basket 30-dayDid the story reach the tokens? | +24.4% | n/a | equal-weight FET/RENDER/TAO, current + 30d window (history limited to 365d by the free API) |
As of 08 September 2026, public attention on AI sits at -0.95 standard deviations versus its own two-year norm — regime QUIET for 39 days — and it is cooling (-0.15 over 30 days). Over the same month NVDA moved +14.9% and the AI-token basket +24.4%.
The divergence alarm is not armed. Attention and price are not yet telling different stories loudly enough for this page to call a distribution pattern.
Honesty check on the lead: over the full window, strongest at lag +2 days (attention leads NVDA), correlation -0.107 across 497 paired days. If that lag is near zero, attention and price are moving together here — the edge is in watching for the gap to open, not in assuming it is always open.
If the attention index prints a new two-year high (above the current peak of +8.18, set 09 Jun 2025) while NVDA fails to make a new 60-day high within the following month, the narrative is exhausted: maximum curiosity, no new buyers. The thesis also dies if the attention floor stops falling between hype waves — a permanently elevated baseline would mean attention has decoupled from the trade.
As of 08 September 2026, the AI attention cycle sits at -0.95σ — regime quiet, 39 days running — after peaking at +8.2σ on 09 Jun 2025. The divergence alarm is off: attention's 30-day move (-0.15σ) and NVDA's distance from its high (4.1%) are not yet telling different stories. Over the same month NVDA moved +14.9% and the AI-token basket +24.4%. What to watch next: whether attention keeps sliding while price holds — if the gap widens, the distribution warning is doing its job. What would prove the thesis wrong: a new two-year attention high (above +8.18σ) that NVDA refuses to follow with a new high of its own — maximum curiosity with no new buyers.
Every answer is composed from the dated readings printed above. If a question asks for a prediction, a target or advice, the page refuses in plain words.
Benchmarks tell you what the models can do; attention tells you what the crowd is about to pay for the story. The AI trade is a narrative trade, and narratives run on curiosity. On 08 September 2026 the attention index reads -0.95 standard deviations versus its two-year norm — that is the input benchmarks never see.
Daily Wikipedia pageview counts for ChatGPT, OpenAI, Artificial general intelligence and Nvidia are summed into one composite, smoothed to a 7-day average, and z-scored against the full two-year window. A score of 0 is a normal day; +2 is twice the typical spread above normal.
They are free, public, daily, and hard to fake at scale — a direct count of people going to read about AI, not a survey or an engagement-weighted feed metric. The trade-off: Wikipedia skews to research curiosity, so the page treats it as a proxy, not the thing itself.
They are bands on the z-score: Quiet below 0, Buzz from 0 to +1, Frenzy from +1 to +2, Euphoria above +2. A hysteresis band stops the label flickering on one noisy day — it takes more evidence to enter a hotter regime than to stay in it.
Attention falling more than 0.4 standard deviations in 30 days while NVDA holds within 8% of its one-year high. Price holding a story the crowd has stopped checking is what distribution looks like. It is not active right now.
On this page's own test, strongest at lag +2 days (attention leads NVDA), correlation -0.107 across 497 paired days. The correlation curve over ±20 days is printed in full, because an honest zero is more useful than a claimed lead.
OpenAI is private — there is no price to chart. So the page works in proxies: public attention, the developer catalogue, and the listed supply chain (NVDA, the semiconductor ETF, the AI tokens). A proxy is a stand-in, and the page labels it as one.
Equal-weight FET, RENDER and TAO — tokens whose prices trade on the AI story. The free data window covers 365 days of history, so longer windows on the chart show attention and equities only. That boundary is stated, not hidden.
If the attention index prints a new two-year high (above the current peak of +8.18, set 09 Jun 2025) while NVDA fails to make a new 60-day high within the following month, the narrative is exhausted: maximum curiosity, no new buyers. The thesis also dies if the attention floor stops falling between hype waves — a permanently elevated baseline would mean attention has decoupled from the trade.
No. Euphoria readings can run for months while price rises with them. The trade signal is not the level — it is the gap: attention rolling over while price holds the high, or new attention highs that price refuses to follow.
Once a day, automatically. The slowest required input sets the page date, and every figure carries its own observation window.
The number of models listed on OpenRouter's public catalogue and how many were added in the last 90 days — a developer-supply proxy. Their usage-rankings feed is not exposed without a key, so this page measures supply, not usage, and says so.
Because the crowd reads about the stock as part of the story. Including it also makes the test fairer: if attention only tracked the product articles, a link to NVDA would be assumed rather than examined.
Zero means attention is at its two-year average. +1 means one typical spread above it. The current reading is -0.95. About 16% of days sit above +1 in a normal distribution, so Frenzy and Euphoria are meant to be rare.
Google Trends weekly, ChatGPT search interest, 12 months
Real-time search volumes, app downloads, API call counts and enterprise contract signings — all either private or paid. The page publishes what can be checked by anyone and names the gaps rather than filling them.
This page does not give trade timing or price targets. It measures the state of the story: how much attention the AI narrative is getting, whether that is rising or decaying, and whether price is agreeing. What you do with that is yours.
The two-year attention peak was +8.18 standard deviations on 09 Jun 2025. The chart overlays every regime change since, so past peaks and what price did next are visible rather than asserted.