While Everyone is Focused on Data Centers: Other Bubbles that Could Pop

People-powered, AI-Generated AI-assisted draft, verified, and corrected: Anthropic Claude Sonnet 5.5 Edited and curated by: Guy Wolf *Disclaimer: AI has been known to malfunction “hallucinate” therefore, a human-in-the-middle approach with a separate draft and verification AI has been used to provide the best quality, most accurate apolitical research on this topic Subscribe now A $2 trillion private credit market, a software fund that paid out 5 percent against requests near 38 percent, and memory-chip profits that analysts expect to peak in 2028: what to watch while everyone stares at data centers. A Cap, Honored On July 2, 2026, Semafor reported that investors in Blue Owl’s funds had asked for $4.7 billion back in the second quarter (Semafor, July 2, 2026). In the firm’s software-focused fund, requests came to about 38 percent of the fund, roughly $1.1 billion. The fund’s rules let it pay out only 5 percent of its value in a quarter. It paid 5 percent. Nobody broke a rule. The cap was written into the fund’s terms, and Blue Owl honored it. But do the arithmetic. Of every 38 dollars investors asked for, about 5 came back. That is roughly one dollar in eight. (That ratio is my own calculation from the reported figures, not a number the firm gave.) Three months later the pattern repeated. Third-quarter requests were about $4.2 billion, and the software fund again drew requests near 39 percent, while Blue Owl’s shares were down about 35 percent for the year (24/7 Wall St., October 5, 2026). Now notice what else was on the front pages in those same months. It was chips, power plants and the price of building data centers. Almost no one outside finance was talking about a fund that politely told its investors “not yet.” Why the Shiny Thing Works as a Distraction, and a Test for What Isn’t Shiny The first two articles in this series asked whether AI data centers are a repeat of the dot-com fiber build-out. The short answer so far: the core looks like a late boom with strain at the edges, not a glut. This piece asks a different question. If everyone is staring at one possible bubble, where else might money, debt or promises be piling up unseen? There is a plain reason a loud story hides quiet ones. Attention is limited. A data center is easy to picture: a huge building, a famous company, a big number. A redemption queue at a credit fund, or a rise in a household power bill, is hard to photograph. Both can still matter more. There is also a second reason. One analyst quoted by Fortune in March argued that the AI valuation bubble had already burst, while a rarer earnings bubble was still growing. The same piece cited Bloomberg Intelligence figures showing the biggest seven tech firms growing earnings at about 18 percent against about 11 percent for the other 493 companies in the S&P 500 (Fortune, March 29, 2026). That is one analyst’s view, not settled fact. But if it is even partly right, it means much of the economy’s apparent strength rests on one theme, and trouble elsewhere could be mistaken for noise. The test I am using The test below is my own working framework, not an official definition. It borrows from how economists describe credit cycles in general. I call something a candidate for the next boom and bust only if it shows all four of these: A fast build-up. Money, prices or debt rose far faster than the underlying need for several years. Financing that needs the good times to continue. Borrowers or funds depend on rising prices, easy refinancing or new investors to stay whole. A time mismatch. Promises are made to people who may want their money soon, while the money is tied up for years. A bank run is the classic example. A plausible trigger. There is a specific event that could make everyone want out at once. A thing that shows only one or two of these is a watch item, not a candidate. I will say plainly which is which, and I will say which areas I only screened rather than researched. The Short List, With One-Line Verdicts These verdicts are my own judgments from the sources I opened, as of November 11, 2026. They are not predictions. Researched Private credit and the insurers behind it: candidate. Fast build-up (a market of about $2 trillion), redemption caps already being hit, heavy exposure to software companies whose business AI may disrupt, and rising rates as a trigger. The memory-chip cycle (DRAM and HBM): candidate for a classic cyclical bust. Record spending on new factories is arriving into a market where prices and profits are at a peak. This one is a normal boom-and-bust pattern for the industry, and it is tied to the data center story. Electricity prices and the power grid: watch item, with a political edge. Shortages are real and bills are rising, but this is more likely to produce a backlash than a financial collapse. Federal interest costs: watch item. A slow squeeze, not a sudden event, though it sets the rate environment for everything else. Household and student debt: watch item. Stress is visible but, in the data I opened, not accelerating sharply. Only screened, not researched I did not research the following in enough depth to give a verdict, and I do not want to imply otherwise: gold and crypto (I have no current prices), commercial real estate, quantum computing, humanoid robots, defense and drone spending, critical minerals, China’s property sector, and climate-driven insurance and real estate. They are listed because a reader will ask, not because I have cleared them. A Close Look: Private Credit and Memory Chips Each claim below is sorted into one of three tiers. Fact means a figure from a source I opened. Consensus means a view many analysts share but that is still interpretation. Speculation means a reasoned guess that could be wrong. Private credit and the insurers behind it Fact The market is about $2.02 trillion. Quarterly redemptions from the main non-traded funds are capped at 5 percent of their value. Moody’s recorded the first-ever outflow for perpetual non-traded business development companies (BDCs) in the first quarter of 2026 and put its outlook for them at negative (CAIA, April 20, 2026). Software is about 25 percent of BDC portfolios (Moody’s median). Loans to software firms grew from about $8 billion in 2015 to over $500 billion by the end of 2025, about 19 percent of direct loans, according to the Bank for International Settlements, as cited in the same CAIA piece. Blue Owl’s second-quarter requests were $4.7 billion, with the software fund near 38 percent. Reported delinquencies on its software loans were none, and about 0.2 percent of that fund was judged unlikely to be repaid (Semafor). In other words, the run is ahead of the losses. Third-quarter requests were about $4.2 billion. The 10-year Treasury yield was 5.24 percent on October 1, up from 4.77 percent on September 3, and the coming test was described as refinancing risk (24/7 Wall St.). Insurers and pension funds supply about 70 percent of private credit fund money, per the European Central Bank as reported in May. That figure comes from a news report of the ECB, not the ECB document itself, which I did not open. Consensus Most analysts I read call this stress serious but not yet systemic. Morgan Stanley’s March note put direct-lending defaults near 5.6 percent, possibly 8 percent, against a 2 to 2.5 percent long-run average, and called it “significant but not systemic.” Another firm, Lincoln International, estimates a “shadow” default rate near 6 percent against about 2 percent in the headline figure. Both are secondary reports. Speculation (mine) The weak point is the mismatch: funds promise investors a way out every quarter while holding loans that last for years. The cap protects the fund and traps the investor. If investors lose faith in the cap, it becomes a queue. I cannot say whether that will happen, only that it is the mechanism to watch. The memory-chip cycle Fact Samsung and SK hynix spent about 43.2 trillion won on factories in the first half of 2026, up 35 percent, and Micron spent $7.1 billion in one quarter. Paul Kedrosky called it the largest supply response in memory-market history and thought a 2028 to 2029 downturn plausible (Kedrosky). The first part is reported spending; the downturn is his opinion. SK Group’s chairman said the shortage could last four to five more years (Tom’s Hardware, March 2026). Producers’ own forecasts of tight supply are exactly what one would expect at a peak, so this is evidence of conviction, not of safety. Micron has non-cancellable deals with 16 customers through 2030 covering about 40 percent of revenue, and analysts expect its earnings to peak in fiscal 2028 and then fall sharply (Nasdaq). Consensus Memory has always swung between shortage and glut. Most observers expect this cycle to end eventually; the argument is about when, and whether long contracts soften the fall. Speculation (mine) A memory bust would not look like the 2000 fiber collapse. It would hit chipmakers’ profits and prices first, and could help buyers such as PC and phone makers. Gartner expected 2026 PC shipments to fall 10.4 percent and smartphones 8.4 percent on memory costs. For a data center builder, cheaper memory would be good news. The risk is to the companies and regions that have bet on today’s prices lasting. Three Historical Parallels, and Where Each Breaks The three episodes below come from my training knowledge, not from live searches in this session. The dates and broad facts are well documented, but I have not re-opened primary documents for them here, so treat the details as a starting point for checking. August 2007: BNP Paribas freezes three funds. On August 9, 2007, the French bank BNP Paribas stopped investors from withdrawing from three funds, saying it could not value some of the US mortgage-linked assets inside them. Many date the start of the financial crisis to that week. Where it holds: a fund promised easy exits while holding assets that were hard to price. Where it breaks: those funds held securities tied to US home loans, and the damage spread through a banking system with far less capital than today’s. Private credit sits more outside the banks, though banks still lend to the funds (the Financial Stability Board put that at about $220 billion, as I noted in my research). June 2016: UK property funds close their doors. After the Brexit vote that month, several large UK property funds, including those run by Standard Life, Aviva and M&G, suspended withdrawals because too many investors wanted out and buildings cannot be sold in a week. Where it holds: a fund with quarterly or daily exits holding assets that take months to sell. Where it breaks: the gates worked as designed, the funds reopened after a period, and there was no wider collapse. That is the optimistic reading of today’s 5 percent cap: it may be a pressure valve, not a crack. 2017 to 2019: the last memory supercycle. Memory prices rose sharply in 2017 and 2018, then fell as supply caught up and demand cooled, hitting chipmakers’ profits. Where it holds: record profits led to record factory spending, which led to oversupply. Where it breaks: this time a large share of output is committed under long contracts, and demand is tied to AI systems rather than phones and PCs. Whether that makes the fall softer or just later is genuinely unknown. What the three have in common is not a villain. It is ordinary people and firms making reasonable-looking choices that add up to a problem no one chose. Competing Explanations, Tested Against the Evidence This section uses two structured techniques. Analysis of Competing Hypotheses (ACH) lists several explanations and asks which evidence rules each one out, rather than collecting facts that fit a favorite. It guards against motivated reasoning, because the explanation with the most support is the one that survives the most tests, not the one that feels best. Indicators and Warning then names what we would see early if each explanation were true. The question: is the strain in private credit a contained repricing, or the start of something larger? The hypotheses A. Contained repricing. Software loans lose value because AI changed their prospects, funds gate redemptions, losses stay inside the funds, and the system absorbs it. B. Slow grind. Losses build over several years as loans come due and fail to refinance. Insurers and pension funds take steady write-downs. No single dramatic week. C. Sudden systemic event. A refinancing wall, higher rates and a loss of confidence combine so that investors, insurers and bank lenders pull back at once. D. A mostly temporary scare. Redemptions fade as rates ease and the first losses prove smaller than feared. The evidence For each item, I mark whether it is consistent (+), inconsistent (−) or neutral (0) with each hypothesis, in the order A, B, C, D. These ratings are my judgment. The cap was honored and Blue Owl’s software fund reported no delinquent loans in Q2 (Semafor): A +, B 0, C −, D +. The structure held, and the run is ahead of actual losses. Requests did not fade in Q3; the software fund stayed near 39 percent (24/7 Wall St.): A +, B +, C 0, D −. A passing scare should be fading by now. For the flagship fund, requests did fall slightly (18.8 to 16.8 percent), which is the only data point that fits D. The 10-year yield rose from 4.77 to 5.24 percent in about a month (24/7 Wall St.): A 0, B +, C +, D −. Higher rates make refinancing harder, and D needs rates to ease. Estimated “shadow” defaults near 6 percent versus about 2 percent headline (Lincoln International, via CAIA): A 0, B +, C +, D −. Insurers and pensions supply about 70 percent of the money (ECB, via press): A +, B +, C 0, D 0. Long-term money does not run quickly, which argues against C. But losses there are slow to surface, which fits B. Regulators asked questions in the spring (Treasury met life insurers in May; the Fed asked banks about private-credit lending in April): A 0, B 0, C 0, D 0. This shows attention, not evidence either way, so I count it as neutral. What survives Explanation D has the most inconsistent marks, so it is the weakest. Explanation C is weakened by long-term funding but not ruled out, because the rate move is going the wrong way. A and B survive best, and they are not rivals: a contained repricing in the first year can turn into a slow grind afterward. My reading is that the most likely path is A turning into B, with C a minority risk. I have not put a number on C, because I do not have a defensible basis for one. Indicators and warning These are the signs I would watch. None of them means disaster alone. Requests above the cap for several quarters running, especially if funds start selling loans to meet them. A fund that changes its cap or its rules, or that pays in something other than cash. Non-accrual rates (loans that stopped paying) rising from about 1 percent at the largest funds, as 24/7 Wall St. noted for Blue Owl’s OBDC. Payment-in-kind (PIK) interest rising. PIK means the borrower pays interest by adding to the debt instead of paying cash. The CAIA piece put “bad” PIK at 6.4 percent of loans in Q4 2025. The 10-year Treasury yield staying above 5 percent as loans reach their refinancing dates. Insurers reporting write-downs, or regulators changing how private ratings are reviewed. A Lens From Another Field, and the Assumption Holding It Up The outside lens: tight coupling In 1984 the sociologist Charles Perrow published Normal Accidents, a study of failures in nuclear plants, chemical works and aircraft. His idea was that some systems are both complex (many parts that interact in ways no one fully sees) and tightly coupled (a problem in one part passes to the next quickly, with little slack). In such systems, he argued, serious accidents are not freak events. They are a normal result of the design. Borrowing this for finance reveals something a list of numbers does not. Private credit, insurers, bank credit lines and software companies are linked in a chain. The AI story pressures the software firm; the software loan sits in a fund; the fund’s investors are wealthy individuals and, through insurers, ordinary policyholders; the fund also borrows from banks. Each link seems fine alone. The question is how much slack sits between the links. Where the lens strains: Perrow studied machines. People in finance adapt, regulators change rules mid-crisis, and central banks can step in. A reactor cannot decide to behave differently. So the lens points to where to look, but it overstates how mechanical the outcome is. It was also built from my training knowledge of the book, not re-read for this piece. The load-bearing assumption: “The gate keeps the fund safe” A great deal of today’s calm rests on one idea: a cap on redemptions protects the fund and the remaining investors, because it stops forced sales. This is not a new thought. The economists Douglas Diamond and Philip Dybvig published the standard model of bank runs in 1983 (Journal of Political Economy). As I understand it, they showed that a run can be self-fulfilling: if you expect others to withdraw, withdrawing is the rational thing to do. They also showed that suspending withdrawals can stop a run, but only under conditions, such as when the number of people who truly need cash is predictable. I have not re-opened the paper here, so check the details before relying on them. The popular version, “gates stop runs,” drops those conditions. Where it holds: the UK property funds of 2016, described above. The gates closed, buyers returned, and the funds reopened. Where it can fail: if a gate is read as a warning sign, investors may rush to get in line first in the next window. I recall regulatory and academic work on money-market funds reaching a similar worry, which was part of why US regulators dropped the idea of gates for those funds in 2023. I am recalling this from training and have not verified it live. The 5 percent cap in private credit is not the same instrument, so the comparison is only suggestive. The honest, conditional answer: a gate helps if redemptions come from a temporary scare and the assets are sound. It hurts if redemptions come from a belief that the assets are not sound, because then every investor wants to be first. The evidence in this piece does not yet tell us which case we are in. The software fund’s requests staying near 39 percent for two quarters is consistent with either. Who Is Hurt, and How the Areas Connect The watch items, briefly Electricity. In the largest US grid region (PJM), the capacity auction held in December 2025 cleared at $333 per megawatt-day for 2027/28, the cap, and the region was about 6,600 megawatts short of its reliability target (DatacenterDynamics). US electricity prices were up 4.4 percent over the year to September 2026 (FRED), down from over 5 percent a year earlier, so the rise is easing but still above many household budgets. Rystad Energy reports turbine orders committed years ahead, transformer waits over 160 weeks, and that over half of announced data center power plans may never reach the grid (Rystad). My judgment: this produces a political backlash against data centers, not a financial bust. Federal interest costs. Interest payments are nearly $1 trillion a year, about 14 percent of federal revenue; one budget group projects 30 percent by 2036 (Fortune, May 30, 2026). This is a slow squeeze. Its main effect on everything else is keeping long-term rates high. Household and student debt. The New York Fed’s Q2 2026 report showed $18.8 trillion of household debt, 4.7 percent of it delinquent, and 10.6 percent of student loan balances 90 or more days late (NY Fed). Real stress, but not a sharp break in this report. Second- and third-order effects These are my reasoned expectations, not observed outcomes. I use confidence words on purpose. Likely: if private credit tightens, healthy software and mid-size firms that depend on it for loans face higher costs or fewer lenders, even if they are not distressed. Private credit funded 84 percent of leveraged buyouts in 2024 (CAIA), so buyout-owned companies are the first exposed. Possible, not yet evidenced: insurers that hold these loans could cut back on other lending or raise prices on annuities and policies. The channel exists, but I have seen no data showing it working yet. Possible: a memory-price fall could lower costs for device makers while straining chipmakers and the regions, such as South Korea, that depend on them. Well-documented as a pattern: when savers are told they cannot withdraw, trust in the whole category suffers, even in funds that are sound. Possible solutions’ own side effects: tighter rules on how insurers rate private loans could slow the market and push lending to less regulated corners; looser gates could cause the very run they are meant to prevent. How the areas connect The candidates are not separate stories. AI hurts software firms, which hurts the loans to them, which hurts the funds. High federal borrowing keeps rates up, which makes all refinancing harder. Power shortages raise the cost of the data center build-out that memory makers are building for. This is the tight coupling from the last section, and it is why I would rather ask where the slack is than which single area fails first. Two fair objections “Private credit is too small and too slow-moving to matter.” At roughly $2 trillion it is large, but smaller than the US banking system or the mortgage market of 2007. Morgan Stanley’s note called the stress significant but not systemic. This is a serious point, and I weight it heavily. My answer is only that the ownership by insurers and pensions makes it matter to ordinary people in a way its size alone would not. “You are inventing a bubble because bubbles are fashionable.” Fair. That is why the test in this piece requires four conditions, and why only two areas pass it. I am not claiming a collapse. I am claiming these two deserve more attention than they are getting. What Is Being Done, and Three Questions to Ask Efforts already underway These are real, but early. None has yet been tested by a serious run. Supervisors are asking questions. The US Treasury met life insurers on May 7, 2026 about hidden risk, offshore opacity and the process by which insurers challenge the private ratings given to their holdings; the Federal Reserve asked banks about private-credit lending in April. These were reported by the press and I did not open the meeting records, so the details may differ. International bodies are measuring it. The Bank for International Settlements (BIS) tracked the growth of loans to software firms, and the Financial Stability Board estimated bank credit lines to private credit funds at about $220 billion (both cited in the sources above). Measurement comes first; rules usually come after. Rating agencies are signaling. Moody’s put its outlook for non-traded BDCs at negative in April 2026 (CAIA). The funds’ own design. The 5 percent cap exists to prevent forced fire sales, and Blue Owl honored it in both quarters I reviewed. Whether that is a strength depends on the load-bearing question above. Grid planning for power. The December 2025 PJM auction price cap and state efforts to shield ratepayers, described by DatacenterDynamics, show regulators already reacting on electricity. Three questions to ask These are for a reader who invests, for one who works in finance, and for anyone who simply wants to read the news better. If you own or are offered a private credit or similar fund: what happens if many investors ask to leave at once, and have I read that part of the terms? A cap is a promise about how little you can take out in a bad quarter. If you hold insurance, an annuity or a pension: what share of what backs it is in private loans, and who rates them? Many providers publish this, and it is worth asking for. If you follow the AI story: who lent the money, not just who spent it? A boom is easiest to judge by following the debt back to the people whose savings are inside it. One thing to do Pick one fund, insurer or pension you have a stake in, and find the one-page explanation of how redemptions or withdrawals work. That single page tells you more about your real risk than most headlines. The Question Behind the Question A fund that says “not yet” is doing what its rules say, and that is the point. The danger in a boom is rarely a villain. It is a promise made in good faith to many people at once, backed by something that takes years to turn into cash. Data centers may or may not be the next fiber glut. But the place to look for trouble is wherever a quiet promise outruns a slow asset, and I have tried to show where that looks most true today. The lens also splits into two jobs that are easy to confuse. As an investor, the lesson is to read the exit terms before you need them. As a citizen, it is to ask who is holding the risk when a theme is popular, because it is often savers and policyholders who never chose to hold it. 3-Minute Summary Data centers are the visible AI story. Quieter places may carry more risk. I used a four-part test: a fast build-up, financing that needs good times, a time mismatch, and a plausible trigger. Two areas passed: private credit (with the insurers behind it) and the memory-chip cycle. In private credit, Blue Owl’s software fund received requests near 38 to 39 percent of its value in two straight quarters and paid out 5 percent, as its rules allow. Losses on those loans had not yet shown up in the reported data. The 10-year Treasury yield rose above 5 percent in October, so refinancing is the next test. In memory chips, record factory spending is arriving at a price peak. A downturn around 2028 to 2029 is plausible but contested. It would hit chipmakers more than data center builders. Electricity prices, federal interest costs and household debt are real pressures but, in my reading, watch items rather than candidates. My most likely path for private credit is a contained repricing that turns into a slow grind. A sudden systemic event is a minority risk, and I have not put a number on it. Several areas were only screened and not researched: gold and crypto, commercial real estate, quantum, humanoid robots, defense, critical minerals, China property, and climate-driven insurance. Something You Can Do Today Find the one-page explanation of how withdrawals work for one fund, insurer or pension you have a stake in, and read it. Something You Can Share A boom isn’t only who spent the money. It’s who lent it, and how fast they can ask for it back. About this Analysis Original source analyzed: none. This is a new piece prompted by a request, not a fact-check of one article. Date written: November 10 to 11, 2026. Data are as of November 10, 2026. AI model and version: Claude Sonnet 5.5 (model ID claude-sonnet-5-5). The serving model can differ from the configured one during a session. Training cutoff: end of June 2026. Verified through live search in this session: the figures from the pages listed under Sources below, which I opened. Drawn from training knowledge and not re-verified here: the three historical parallels, Perrow’s Normal Accidents, the Diamond and Dybvig model, and the 2023 US money-market-fund rule change. Several figures (the ECB’s 70 percent, Morgan Stanley’s default estimate, Lincoln International’s shadow default rate, the Treasury and Fed meetings) are as reported by secondary sources, and I did not open the primary documents. The test for a “candidate” and the A-to-D evidence ratings are my own judgment. Sources & Further Reading Pages I opened for this piece. Items marked (secondary) are interpretations by a news or analyst outlet. Private Credit: Redemptions, Defaults, and Wrappers, Oh My, CAIA, April 20, 2026 (secondary). Blue Owl hit by billions in redemptions again, Semafor, July 2, 2026 (secondary). The $2 trillion private credit market just got some good news, but its biggest test may still be coming, 24/7 Wall St., October 5, 2026 (secondary). The AI bubble has already burst; a rare one is still growing, Fortune, March 29, 2026 (secondary). Monday Signal: the memory supercycle and how it ends, Paul Kedrosky (commentary). SK Group chairman says memory chip shortage will last until 2030, Tom’s Hardware, March 2026 (secondary). Memory supercycle peak near: Micron and SK hynix, Nasdaq (secondary). Data center boom fuels PJM power cost increases as states move to shield ratepayers, DatacenterDynamics (secondary). Data centers and electrification test the limits of power delivery, Rystad Energy. Consumer Price Index: electricity, Federal Reserve Bank of St. Louis (FRED), data through September 2026. National debt and Treasury rates, Fortune, May 30, 2026 (secondary). Quarterly Report on Household Debt and Credit, Q2 2026, Federal Reserve Bank of New York. Not opened, cited only through secondary reports or my training knowledge: the ECB, BIS, Financial Stability Board, Moody’s, Morgan Stanley and Lincoln International publications; Diamond and Dybvig (1983); Perrow (1984).