Liquidity, capex growth, and valuation-stretch indicators for spotting an AI-market bubble peak — hyperscaler capex growth, Fed reserves, Treasury General Account, SOFR-RRP spread, M2 vs. valuation, and a crypto funding-rate canary.
These seven gauges are the latest reading for each signal, colour-coded against the thresholds most likely to matter at an AI-market bubble peak: green means the underlying condition still supports the current melt-up, orange means it is worth watching closely, and red means the threshold has already been breached. Only Fed Bank Reserves and the SOFR − RRP Spread have a hard, colour-coded threshold today: reserves staying above roughly $2.8 trillion is read as liquidity still being ample enough to support risk assets, while a fast break below about $2.5 trillion (outside of an actual banking-system crisis) is treated as an exit signal — the point liquidity is being withdrawn fast enough to threaten valuations, including AI stocks. A SOFR − RRP spread sustained above roughly 3 basis points is treated as an early liquidity-stress warning, since it means banks are paying up for overnight cash rather than resting on the Fed's reverse-repo floor. The remaining five gauges (Treasury General Account, M2 YoY %, Buffett Indicator, Nasdaq-100 P/E, and BTC Funding Rate) are shown as reference readings rather than hard triggers — there is no single "safe" number for any of them. Instead, they matter in the context of their own trend: a rapidly draining Treasury account, decelerating M2 growth, a Buffett Indicator or P/E climbing to a multi-decade high, or a persistently rich BTC funding rate should all be read alongside the charts further down the page, which show how each one is actually trending over time.
The single highest-priority "escape signal": watch the growth RATE, not the absolute capex level. A decelerating or negative growth rate — even while capex dollars keep rising — signals the AI infrastructure cycle is rolling over. Sourced directly from each company's SEC 10-Q/10-K filings.
What it shows: the quarter-over-quarter percentage change in capital expenditure reported by Microsoft, Alphabet, Amazon and Meta — the four companies building most of the world's AI data-center capacity. Each line is a growth rate, not a dollar amount. What it could indicate: as long as growth rates stay positive and roughly flat or rising, hyperscalers still expect enough AI demand to justify spending even more each quarter. A sustained slowdown means they are pulling back their forward bets on AI demand, usually well before that shows up anywhere else. Bubble-peak relevance: this is treated as the single highest-priority "escape signal" on this page. Because much of this spending is locked in by multi-year, take-or-pay contracts, the absolute capex level can keep climbing for quarters after real demand has already peaked — so watch the RATE of growth rolling over, not the headline dollar figure. A synchronised deceleration or a swing to negative growth across two or more of these companies is the clearest early warning that the AI infrastructure build-out cycle is turning.
Bank reserves and Treasury General Account balance, normalised to a 0–100 scale (tooltip shows original values). SOFR-RRP spread shown separately in basis points.
What it shows: two measures of dollar liquidity in the financial system — bank reserves held at the Federal Reserve, and the Treasury General Account (TGA), the US government's own checking account at the Fed. Both are normalised here so their shapes can be compared on one chart; hover any point to see the real dollar values. What it could indicate: rising bank reserves generally mean more cash is available in the system to flow into risk assets, including AI stocks. A rapidly draining TGA (the government spending down its cash balance) temporarily adds liquidity to the banking system, while TGA rebuilding (e.g. after a debt-ceiling resolution) drains it. Bubble-peak relevance: the melt-up in risk assets is considered intact as long as bank reserves stay comfortably above roughly $2.8 trillion. A fast break below about $2.5 trillion — outside of an actual banking-system crisis — is flagged as an exit signal, because it suggests the liquidity that has been inflating asset prices, including AI valuations, is being withdrawn from the system.
A sustained spread above ~3bp signals repo-market funding stress and tightening liquidity.
What it shows: the gap, in basis points, between SOFR (the Secured Overnight Financing Rate — what banks actually pay to borrow cash overnight against Treasury collateral) and the interest rate the Fed pays on its overnight reverse repo (RRP) facility. What it could indicate: under normal conditions this spread sits close to zero, since RRP acts as a floor beneath overnight rates. A widening spread means banks and dealers are willing to pay noticeably more than the Fed's floor rate to get overnight cash — a sign that reserves are scarcer and short-term funding markets are tightening. Bubble-peak relevance: a spread sustained above roughly 3 basis points is treated as a yellow-alert for repo-market funding stress. Because leveraged positions (including margin used to fund concentrated bets on AI stocks) depend on cheap, abundant overnight funding, a persistently elevated spread can force deleveraging across risk assets well before it is visible in equity prices themselves.
M2 money supply growth, the Buffett Indicator, and Nasdaq-100 trailing P/E — normalised to a 0–100 scale.
What it shows: three unrelated measures of how "stretched" asset prices and money supply are relative to their own history — M2 money supply growth (how fast the broad money stock is expanding year-over-year), the Buffett Indicator (total US stock market capitalisation divided by GDP), and the Nasdaq-100's trailing price/earnings ratio (using QQQ as a proxy). What it could indicate: faster M2 growth generally supports higher asset prices by increasing the money available to chase them. A rising Buffett Indicator shows equity values pulling further away from the size of the real economy that ultimately supports corporate earnings. A climbing trailing P/E shows investors paying more per dollar of current earnings, often on the assumption that future (AI-driven) earnings growth will justify today's price. Bubble-peak relevance: none of these three signals is a precise timing tool on its own, but together they describe how much of current AI-era valuations rest on optimism rather than realised fundamentals. Historically, previous market peaks have coincided with the Buffett Indicator and P/E multiples reaching multi-decade extremes at the same time M2 growth was decelerating — money creation slowing just as valuations were most stretched.
Crypto often turns before equities. A stabilising or declining perpetual funding rate signals leveraged longs unwinding ahead of a broader market top.
What it shows: the daily median funding rate on Bitcoin perpetual futures — the periodic payment traders holding leveraged long positions pay to those holding shorts (or vice versa, when negative) to keep the perpetual contract's price tied to the spot price. What it could indicate: a persistently high, positive funding rate means the market is crowded with leveraged long speculators willing to pay a premium to stay long — a sign of excess speculative leverage and one-sided positioning. A rate that stabilises or turns negative shows that leveraged longs are being unwound, whether through profit-taking or forced liquidations. Bubble-peak relevance: crypto is a smaller, faster, more retail-driven market than equities, so leverage typically builds up — and unwinds — there first. Bitcoin's funding rate is treated as a "canary in the coal mine" here: speculative excess in crypto tends to break down before a broader risk-asset top, including in AI-related equities, becomes visible.
Disclaimer: The information and data published on this website are sourced from publicly available internet content and curated by an automated AI agent. We make no representations or warranties of any kind, express or implied, regarding the accuracy, completeness, reliability, or timeliness of any data or information presented herein. As data is collected and updated on a continuous basis, certain historical records may be incomplete, subject to revision, or no longer reflective of current conditions. This website and its operators expressly disclaim all liability for any loss, damage, or adverse consequence arising directly or indirectly from reliance on the information contained herein. All users are advised to independently verify any information before acting upon it.