Railroads and 1990s telcos were heavily debt‑financed; AI infra today is mostly funded by cash‑rich companies.
Dark fiber was actually quite reusable (once cheap enough), which eventually enabled broadband and cloud, but it took years for demand to catch up. AI data centers are general compute + storage + networking: even if AI inference demand is weaker than hoped, much of that capacity can still run generic cloud, storage, or non‑AI workloads.
In rail and fiber, physical infra was fixed and demand was slow to materialize. With AI, software and business models iterate fast. It doesn’t remove bubble risk, but it shortens the lag between infra and profitable use.
Cycles take 3–7 years from early mania to obvious bust (railroads, 90s telco, dot‑com). Given the AI infra boom really accelerated in 2023–2024, a 2026–2029 window for a serious shakeout is consistent with prior episodes, but it can be front‑loaded by a macro shock or delayed if genuine killer apps ramp faster than expected. Aspects of AI are doubling every 3 1/2 (code) to 6 months (compute).
Overcapacity is a credible path to an AI bubble “pop,” but it’s better understood as a probable regime shift over the next few years, triggered when utilization, pricing, and capex guidance finally force investors to mark expectations back to reality. Prediction markets say 1/3 chance of a burst by end of 2026 (2/3 chance no burst, bull market continues). This is in keeping with rate cuts and S&P 500 history. The last 22 times the Fed has lowered rates with the S&P 500 within 2% of record highs, stocks have been higher 12 months later 100% of the time. So the fears that rate cuts lead to recession is only based on two data points (2008 and 2020) but both years were caused by exogenous one-off events which tanked the economy. Cycles in 1984, 1995, and 1998 eased policy without a subsequent recession as markets hit new highs.
Global liquidity
In practice, GDP growth rides on population and productivity with governments increasingly relying on debt to pull future demand into the present as aging demographics slow labor‑force growth. Over the last cycle, broad money and central‑bank balance sheets have grown at mid‑ to high‑single‑digit rates, steadily debasing cash and driving capital into scarce assets. Bitcoin, whose direction aligns with global liquidity over 80% of the time and has posted correlations above 0.9 with some M2 composites, is one of the purest beneficiaries of this dynamic.
Looking ahead, 2025’s relatively tight liquidity and higher real yields help explain today’s ‘alligator jaws’ between Bitcoin and the NASDAQ. But 2026 faces a massive Treasury maturity wall, roughly a third of outstanding US debt, about $10T, comes due, and regulatory easing of leverage ratios could unlock trillions in additional bank balance‑sheet capacity, especially for Treasuries and repo. Layered on top of whatever fiscal packages Washington passes, that combination sets the stage for a renewed global liquidity updraft in 2026, with obvious implications for assets most sensitive to money creation such as stocks and bitcoin.
Future buy pointsWith the S&P 500 hitting new highs, keep a close eye on leading groups including uranium, data center, and big tech stocks as they round out their bases. Some may offer pocket pivot, volume dry-up, and undercut & rally entry points.