Where Capital Goes Next

Start with the largest capital migration in the modern economy. The scale of money moving into artificial intelligence is often reported as venture capital, and the venture numbers alone are striking: AI companies raised roughly $212 billion globally in 2025, nearly half of all venture funding worldwide, according to Crunchbase. OpenAI's $40 billion round in March 2025 was the largest private financing ever arranged at the time; a year later the company closed a $122 billion raise at an $852 billion valuation, a deal so large that its biggest checks came not from venture firms but from Amazon, Nvidia and SoftBank.

But venture capital is the smallest layer of the stack. The heavier money sits in capital expenditure. Microsoft, Alphabet, Amazon and Meta together spent roughly $400 billion on capital investment in 2025, most of it on data centers and the chips inside them, and their guidance for 2026 points well above $600 billion. Goldman Sachs Research projects the four will spend a combined $5.3 trillion between 2025 and 2030. McKinsey estimates the world may need $6.7 trillion of data center investment by 2030, counting the chips, the land, the buildings, the cooling and the power systems.

Follow one dollar of that spending and it fans out across an entire industrial economy. It buys accelerators: Nvidia's data center business alone generated $193.7 billion of revenue in its last fiscal year, up 68 percent. It buys fabrication capacity: TSMC lifted its own capital budget from $40.9 billion in 2025 toward more than $50 billion in 2026. It buys steel, concrete, transformers, fiber and the labor of tens of thousands of electricians and pipefitters.

And increasingly, it is borrowed. Hyperscalers issued roughly $121 billion of investment-grade bonds in 2025, more than four times their five-year average. Meta's $30 billion October bond sale, the largest US corporate offering of the year, drew $125 billion of orders; separately, the company financed its Louisiana mega-campus through a roughly $30 billion private-credit joint venture with Blue Owl, the largest private-capital transaction on record. A market for securities backed by data center leases, which barely existed in 2020, passed $26 billion of issuance in 2025. Morgan Stanley expects private credit to supply about $800 billion of the financing the buildout still needs.

The lesson in these numbers is a change of category. For thirty years, software was the asset-light business: a few laptops, some cloud credits, gross margins near ninety percent. The AI era inverted that. Frontier intelligence now requires gigawatt campuses, five-year construction schedules and debt structures borrowed from the pipeline and power industries. Intelligence became infrastructure. And infrastructure is financed differently than code: with more capital, longer horizons, and lenders alongside the believers.

Here is a rule worth keeping: capital moves toward a breakthrough first, then toward whatever constrains it. The railroads created a steel boom. The automobile created an oil boom. The data center is creating an electricity boom.

The arithmetic is simple. A modern AI campus draws as much power as a mid-sized city, and there are hundreds under construction. US data centers consumed about 4.4 percent of the country's electricity in 2023; the Department of Energy's Lawrence Berkeley National Laboratory projects between 6.7 and 12 percent by 2028. Goldman Sachs forecasts data center power demand growing 160 percent by 2030. After two decades of essentially flat American electricity consumption, demand is rising again, and the constraint on building intelligence has become the ability to generate and deliver power.

Money noticed. The International Energy Agency counted $3.3 trillion of global energy investment in 2025, the most ever recorded, with $1.5 trillion of it flowing to the electricity system — half again more than the world spent bringing oil, gas and coal to market, a ratio that has fully inverted in a decade. In the United States, investor-owned utilities plan $1.1 trillion of capital spending between 2025 and 2029, nearly as much as they spent in the entire previous decade.

The most telling deals are the strangest ones. Microsoft signed a twenty-year power agreement to restart the Three Mile Island nuclear plant. Amazon contracted for 1.92 gigawatts from the Susquehanna station and anchored a $700 million round in the small-reactor developer X-energy; Nvidia's venture arm and Bill Gates put $650 million into TerraPower. Meta locked up the output of an Illinois nuclear plant for two decades. Buyers of gas turbines now wait years: GE Vernova's order book and paid reservations reached 116 gigawatts by mid-2026, with slots being reserved for delivery in 2031. Even the humble transformer has become a chokepoint, with lead times averaging 128 weeks.

Institutional money is arriving in size to match. Brookfield closed a $20 billion energy-transition fund, the largest ever raised; KKR closed a record $19.2 billion infrastructure fund this August with data centers named as a priority. None of this capital is betting on artificial intelligence directly. It is betting on the thing artificial intelligence cannot happen without, which is often the safer side of a boom.

For most of the 2010s, the venture industry had a revealed preference: bits over atoms. Software scaled; factories did not. That preference is now reversing, and the reversal shows up in the money before it shows up anywhere else.

Defense technology is the clearest case. Venture investors put $49.1 billion into defense startups in 2025, up from $27.2 billion the year before, according to PitchBook — the sector's best year on record. Anduril, the sector's flagship, was valued at $30.5 billion in June 2025 and $61 billion eleven months later, after revenue doubled to $2.2 billion. Europe produced its own version: Munich's Helsing reached an $18 billion valuation in the continent's largest defense-startup round. Behind the private money stands an enormous public commitment: world military spending reached roughly $2.9 trillion in 2025, and NATO members have pledged to reach five percent of GDP by 2035. Whatever one thinks of the causes, the capital formation is real, and startups are being funded to compete for it.

Space tells a similar story. Private investment in space companies hit a record $55.3 billion in 2025, per Space Capital, and SpaceX — valued at roughly $350 billion at the start of the year — was marked near $800 billion by December's insider share sale, making it the most valuable private company in the world. Robotics venture funding rose about 70 percent to roughly $14 billion, led by a $39 billion valuation for the humanoid-robot maker Figure. And beneath all of it runs a manufacturing buildout: chipmakers have announced more than $540 billion of US fabrication investment, and American factory construction has run at roughly triple its 2021 pace.

These companies share a financial signature. They need vastly more capital than a software startup, they take years longer to prove out, and when they work, they own something a competitor cannot copy in a weekend. Software spent decades digitizing the world. The next wave may have to rebuild parts of it — and rebuilding is expensive, which is precisely why the money is moving there.

Every other capital flow in this story purchases something you can photograph. A data center financing buys servers and switchgear. A factory financing buys machines. A biotechnology financing buys something stranger and more fragile: another chance to find out whether an idea about the human body is true.

That is the essential mechanics of biotech capital, and it is worth slowing down for. A drug company begins with a molecular hypothesis — that blocking this receptor, or silencing this gene, will change the course of a disease. Capital buys the preclinical work. If the evidence holds, more capital buys the first human dose. Then Phase I, for safety. Then Phase II, for a signal of efficacy. Then Phase III, at enormous cost, for proof. Then regulatory review, then manufacturing, then a medicine. At every stage, the financing purchases the next experiment, and the experiment converts uncertainty into evidence — in either direction. An early-stage asset can be simultaneously worth billions and worth nothing; each successful trial collapses that range. In biotechnology, money buys time and probability.

Right now, the money is unusually discriminating. Biopharma venture funding was essentially flat in 2025 at $33.8 billion across 1,171 deals, per PitchBook, but it concentrated hard: fewer, larger rounds for later-stage companies with clinical data, while early-stage firms struggled. The IPO window nearly closed — roughly $3 billion raised across about fifteen US listings, a ten-year low, against more than a hundred IPOs at the 2021 peak. Investors who once funded platforms and promises now want proof.

Where the capital is concentrating tells you what it believes. Obesity and cardiometabolic disease sit at the center: Eli Lilly's two incretin drugs alone generated about $36.5 billion in 2025, the global market for obesity medicines reached roughly $66 billion, and Morgan Stanley has projected the class could approach $150 billion by 2035. A successful oral pill — Lilly's orforglipron cut body weight by 12.4 percent in Phase III — would widen the market further. Neuroscience re-rated after Johnson & Johnson paid $14.6 billion for Intra-Cellular Therapies and its psychiatric franchise. Radiopharmaceuticals, which deliver radiation directly to tumors, have become a dealmaking magnet. AI-enabled discovery drew its first mega-rounds, led by Alphabet's Isomorphic Labs. Meanwhile cell and gene therapy — the most celebrated science of the last cycle — saw venture funding fall by more than sixty percent from its 2021 peak, with a wave of closures. The science did not stop working; the economics of one-time cures for small populations did not yet work. Selectivity cuts both ways.

Then there is the force compressing all of it: Big Pharma's calendar. The industry faces one of the largest patent cliffs in its history, with EY estimating roughly $180 billion of top-twenty-company revenue exposed to expirations through 2028. Merck's Keytruda, the world's best-selling drug at over $30 billion a year and roughly forty percent of the company's revenue, begins losing US exclusivity in 2028. Revenue that disappears must be replaced, and it is far faster to buy evidence than to grow it. Biopharma companies signed $240 billion of acquisitions in 2025, up 81 percent from the year before, per EY: Pfizer outbid Novo Nordisk in a rare public duel for the obesity startup Metsera; Novartis paid $12 billion for Avidity Biosciences; Merck bought Verona for $10 billion. A parallel market boomed in licensing, with Chinese biotechs signing partnerships carrying a record $135.7 billion in potential deal value — a number mostly made of milestones, but with upfront cash payments quintupling since 2022.

The result is a capital chain that runs, link by link, from a pension fund's allocation decision to a pharmacy shelf:

It is worth stating plainly what sits at the end of that chain. The return on a successful biotech investment is financial, and should be judged that way. But the underlying asset is unlike any other in this story: a disease that becomes treatable, suffering that becomes optional, time that a patient gets back. No spreadsheet line captures that, and no honest account of these markets should ignore it.

Everything to this point describes money that moved. The more consequential fact is the money that did not. There is always more ambition than capital: more companies than investors, more hypotheses than trials, more startups than there will ever be successful businesses. In 2025, over a thousand biopharma companies won venture backing; about fifteen reached the public markets. Gene therapy pioneers shut their doors in the same quarters that obesity startups sparked bidding wars. For every gigawatt of contracted power, other projects sit in interconnection queues that now hold more than 2,500 gigawatts of proposed generation worldwide.

Capital allocation, in other words, is a selection mechanism — the largest one humanity operates. Every day, investors decide: which founder, which drug, which reactor design, which factory, which price, which risk, which moment. Markets are constantly ranking competing versions of the future, promoting some to existence and quietly declining the rest.

The mechanism is fallible, and its fallibility is the point. Some of today's allocations will look brilliant in a decade; others will destroy capital on a historic scale, as every buildout from railways to telecoms has managed to do. But that uncertainty is not a defect of the system. It is why the returns exist. An investor who demands certainty is describing a bond. Everything in this story — the data centers, the reactors, the rockets, the molecules — is being financed precisely because nobody yet knows how it ends.

Intelligence. Power. Machines. Space. Medicine. On an exchange, these are different sectors with different tickers and different analysts. Stand further back and they are one phenomenon: enormous pools of capital committing to projects whose ultimate value cannot be known in advance. A frontier model, a small modular reactor, an autonomous aircraft and a Phase II molecule have more in common with each other than any of them has with the asset-light software that defined the last cycle. All of them require the money first and deliver the answer later.

That is the thread running through every chapter of this story. The future is not merely invented. It is financed — argued over at those tables where one side has an idea and the other has capital, priced in term sheets and bond indentures and merger agreements, and only then, sometimes, built.

Every generation inherits a finite pool of capital. Thousands of people compete for it. Some want to build intelligence. Some want to generate energy. Some want to manufacture machines. Some want to reach space. Some want to cure disease. Investors decide which of those ambitions receive the resources to become real.

Capital is a vote on the future. Where it goes next will help determine what the future becomes.