Follow the money

Code can be hidden. Money is much harder to hide. If a lab were sitting on something far beyond its public models, spending, revenue and markets might show it before any announcement. Every figure links to its source: company filings for capital spending and Nvidia's revenue, company statements or press reports for lab revenue, and live prices for prediction markets.

Big Tech capital spending, per quarter

All property and equipment: mostly data centers and chips, also Amazon logistics. Combined cash capital expenditure of Microsoft, Alphabet, Amazon and Meta, from their SEC filings.

Spending and capability, side by side (context only)

Both lines are indexed to 100 at the first quarter, on one log scale. The capability line is the best METR-measured frontier time horizon released by each quarter's end. The two lines measure different things; capability tends to rise faster, so the gap is not a test for hidden spending. The check that matters is trigger X4: a run or cluster 2x the largest known with no matching release.

Nvidia data-center revenue

What the chip supplier sells to everyone building AI, per quarter.

AI lab revenue

If labs kept their best models for their own use, revenue from selling access would flatten while their own ventures grew. This is the exclusive-use tripwire.

Prediction markets

What would be a red flag

Spending with no matching release: a training run or cluster at least twice the largest known, with no public model within 6 months (trigger X4).
Revenue mix shifting: revenue from selling access flattening while labs' own research, trading or products grow (W6 and the exclusive-use tripwire).
Insiders: heavy insider buying, or departures that give up equity with warnings attached.
Markets: a sharp repricing of AGI prediction markets with no public news behind it.

Caution: in an AI investment boom, a spending spike means "bubble" far more often than "hidden AGI". Money signals feed the gauges and triggers. They never set the alarm level on their own.