How to Predict the AI Bubble Peak: A Signal-Based Framework

Disclaimer: This article reflects the author's personal opinions and analysis only. It is provided for general informational and educational purposes and does not constitute financial, investment, legal, or tax advice, nor a recommendation or solicitation to buy, sell, or hold any security, cryptocurrency, or other asset. The views expressed are not those of any employer, client, or affiliated organisation. All data, figures, and interpretations are based on publicly available sources believed to be accurate at the time of writing but are not guaranteed and may be incomplete, outdated, or subject to revision. Markets are inherently unpredictable, and no signal, gauge, or threshold discussed here guarantees any future outcome. Readers should conduct their own independent research and consult a licensed financial advisor before making any investment decision. The author accepts no liability for any loss or damage arising from reliance on this content.

Market Signals Dashboard: https://aistorm.cc/market-signals/

Every speculative boom ends the same way in hindsight - obvious, inevitable, and somehow still a surprise to almost everyone holding the bag when it breaks. The AI buildout of the mid-2020s will likely be no different. The genuinely useful question isn't whether this ends, but what will be visible in the data before it does. Below is a practical framework - four signal families, tracked continuously - for spotting the top before the headlines do. The underlying data for each of these signals is tracked live on our Market Signals dashboard (https://aistorm.cc/market-signals/), updated as new figures become available.

Fed liquidity is the tide; watch it before you watch the boats

Nothing in markets happens in a vacuum of liquidity. Right now, Fed Bank Reserves sit at roughly $2.99 trillion - comfortably above the $2.8 trillion line we treat as "melt-up intact." As long as reserves stay above that level, the system has enough spare liquidity sloshing around to keep funding risk assets, AI stocks included. The signal to watch for isn't a slow drift - it's a fast break below roughly $2.5 trillion outside of an actual banking-system crisis. That is the point at which liquidity is being withdrawn quickly enough to threaten valuations broadly, and AI names - which have been among the largest beneficiaries of cheap, abundant capital - would be first in line to reprice.

The companion metric here is the SOFR − RRP spread, currently 21.0 basis points - already well past the roughly 3bp threshold we treat as an early liquidity-stress warning, and it's flashing orange (watch) on our gauge board. A rising spread means banks are increasingly paying up for overnight cash instead of resting comfortably on the Fed's reverse-repo floor. It's a leading indicator precisely because funding markets usually crack before equity markets notice - banks feel the squeeze weeks or months before it shows up in stock prices.

Two supporting reference gauges round out the liquidity picture: the Treasury General Account, at $907.3 billion, and M2 year-over-year growth, at 5.5%. Neither has a hard trigger on its own, but a TGA that's rapidly rebuilding (draining reserves in the process) or M2 growth decelerating sharply toward zero would both tighten the liquidity backdrop further - worth reading alongside the reserves and spread gauges, not in isolation.

The capex "escape signal" - and it just started flashing

If liquidity is the tide, hyperscaler capital expenditure is the boat that tells you how much conviction is actually being deployed into AI infrastructure. Four companies - Amazon, Google, Microsoft, and Meta - collectively account for the overwhelming majority of AI-related data-center and GPU spend, and their quarterly capex growth rates are, in our view, the single highest-priority signal on this entire page.

The current picture is mixed in a way that deserves real attention. Amazon's capex is still growing 22.6% year-over-year (to roughly $54.2 billion), Google's 28.1% (to $35.7 billion), and Microsoft's 16.0% (to $35.8 billion) - all still expanding, still building. But Meta's capex growth has gone negative - down 11.2% year-over-year, to roughly $19 billion. That is the tell to watch closely. A single hyperscaler pulling back doesn't confirm a peak on its own - company-specific reallocation happens for all sorts of reasons - but when the largest, best-capitalized buyers of AI infrastructure start cutting rather than accelerating spend, it is historically one of the earliest and most credible signs that the investment cycle has crested. The reasoning is simple: these companies have the best forward visibility into AI demand, utilisation, and return on invested capital of anyone in the market. When they slow down, they are voting with the one thing that matters more than sentiment - capital.

The signal to watch going forward: does Meta's pullback stay isolated, or does it spread to a second or third hyperscaler in the next one or two earnings cycles? A synchronised deceleration across two or more of the big four would be a far stronger confirmation than any single company's guidance call.

Valuation stretch - the slow-burn confirmation, not the trigger

Two backward-looking but historically reliable valuation gauges round out the picture: the Buffett Indicator (total market cap to GDP), currently at 224%, and the Nasdaq-100 forward P/E, currently 30.9. Neither number is, by itself, a timing tool - a stretched valuation can persist for years before it matters, and both of these have printed even higher during past manias. What they're useful for is context: confirming that whatever liquidity and capex signals are doing, they're doing it against a backdrop where prices already embed a lot of optimism. A rising Buffett Indicator or P/E climbing to new multi-decade highs while liquidity gauges turn red and capex growth broadens its deceleration is a very different situation than the same valuation levels persisting alongside ample liquidity and accelerating capex. Valuation stretch is the amplifier, not the trigger.

The crypto canary

Bitcoin perpetual funding rates are the fastest-moving, most sentiment-sensitive gauge on the board - a real-time read on how much leveraged long speculation is being financed in the most liquidity-sensitive corner of markets. The current daily-median funding rate is 0.0075% on 8-hour prints - modest, not obviously frothy. Because crypto derivatives markets are hyper-levered and react to liquidity conditions almost instantly, a sharp spike in funding rates (leveraged longs paying up aggressively) or a violent flip negative (a rush to cover, deleveraging) often precedes broader risk-asset stress by days to weeks. It won't call the top in AI stocks directly, but it's frequently the first place cracks in market-wide risk appetite show up.

Putting it together

No single gauge calls the peak. The framework that matters is convergence: Fed reserves breaking down through $2.5 trillion, the SOFR-RRP spread staying elevated or widening further, capex deceleration spreading beyond one hyperscaler, valuation multiples at extremes, and crypto funding rates swinging violently - all showing up in the same window. Today, the picture is mixed-to-early: liquidity is still broadly supportive but stressed at the margin (SOFR-RRP already in watch territory), capex is still expanding in aggregate but showing its first real crack at Meta, and valuations are stretched but not yet paired with a liquidity break. That combination doesn't say "peak now" - it says "watch closely, especially the next round of hyperscaler earnings and the reserves/spread trend." The AI bubble, if it is one, will very likely announce its top through exactly this kind of data - not a headline, but a shift across two or more of these gauges at once.