The Singularity Has a Balance Sheet

A founder said something to me recently that stuck.

He was raising money for an AI company, and the pitch was good. Maybe too good. The product was working, customers were interested, the demo had that little electric feeling every investor wants to feel before writing a check. Then he got to the market slide and said, almost casually, “If models keep improving at this pace, this becomes a labor market replacement.”

He meant it as upside.

Everyone in the room understood the sentence in two ways at once.

As a startup pitch, it meant the company might be huge. As a description of the economy, it sounded like a grenade rolling under the table.

That is the strange mood right now. AI is being discussed like software, priced like infrastructure, funded like a gold rush, and feared like a constitutional crisis. One minute the conversation is about revenue per customer. The next it is about whether work, wages, interest rates, sovereign debt, and money itself are about to be pulled into a different gravitational field.

People feel the ground moving, then reach for the biggest available words. Singularity. Collapse. Bubble. End of capitalism.

The end of capitalism is probably the wrong frame. Capitalism has a long history of swallowing its supposed replacements and issuing them a subscription business model. AI fits that pattern for now. The first winners are chip companies, cloud providers, model labs, data-center operators, energy suppliers, software incumbents, and firms with distribution. In plain English, capital is winning.

The sharper question is whether the version of capitalism we have been living inside can survive the arrival of machine intelligence without rewriting its financial plumbing.

Because the old plumbing is already under pressure.

For forty years, the rich world got used to falling rates. That sounds technical. It was everything.

Falling rates made houses more expensive and monthly payments feel manageable. They made stocks more valuable because future profits were discounted at cheaper prices. They let governments borrow without voters feeling the cost. They made venture capital feel rational even when startups burned money for years. They allowed private equity to buy companies with debt and call the arithmetic strategy. They trained everyone to believe the future could be pulled into the present and financed.

Then inflation came back.

It did not return as a theory. It returned as rent, groceries, insurance, energy bills, wages, mortgage payments, and government interest expense. Suddenly money had a price again. Time had a price again. The future stopped being cheap.

That is why this moment feels larger than another rate cycle. AI is arriving at the same time the debt machine is losing its lubricant.

The uncomfortable part is that both forces point in opposite directions. AI promises abundance. The bond market is pricing scarcity. AI says output can explode. Debt says tomorrow has already been spent. AI says labor may become optional in more parts of the economy. Governments still tax wages, payrolls, income, consumption, and corporate profits as if the industrial economy remains the template.

The story gets messy fast. Good. The clean versions are usually wrong.

The low-rate world made everyone a little delusional

Low interest rates do something to judgment. They make distant possibilities feel close.

A dollar ten years from now is worth more today when rates are low. That single fact shaped the whole financial culture of the last decade. Growth stocks, venture funds, crypto tokens, real estate, private credit, and government deficits all benefited from the same background condition. If the cost of waiting is low, dreams get expensive.

Startup investing became one of the purest expressions of that world. A seed company with no profit could raise at a high valuation because the market was willing to pay for a story that might become real five or ten years later. The story did not need to be fake. Many were real. The issue was timing. Low rates compressed the distance between promise and proof.

Governments behaved in a similar way. They did not pitch investors with product demos, but they sold the same basic idea. Lend us money now. Growth, credibility, and future tax receipts will take care of the rest.

For a long time, markets accepted the pitch.

That pitch becomes harder when rates rise.

A government with high debt can function when interest costs are low. It can even look prudent for a while. The danger shows up when old debt rolls into a higher-rate world. The country does not need to default for the pressure to become real. Interest payments crowd out other spending. Voters dislike tax increases. Politicians dislike spending cuts. Central banks dislike inflation. Bond markets dislike fiscal denial.

Everyone wants someone else to absorb the pain.

This is where Japan matters.

Japan spent decades as the example people waved around when they wanted to prove that large sovereign debt could be sustained. They had a point. Japan carried enormous debt, weak demographics, and low growth without the clean collapse predicted by simple textbooks. Domestic savings, institutional bond ownership, low inflation, and central-bank credibility gave the system room.

Now Japan is more interesting as a warning label. Once an economy builds itself around ultra-low rates, leaving that world becomes dangerous. Higher rates can support currency credibility and fight inflation, but they also increase the state’s debt-service burden. The central bank becomes trapped between price stability and fiscal arithmetic.

That trap is not uniquely Japanese. Japan just got there earlier.

The United States has more growth, better demographics, deeper capital markets, and the global reserve currency. Those advantages are enormous. They are also easy to abuse. A reserve currency is a privilege until politicians start treating it like a magic trick.

The old assumption was that developed-market sovereign debt sat at the bottom of the financial system as the safest asset. Increasingly, it looks more like the central political asset. The risk is less about formal default and more about repayment through inflation, financial repression, taxation, currency weakness, or pressure on domestic institutions to absorb government debt at unattractive real returns.

That sounds abstract. It means savers pay. Sometimes slowly. Sometimes without a vote.

AI enters the room at the worst and best possible time

If you wanted to design a technology that could rescue indebted economies, you would design something like AI.

Debt becomes easier to handle when real growth rises. More output means more income. More income means more tax revenue. More tax revenue makes interest costs less painful. If AI lifts productivity enough, a lot of terrifying debt charts start to look less terrifying.

That is the optimistic case. Intelligence has always been the scarcest input. Make intelligence cheap and everything changes. Drug discovery accelerates. Software gets easier to build. Education improves. Customer support becomes cheaper. Manufacturing gets more automated. Scientific research speeds up. Small teams do work that once required entire departments. Governments deliver services with less waste. Healthcare gets less broken. Maybe.

The word “maybe” is carrying a lot of weight.

AI has to pass through the real economy before it saves the fiscal state. The real economy is full of stubborn objects. Power grids. Hospitals. Procurement departments. Lawsuits. Unions. Insurance codes. School boards. Permits. Data silos. Legacy software. Human fear. Human vanity. Human incompetence in high places.

Anyone who has watched a large company buy software knows the problem. The demo takes twelve minutes. The deployment takes twelve months. Sometimes longer. Then half the team refuses to use it because the old spreadsheet still feels safer.

This is why the AI boom can be real and overhyped at the same time. That combination is common in major technology shifts. The technology changes the world. The first wave of financial claims on that change still gets too excited.

The internet was real. The dot-com bubble still burst.

AI may follow a similar pattern, with larger numbers and more political heat. The infrastructure buildout comes first. Chips, cloud capacity, data centers, grid connections, cooling systems, software tooling, security layers, enterprise pilots. Money pours into the rails before anyone knows the final shape of the train schedule.

The early AI economy is less like magic abundance and more like a massive construction project with better marketing.

That matters for inflation and rates. People talk about AI as if it will make everything cheaper. Over time, it might. In the early phase, it demands scarce inputs. Advanced chips. Electricity. Land. Engineers. Transformers. Fiber. Water. Specialized construction. Political permission.

Scarce inputs do not become cheaper because a slide says “exponential.”

So the transition can be inflationary before it becomes deflationary. Companies spend heavily. Utilities strain. Supply chains tighten. Workers with relevant skills become expensive. Asset prices rise because investors believe the prize is huge.

Then everyone looks around and asks why the abundance feels so costly.

The bubble is built into the structure

AI is perfect bubble material because the upside is huge and hard to bound.

A normal business has a market. AI keeps trying to become the market underneath other markets. Software, services, education, law, medicine, finance, advertising, entertainment, coding, sales, operations. Every category can be reimagined if machine intelligence gets cheap enough and reliable enough.

Investors hate missing platforms. They especially hate missing platforms after seeing the last generation of platform companies become some of the most valuable firms in history. That memory changes behavior. It makes capital faster, more anxious, and more forgiving of vague answers.

The most dangerous sentence in a financing meeting is “this could be the interface for all work.” It may be true. It may also justify almost any valuation if said with enough conviction and accompanied by a clean demo.

The AI market has another bubble ingredient: reflexivity.

A company valued highly can raise more money. More money buys more compute and talent. More compute and talent improve the product. A better product attracts more users. More users improve distribution and data. The improved story raises the valuation again.

For a while, the financial loop and the product loop reinforce each other. It feels like destiny from the inside.

Then the boring questions arrive.

How much will customers pay?
How much gross margin survives when inference costs are included?
Who owns the workflow?
Can open-source models compress pricing?
Does the product reduce headcount or just create another software line item?
Will enterprises trust it with sensitive decisions?
What happens when every competitor has similar model access?

Experienced investors ask these questions early because they have seen the movie where a product is impressive and the business is mediocre. Intelligent newcomers often underestimate this distinction. A demo can be magical while the profit pool remains slippery.

AI makes that distinction harder because the demos are so good. The machine talks back. It writes code. It makes images. It drafts contracts. It answers customer tickets. It feels like labor on a screen.

But a company does not get paid for sounding like the future. It gets paid for controlling something valuable that customers cannot easily get elsewhere.

That control might come from distribution, proprietary data, deep workflow integration, regulatory trust, brand, switching costs, hardware access, or ownership of the customer relationship. Without control, AI features become table stakes. Table stakes are useful. They do not always create great businesses.

This is one reason AI can inflate public markets and disappoint private investors at the same time. The biggest incumbents may capture much of the value because they already own the customers, infrastructure, and balance sheets. Startups create the heat. Platforms collect the rent.

The labor problem is the real problem

People talk about money supply because it is visible. Central-bank balance sheets have charts. Interest rates have announcements. Debt has numbers with commas.

The deeper issue is income.

Modern capitalism works because most people get purchasing power by selling labor. They work, earn wages, buy goods and services, pay taxes, and support the state. The system has plenty of unfairness, but the loop is understandable.

AI threatens to weaken that loop.

The first affected workers are not only factory workers or call-center workers. They include programmers, analysts, designers, marketers, lawyers, consultants, support reps, junior bankers, copywriters, recruiters, accountants, tutors, and operations managers. The polite phrase is “productivity enhancement.” The less polite version is that many white-collar tasks are being repriced.

The job does not need to disappear for wages to come under pressure. If one person with AI can do the work of three people, the labor market changes. If junior tasks get automated, the training ladder changes. If software becomes easier to build, some engineering work becomes more valuable and some becomes more commoditized. If every company can produce more content, content gets cheaper. If every analyst has a tireless assistant, basic analysis loses scarcity.

This is where the social contract starts to creak.

An economy can grow while many workers feel poorer. Asset owners can become richer while wage earners feel replaceable. Stock indices can rise while households feel trapped by rent, insurance, childcare, healthcare, and debt.

That combination is politically explosive because people do not experience GDP. They experience cash flow.

If AI raises output while reducing labor’s share of income, governments will need new ways to distribute purchasing power. That does not require a utopian ideology. It follows from arithmetic. If wages become a weaker channel for mass income, some other channel has to carry more weight.

That could mean larger earned-income credits, wage subsidies, public services, universal basic income, sovereign wealth funds, public stakes in AI infrastructure, heavier taxes on monopoly profits, taxes on capital gains, or new forms of citizen dividends.

The exact policy will vary by country. The direction is harder to avoid. A labor-light economy needs a broader income-distribution mechanism or it becomes socially unstable.

This is where the “end of capitalism” conversation gets confused. AI does not automatically end private ownership. In the early phase, it strengthens private ownership. The owners of compute, energy, distribution, models, and capital gain more leverage.

The pressure comes later, when voters ask why the machine produces abundance and their lives still feel financially narrow.

GDP may start lying more than usual

Gross domestic product measures market activity. It is useful. It is also clumsy.

AI will make it clumsier.

Suppose an AI tutor gives every child access to personalized instruction at near-zero cost. Human welfare could rise enormously. Measured GDP might barely notice if the service is cheap or bundled into a subscription.

Now suppose companies spend hundreds of billions on data centers, chips, consulting, and enterprise software, then use much of it to generate ads, spam, synthetic media, and internal slide decks nobody reads. GDP rises. Human welfare may not rise much.

Both outcomes can happen at once.

This matters because debt lives in the measured economy. Governments collect taxes from transactions, wages, profits, capital gains, property, and consumption. Bond markets care about nominal growth, inflation, and fiscal receipts. They do not give a government full credit for consumer surplus that never turns into taxable cash flow.

A free AI doctor would be socially extraordinary. It helps the state’s finances only if it reduces public healthcare costs, creates taxable income, or replaces expensive labor in a way the government can capture.

That last phrase is ugly because the topic is ugly. Fiscal systems need claims on value. If value migrates into cheap digital abundance, untaxed capital gains, offshore profits, open-source models, or consumer surplus, the state can become fiscally stressed while citizens are surrounded by powerful technology.

A country can be technologically rich and fiscally weak.

That is one of the strange possibilities ahead.

Money will still matter

The dream version of the singularity says money fades because abundance wins. Maybe in the distant limit. Before that, money may matter more because claims on the future will become more contested.

Money is a claim on output. Debt is a claim on future money. Equity is a claim on future profits. Wages are claims earned by selling labor. Government bonds are claims on future taxpayers. If AI changes production, every claim built on top of production starts to move.

That is why the topic feels so large. It is not just about better chatbots or faster coding. It is about who gets paid when intelligence becomes software.

Money supply matters, but the next phase may depend even more on collateral. Modern credit systems lend against assets. When AI inflates the market value of certain assets, it expands the collateral base. Companies can raise more capital, borrow more cheaply, pay employees in stock, acquire competitors, and fund larger infrastructure plans.

The asset price becomes part of the engine.

This can work beautifully while confidence holds. Rising valuations fund real investment. Real investment improves products. Better products support the original valuations. Everyone feels brilliant.

If expectations break, the loop runs in reverse. Equity falls. Credit tightens. Capex slows. Tax receipts weaken. Governments respond with support. Central banks feel pressure. The private bubble becomes a public problem.

This is the part people tend to miss. Large asset bubbles rarely stay private. Once enough jobs, tax receipts, pensions, retirement accounts, and infrastructure plans depend on inflated values, the state gets dragged in.

AI could produce the largest version of that story because the investment needs are enormous and the promised prize is civilizational.

Interest rates could go either way

The simple view says AI means deflation, so rates should fall.

That may be right eventually. It is a weak guide for the transition.

In the near term, AI can push rates higher by increasing investment demand. Everyone wants compute. Governments want national AI capacity. Militaries want autonomy and intelligence systems. Companies want automation. Cloud providers want data centers. Data centers want power. Power wants grids, permitting, turbines, substations, transformers, and fuel.

That is capital demand. Capital demand can push real rates up.

AI can also create inflation through wealth effects. If asset owners get richer, they spend more. If companies invest aggressively, demand rises. If electricity and commodities become bottlenecks, prices rise. If governments subsidize domestic AI infrastructure, deficits widen.

Over time, AI may push in the other direction. Better software lowers costs. Better science improves energy. Better robotics reduce manufacturing costs. Better logistics reduce waste. Better education raises human capability. Better medicine reduces expensive failure.

So the path could be high-pressure first, disinflation later.

Markets are tempted to price the later world before the current one has paid for the buildout. That is how long-duration bubbles form. Investors discount the promised abundance while the economy is still buying the servers.

Sovereign debt becomes a bet on productivity

Government debt is usually discussed as a moral drama. Too much spending. Too little discipline. Irresponsible politicians. Entitled voters.

Some of that is true. It is also incomplete.

Debt sustainability depends on growth, rates, inflation, and political capacity. A country can carry high debt if growth is strong, rates are manageable, and investors trust the system. A country can struggle with lower debt if credibility breaks.

AI enters as the great productivity bet.

If AI meaningfully raises real growth, governments get relief. More income. More profits. More taxable activity. Lower unit costs. Better services. Maybe smaller deficits.

If AI disappoints, debt math becomes harsher. If AI concentrates gains in companies and individuals that avoid taxation, governments may receive less relief than the productivity headlines suggest. If AI disrupts labor income faster than fiscal systems adapt, deficits can widen even as technology improves.

That is the dangerous middle outcome. The economy gets smarter. The state gets poorer. Politics gets angrier.

The fiscal question becomes less about whether AI creates value and more about whether public institutions can capture enough of that value to fund obligations.

The United States has a particular advantage here because many leading AI companies, chip designers, cloud platforms, and capital markets sit within its orbit. That gives the U.S. a better chance of taxing, regulating, and benefiting from the boom. It also creates complacency. Dominant countries often mistake structural advantages for permanent exemption from arithmetic.

Arithmetic has a long memory.

The new bottlenecks

Every age has its bottlenecks.

In the industrial age, machinery, oil, labor, factories, ports, and finance mattered. In the software age, talent, code, distribution, and network effects mattered. In the AI age, the bottlenecks look different.

Compute matters.
Energy matters.
Data matters.
Distribution matters.
Trust matters.
Capital matters.
Regulatory permission matters.
Ownership matters.

That last one may matter most.

If AI becomes a general-purpose production engine, then owning the engine becomes the central economic fact. A person with no ownership may get cheaper services and worse bargaining power. A person with ownership may get compounding claims on machine output.

This is why the singularity debate can sound mystical while the practical question is brutally simple. Who owns the productive assets?

If ownership stays concentrated, AI becomes a machine for widening the gap between capital and labor. If ownership broadens through public markets, pensions, sovereign funds, employee equity, public investment vehicles, or tax-and-transfer systems, AI can support a wider prosperity.

The technology does not decide that. Institutions do.

And institutions are slow. Slower than model releases. Slower than markets. Slower than fear.

What experienced investors are watching

The public conversation loves dramatic questions. Will AI take all jobs? Will money disappear? Will capitalism end?

Experienced investors tend to ask more grounded questions because grounded questions reveal where the money goes.

They ask whether customers are paying more because the product is useful or because the budget cycle is full of AI experiments.

They ask whether the company owns the customer relationship or sits as a feature inside someone else’s platform.

They ask whether gross margins improve with scale or get eaten by compute costs.

They ask whether the product replaces labor, increases labor productivity, or creates more work under a shinier label.

They ask whether the company has proprietary data that matters.

They ask whether the buyer can measure return on investment without pretending.

They ask whether a model upgrade from OpenAI, Anthropic, Google, Meta, or an open-source project can erase the company’s advantage.

They ask whether the founder is describing a business or narrating a future.

That last distinction matters. A future can be correct while a company inside it fails.

The same discipline applies at the macro level. AI can transform the economy while many AI investments lose money. AI can raise GDP while worsening inequality. AI can reduce prices in some categories while increasing pressure on energy and capital markets. AI can help sovereign debt over time while worsening deficits during the buildout.

The world can get richer and more unstable at the same time.

The end state is allocation

The more I think about the singularity economy, the less I think the central question is production.

Production is the fun part. More intelligence means more discovery, more automation, more optimization, more software, more science. It is easy to get excited about that because the upside is real.

The harder question is allocation.

Who gets the output?
Who gets the income?
Who pays the debts?
Who owns the machines?
Who taxes the rents?
Who absorbs workers whose tasks have been repriced?
Who controls the compute?
Who decides whether abundance arrives as lower prices, higher profits, larger transfers, or stronger monopolies?

Those questions sound political because they are. Every economic regime eventually becomes a political settlement. The AI regime will be no different.

Money will not vanish early in that process. It will become the scoreboard for a more intense fight over claims. Wages, profits, taxes, benefits, debt payments, capital gains, subsidies, and inflation will all become ways of deciding who receives the gains from machine intelligence.

The cleanest positive scenario looks like this. AI raises productivity. Growth improves. Governments capture enough tax revenue to manage debt. Competition and open-source models prevent permanent monopoly rents. Prices fall in important categories. Workers use AI to become more productive rather than broadly replaceable. Public policy updates the income system before social trust breaks. Energy supply expands fast enough to support the compute buildout.

That scenario is possible.

The darker scenario is just as easy to describe. AI raises asset prices faster than real productivity. Infrastructure spending booms. Labor income weakens. Sovereign debt costs keep rising. Governments become dependent on bubbly tax receipts. Incumbents capture the profit pools. Central banks face inflation they cannot cleanly fight. Voters grow hostile toward a system that produces technological miracles and household anxiety in the same year.

That scenario is also possible.

The likely path has pieces of both. A boom, a bust, real productivity, wasted capital, monopoly profits, open-source deflation, better tools, worse politics, sovereign intervention, and a long argument over ownership.

Messy. Human. Expensive.

What this moment really is

The singularity will not first arrive as a glowing line on a chart. It will arrive through budgets.

A cloud provider signs another power agreement. A government pays more interest on old debt. A startup raises money at a valuation that assumes a labor market will bend. A household refinances nothing because the rate is too high. A company freezes hiring because its AI tools are good enough. A central bank holds rates while politicians complain. A pension fund buys the index because it cannot afford to miss the only growth story large enough to matter.

That is how a new regime starts to feel real. Not as one grand announcement. As a thousand decisions that begin to rhyme.

So no, capitalism is not ending in some clean cinematic sense. The stronger claim is that the low-rate, labor-centered, debt-tolerant version of capitalism is under strain. AI gives it a possible escape route and a possible accelerant for its worst tendencies.

It can grow us out of debt or inflate the biggest asset bubble in history.

It can broaden access to intelligence or concentrate power around the owners of compute and distribution.

It can lower the cost of living or raise the value of assets faster than wages can follow.

It can make governments more capable or expose how poorly their fiscal systems fit a post-labor economy.

The mistake is expecting one answer.

For a while, AI will be productivity miracle, investment mania, fiscal hope, labor shock, and political accelerant all at once. That is why the moment feels so hard to name. We are trying to describe a technological transition using financial language built for a slower world.

The old economy asked how much labor and capital could produce.

The new economy asks what happens when intelligence itself becomes capital.

Once that question is live, everything downstream starts to move. Money. Debt. Rates. Wages. GDP. Asset prices. The state.

The singularity has a balance sheet.

We are only beginning to read it.