In the autumn of 1907, a single financier sat in the library of his Madison Avenue brownstone and decided whether the United States would have a functioning banking system the following morning. J. Pierpont Morgan had no formal authority over the Treasury, no seat at the Federal Reserve (which did not yet exist), and no public mandate. He had something more useful. He controlled, or could summon at will, the balance sheets that mattered. Over three weeks of the Panic of 1907, Morgan organized private rescues for trust companies, the New York Stock Exchange, and ultimately the City of New York itself. The episode produced two enduring lessons. The first was that an industrial economy of unprecedented productive capacity had outgrown the financial plumbing meant to support it. The second was that when the state lacked the tools to act, private capital would act in its place, and would extract terms accordingly.
It is worth holding that scene in mind while reading the news of the past ninety days.
Five American companies have committed to roughly $660 to $720 billion in capital expenditure during calendar year 2026, the overwhelming majority of it directed at artificial intelligence infrastructure. Microsoft, Alphabet, Amazon, Meta, and Oracle have collectively committed to spending between $660 billion and $690 billion on capital expenditure in 2026, nearly doubling 2025 levels, with some analysts now placing the figure closer to $725 billion. Wall Street analysts estimate total AI capital expenditures could now climb above $1 trillion in 2027. For context, that is more than the combined defense budgets of every NATO member outside the United States, deployed by five firms, over twenty-four months, on a single technology stack.
This is the central fact of the present moment, and every other story in business, finance, and geopolitics is now organized around it.
The new trusts
The most useful historical frame for understanding what is happening is not the dot-com bubble of 1999, which it superficially resembles, but the period between roughly 1898 and 1907. That window saw the consolidation of American industry into a small number of vertically integrated giants, the rapid build-out of a transformative general-purpose technology in electrification, a great-power scramble for the raw materials that fed it, and a financial architecture that ran ahead of regulators until it broke.
The numbers from that era are instructive. Between 1897 and 1904, some 4,000 firms vanished into larger corporations that served national markets and exercised an unprecedented degree of control over the economy. The first billion-dollar corporation was United States Steel, formed by financier J. P. Morgan in 1901, who purchased and consolidated steel firms built by Andrew Carnegie and others. By the time the dust settled, a handful of trusts in steel, oil, sugar, meatpacking, and rail controlled the commanding heights of the American economy.
Today’s parallel is cleaner than it should be. The buildout of compute is not merely concentrated, it is concentrating further. Alphabet alone is on track for capex approaching $185 billion this year, scaling toward $250 billion by 2027 on current Morgan Stanley estimates. Capital intensity is reaching historically unthinkable levels of 45 to 57 percent of revenue. The pure-play AI vendors that depend on this infrastructure, Anthropic and OpenAI most prominently, generate revenues that remain a small fraction of what is being spent on their behalf. The capital flowing into AI infrastructure has the unmistakable signature of a trust era. A small number of firms with privileged access to capital, energy, land, and proprietary silicon are building an asset base that is becoming increasingly difficult to replicate, and that the rest of the economy will be forced to rent.
The strategic implication is not that this is a bubble, although it may also be one. The strategic implication is that the productive base of the next economy is being built inside a half-dozen corporate balance sheets, and that ownership of that base will determine who gets to participate in the industries it enables.
The return of the state as principal
The first Gilded Age ended because the state eventually caught up. The Sherman Act of 1890 was reactivated under Theodore Roosevelt, the Federal Reserve was created in 1913 in direct response to the Panic of 1907, and the New Deal generation that came of age in this period built the regulatory architecture that defined American capitalism for the next seventy years.
What is striking about the present moment is that the state is not waiting. It is acting as principal, not regulator.
Washington’s fusion of muscular economic interventionism and transactional dealmaking has forged a new industrial strategy playbook that is reshaping the relationship between state and market. In 2026, aspects of the US playbook will go global. As governments adopt variations of the US’ interventionist repertoire, businesses will settle into a new normal in which they must anticipate greater scrutiny, manage competing demands from multiple capitals, and identify opportunities for government support. The age of laissez faire is giving way to an era in which governments are not merely referees but major players in the corporate arena.
The mechanisms are familiar to anyone who has read the diplomatic history of 1900. Washington is using loan guarantees, equity stakes, price floors, and long-term purchase agreements to underwrite critical mineral supply chains. The administration launched a more muscular industrial policy aimed at strengthening U.S. mineral resilience through domestic and allied investment. These efforts deployed a broad toolkit, including loans and loan guarantees, quasi-equity and equity investments, price floors, and long-term purchase agreements. The same era produced the Critical Minerals Action Plan agreed at the G7 in June 2025, the Quad Critical Minerals Initiative in July, and more than $10 billion in joint commitments with Asian partners to finance, build, and stockpile critical mineral supplies.
The 1900 analogue is the scramble for rubber, oil, and the resources of empire. Britain, Germany, France, and the United States competed for the inputs of the new industrial age through a combination of state-chartered companies, direct colonial administration, and treaty diplomacy. Today the contested input is not rubber but indium, gallium, neodymium, and the engineering talent that converts them into compute. Indium is the one critical mineral that remains restricted in the wake of the November 2025 US-China trade deal, throttling global supply of a key input in data centers. The methods are quieter than gunboats but the logic is the same. Whoever controls the supply chain of the dominant general-purpose technology controls the terms on which others can use it.
The financial system runs ahead of its plumbing
The third structural feature linking 1907 to 2026 is the gap between the financial system as it actually operates and the regulatory architecture meant to oversee it.
In 1907, the source of fragility was the trust company, a lightly regulated non-bank financial institution that performed bank-like functions, including taking deposits, but operated outside the clearinghouse system. When Knickerbocker Trust failed, it triggered a cascading run because no one knew the full scope of trust company exposures, and the bank-like assets they held could not be liquidated at par.
The 2026 analogue is private credit. The market has grown from roughly $158 billion in 2010 to a $1.8 trillion market, and is now generating the kind of warning signals that always precede a serious dislocation. A new Financial Stability Board report has flagged the potential risks to banks, insurers and asset managers arising from private credit’s complex lending structures and opaque data. It comes amid growing jitters surrounding private credit in the U.S. spanning software exposures, business development companies, and individual corporate blow-ups. Banks have $95 billion in committed credit lines to private credit vehicles, up from $8 billion in 2013, and the Boston Fed documented $300 billion in total commitments to private equity and private credit funds, 30 times the level of a decade ago.
Senior bank executives are doing in 2026 exactly what their predecessors did in 1906. During Q1 2026 earnings calls, bank CEOs universally downplayed systemic risk. JPMorgan’s Jamie Dimon called the $1.8 trillion market still small enough to pose no significant threat, and Morgan Stanley’s Ted Pick described it as “an adolescent moment” for a young asset class. The Treasury Secretary concurs. The reassurances are likely correct on a one-year horizon and worth very little on a five-year one.
What private credit shares with the 1907 trust system is not a particular vulnerability but a structural one. Rather than triggering an abrupt systemic collapse, risks in private credit are more likely to accumulate beneath the surface, becoming evident only when liquidity constraints bind or valuations are tested under stress. Stale marks, infrequent valuations, and limited disclosure produce an “illusion of stability.” Reported prices remain steady, volatility appears subdued and performance seems consistent. Yet this apparent resilience is, to a significant extent, an artifact of valuation conventions and liquidity management practices rather than a reflection of underlying economic fundamentals. The system looks stable until the moment redemptions test it, and by then the architecture has already failed.
The interaction between private credit and the AI capex cycle is the part regulators are not yet pricing. A meaningful share of the data center buildout is being financed off the hyperscaler balance sheet, through joint ventures, special purpose vehicles, and private credit facilities arranged by the same handful of firms underwriting the AI revolution itself. If AI revenues disappoint, the first place the strain will appear is not the listed equity of the hyperscalers. It will be in the financing vehicles that sit one layer below them, in the BDCs and direct lending funds that hold the paper, and in the insurance balance sheets that own roughly a third of the private credit market.
The geopolitical reorganization
The 1898-1914 period was characterized by the unraveling of the first great age of globalization. Trade volumes peaked in 1913. The institutional architecture that had governed international economic relations for two generations was overtaken by the strategic logic of rising powers, declining ones, and resource competition. The First World War did not cause the unraveling. It ratified it.
The current geopolitical environment is not on the path to a 1914. It is on the path to something with no obvious precedent, a multipolar order in which the major powers cooperate selectively on issues where their interests align and compete openly where they do not. New security alliances and trade deals are emerging despite intensifying economic friction as policymakers seek to compensate for the eroding global order.
The specific developments of the past sixty days illustrate the texture. The Trump administration announced on March 21, 2026, that it would delay a planned meeting with President Xi in China, originally scheduled for late March, due to the ongoing war in Iran, with a rescheduled summit on May 14-15. The UAE announced plans to exit OPEC, the global body that coordinates oil production among members, effective May 1, a move framed as economic necessity but reading as the latest sign that the post-1973 energy architecture is dissolving in real time. Japanese Prime Minister Sanae Takaichi is reportedly planning to visit Australia in early May 2026 to sign a bilateral deal covering energy, rare-earths, food, and other critical commodities, paired with a major revision to its defense export rules, allowing the overseas transfer of a broader range of military equipment including missiles and warships. The first major contract is a $7 billion submarine deal with Canberra.
Each of these moves would have been unthinkable a decade ago. Together they describe a world in which middle powers are hedging against the reliability of the American security guarantee by building their own industrial and defense capacity, often in partnership with each other and outside the institutions that have nominally governed the postwar order. Defense alliances appear poised to transition to more regional initiatives, with the Philippines-Japan Reciprocal Access Agreement and India’s deepening strategic partnerships with Israel and Cyprus as further data points.
The 1898-1914 precedent is instructive here too. The Anglo-Japanese Alliance of 1902, the Entente Cordiale of 1904, and the Anglo-Russian Convention of 1907 were each minor bilateral arrangements that, in aggregate, reorganized the strategic map of the world over the course of a decade. The current wave of bilateral and minilateral defense and resource agreements has the same character. No single deal looks decisive. The pattern is.
The energy question
Underneath every story above sits a single physical constraint. Electricity.
The Second Industrial Revolution was not really about steel, although the steel statistics are easier to cite. It was about the electrification of industrial production, which raised total factor productivity in ways that took decades to fully manifest. The compute revolution is the same kind of event with the same kind of physical bottleneck.
The hyperscalers have figured this out. Power infrastructure will consume another sizable portion of the expense stack. AI training clusters draw loads of electricity, forcing hyperscalers to commit to long-term agreements for renewable and nuclear capacity. The Middle East war has compounded the problem. Oil briefly traded above $100 per barrel earlier this spring, and traffic through the Strait of Hormuz effectively coming to a standstill has reminded markets that the energy transition is not a substitute for energy security in the short run. The energy price and supply shock is expected to constrain household consumption over the medium term, particularly for durable goods.
The strategic question for the next twenty-four months is whether the United States can build enough generation capacity, transmission, and grid resilience to support the compute buildout it has committed to. The answer is currently no, and the gap will be closed through some combination of nuclear restarts, accelerated permitting, behind-the-meter generation at hyperscaler sites, and a level of state intervention in electricity markets that would have been politically impossible five years ago. The states that figure this out first will host the next generation of frontier model training. The ones that do not will lose the industry to those that did.
What the 1907 frame suggests
The Panic of 1907 did not end the Second Industrial Revolution. It accelerated the institutional reforms that allowed it to mature. The Federal Reserve Act of 1913, the Sixteenth Amendment, the Clayton Antitrust Act, and the Federal Trade Commission all flowed downstream from the recognition that an industrial economy of the scale that had been built required new instruments of governance.
The current moment is poised for a similar reckoning. The signals are visible already. Antitrust action against the hyperscalers is now bipartisan in everything but name. The Financial Stability Board is asking the questions about private credit that were asked about money market funds in 2007 and shadow banking in 2009. State capacity is being rebuilt across industrial policy, export controls, and critical minerals, areas where it had atrophied since the 1980s. The institutional architecture of the 2030s will not look like the architecture of the 2010s.
For operators, investors, and policymakers, the implications are practical.
The first is that scale matters more than at any point in the last half-century, and the second tier of every industry is being competitively eliminated by the first. The capital expenditure gap between the top five compute providers and everyone else is now wide enough that it is not closable through private capital alone. Either the state intervenes to maintain a competitive market, or the market consolidates further. Both outcomes have winners.
The second is that the financial plumbing built for the post-2008 world is being stressed in ways that the post-2008 regulatory framework was not designed to address. Private credit, stablecoins, and the increasingly opaque interfaces between insurance balance sheets and alternative assets are the soft spots. The next financial dislocation will not look like 2008 or 2020. It will look like 1907, originating in the lightly regulated periphery of the system and propagating inward.
The third is that geopolitical risk is no longer something institutional investors can diversify away. The map is being redrawn in real time, through bilateral resource deals, defense export liberalization, and the quiet construction of regional security architectures that do not require American consent to function. The investment thesis that worked for the last thirty years assumed a stable, rules-based international economy. The next thirty years will not provide one. Investors who understand this are already repositioning. Most are not.
The fourth is that the historical rhyme is not subtle. A concentrated capital base building a transformative technology stack, a state reasserting itself as an economic actor, an opaque financial periphery, a great-power resource competition, and an energy system that cannot quite keep up. All of it has happened before. The question is whether the institutional response this time will be as effective as the one that followed 1907, or whether the architecture will be built only after a much larger dislocation forces it.
The library on Madison Avenue is gone. The function it performed has not been replaced. It has been distributed across a handful of corporate balance sheets, a few sovereign wealth funds, and the working memory of perhaps two dozen people in Washington, Beijing, Riyadh, and Brussels. That is not a system. It is an interregnum. The question is what comes next, and who builds it.
Intelligence Report covers the intersection of capital, technology, and policy at the level of state and corporate decision-making. This briefing draws on public market data, regulatory filings, and recent reporting from EY-Parthenon, Goldman Sachs, the Financial Stability Board, the Council on Foreign Relations, and the Geopolitical Monitor, among others.
