The Other
Printing Press

01 · The 2,000-foot viewMoney is not the harvest.

The farmer’s loan does two things at different times. It gives workers money to spend today. It pays for trees that can produce fruit later.

That is why useful investment can still raise prices while it gets built. The workers need food before the orchard delivers its next harvest.

Once the trees bear fruit, the market has more to sell. That can ease price pressure compared with a world where spending rose but the harvest never grew.

The money arrives first. The harvest has to catch up.

There is no guarantee that it will. Trees can fail. Costs can rise. Buyers can keep spending more. Calling the loan “productive” tells us what the farmer hopes to do, not what the orchard will deliver.

The useful question is how much extra output the investment makes possible, and when people can buy it.

02 · One small changeSometimes the extra oven is already there.

Some businesses can make more without waiting for anything to be built. Consider a bakery with an unused oven, an available baker, and enough flour and electricity for another shift.

Economists call this unused room slack. An empty hotel room, an idle machine, or a worker who wants another shift can be part of it.

When there is slack, extra spending can bring resources into use. When everyone and everything is already busy, it is more likely to bid up prices, draw in imports, or displace someone else’s activity.

The same dollar has a different effect depending on what is available when it arrives.

03 · The 20,000-foot viewEvery economy has two printing presses.

One press creates money and financial claims. The other is the economy’s ability to turn work, materials, energy, and ideas into useful things.

The first can move quickly. A bank can approve a loan and credit an account. The second has to deliver the food, home, medicine, or software that someone actually needs.

What matters is spending, not just balances in bank accounts. Expectations, trade, exchange rates, and monetary policy also affect inflation. A central bank can respond to productivity growth by allowing more real growth while keeping inflation near its target.

Knowledge changes what is worth building.

David Deutsch’s account of resources starts with knowledge. Silicon becomes useful in a computer chip because we learn how to transform it, then build the equipment to do so. A higher price for sand does not perform that transformation.¹¹

AI could make existing work cheaper. It could also put specialized software, design, and research within reach of businesses that could never justify their cost before.

Imagine a tool that would help a small grower detect crop disease. If developing and running it costs more than the grower could gain, the project stays an idea. If those costs fall enough, someone may build it. A service that did not exist becomes possible.

AI could make each project cheaper while making far more projects worth attempting.

That is the larger possibility. The economy’s capacity to put money to useful work can change as knowledge improves. Whether total investment rises depends on how many new opportunities appear and how much they cost to pursue.

04 · Follow the first paymentWho actually creates the money?

“Money printing” often bundles together three different operations. Keeping them separate helps us follow who can spend, and why.

Bank loans create deposits. The farmer can receive newly created money and owe a matching debt. Repaying the principal reduces deposit money.¹

Government borrowing finances a deficit. It is not, by itself, the same operation as creating central-bank money. QE is a central-bank asset purchase. It changes the assets people hold and can ease financing conditions, but does not order anyone to plant a tree.¹⁰

New opportunities can also lead to new money. A promising project gives an entrepreneur a reason to borrow and a bank a reason to lend. The resulting loan creates a deposit. Technology can help create demand for financing, as financing helps people use the technology.

05 · The destinationWhat did the new dollar actually buy?

To understand what new money does, follow what it pays for.

Buying a business and building a business are not the same physical event. A purchase can lead to better management, fresh investment, or useful reallocation. But the transfer of ownership alone does not add an oven.

A high return on investment does not always mean more production. A person can make an excellent return buying scarce land that becomes more expensive. Society has not necessarily gained another home.

Meanwhile, a research project could generate discoveries that benefit everyone while earning little for the investor who funded it.

A high private return can come from scarcity. A productive return reduces scarcity.

A project can deliver several of these returns, or do well on one and poorly on another. A rising valuation is not proof that the economy has become better at producing things people need.

06 · The missing time dimensionGood investment can raise costs first.

Suppose the spare oven does not exist. The baker borrows to build a new bakery, then orders bricks, hires electricians, and buys equipment. The payments begin well before the extra bread arrives.

Useful compared with what?

The same bakery could be financed with existing savings. It might have been built anyway. A good project is not, on its own, a reason to create more money.

What gets built because of this financing, after accounting for what it displaces? That is the comparison that matters. Existing savings and newly created money can both finance investment, but their effects on total spending can differ.

CBO’s infrastructure analysis shows why financing matters. Borrowing can boost demand early, while productivity benefits arrive gradually. Some private investment can also be crowded out.³

“Inflationary now” and “productive later” can both be true.

Debt adds another deadline. Interest and repayments may begin before the bakery earns revenue. Even a useful project can be a poor borrower if its debt costs too much or comes due too soon.

Nor does public investment automatically pay for itself. More economic activity can benefit society without generating enough extra tax revenue or government savings to cover the debt.

07 · A necessary objectionWe tried “just lend to productive things.”

An old idea called the real bills doctrine argued that money could safely expand with short-term lending against genuine commercial activity.

A baker borrowed to buy flour, sold the bread, and repaid the loan. Money expanded when commerce needed it and contracted when the sale was complete. It looked self-regulating.

The flaw was that a legitimate commercial invoice could grow because prices rose, without more goods being produced. Financing every such invoice could feed a boom. When orders fell, the same rule could shrink credit and deepen the slump. The Federal Reserve’s historical account describes this procyclical problem.⁴

We need to judge productive lending at both the project level and across the economy: is the money financing useful future output, and is total spending outrunning what the economy can deliver now?

Useful projects can still add to an economy-wide spending boom.

08 · How new opportunities appearAI can make the next project possible.

There are three ways to expand what we can produce: use idle resources, build better equipment, and learn how to do something new. They can reinforce each other. A better idea helps us design better tools, which let us test more ideas.

AI could matter across all three. Its most interesting effect may be making projects viable that were previously too expensive to attempt.

Cheaper development could let a company serve a small market it previously had to ignore. Cheaper design could make a specialized product worth manufacturing. Better research tools could help a team explore a promising idea within its budget.

Some gains improve existing businesses. Others allow new ones to start. This is how new knowledge can expand the set of useful things we can make.

There is a catch. Cheaper projects need less money each. Total investment rises only if the number and scale of opportunities more than offset that saving. And software does not remove every constraint. Someone still has to install the sensor, connect the power, and persuade customers to pay.

09 · Check it against the worldWhy 2008 and 2020 were different.

The amount of central-bank money alone cannot explain what happened to prices.

The emergency response protected incomes and helped prevent worse damage. Its value and its inflationary consequences are separate questions.

Low inflation does not prove that money was invested productively. It may simply mean that little of it was spent.

Investment can also become less useful. China’s earlier growth benefited from capital investment and reallocation, while weaker productivity growth and poorer allocation later reduced the payoff.⁷ The next road can be much less valuable than the first.

AI may open new opportunities, but it will still have to pass the same test: what becomes possible once the investment is finished?

10 · The bet and the billWhat would make the AI boom worth it?

The promise is a larger set of worthwhile projects. The first bill is for very ordinary things: chips, electricity, cooling equipment, construction crews, and grid connections.

Those orders add demand before most customers see the benefit. Data-center electricity use is a concrete part of the buildout.⁸

Even optimism can bring spending forward. If firms expect AI to make future investments more profitable, they may try to build now. The supply gains still need time to arrive.

The middle outcome deserves attention. A company can build useful infrastructure and earn a poor return because too many competitors build it too, or because customers capture the benefit through lower prices. Barr’s discussion of AI points to earlier infrastructure booms with this mix of lasting usefulness and investor losses.⁹

That brings us back to the four kinds of return. Society may gain while some shareholders lose. A lender still needs its particular borrower to generate cash. A government still needs revenue or savings to service its debt.

For AI’s benefits to spread, firms need reliable tools, supporting equipment, and changes in how people work. Cheaper software helps less when the main problem is a grid connection, a housing shortage, or a permit.

A model can help design a power plant faster than we can permit and build one.

AI could improve the return on other investments by making design, discovery, coding, and coordination cheaper. But the useful output has to reach customers, and the spending required to get there has to fit the economy’s resources.

The AI bet is that more useful possibilities become real, broadly enough and soon enough to justify the cost of pursuing them.

11 · A better set of questionsUnderwrite the harvest.

Alongside how much money is created, I would ask five questions.

1. Who gets to spend it first?

   Follow the payment from the bank, investor, or government to the person who actually spends it.

2. What becomes possible because of it?

   Ask what gets produced that would not otherwise exist, and what other activity is displaced.

3. What is the actual bottleneck?

   Money may be missing. Or it may be a worker, transformer, permit, customer, or idea.

4. When will people get the output?

   A financed plan, a construction site, and a finished product are different milestones.

5. Who gets paid, and who bears the risk?

   Separate benefits to customers from cash available to repay debt. Check what happens if the payoff is late or smaller than expected.

Productive investment does not make borrowing harmless. It gives us a reason to examine what the borrowing could accomplish, what it costs, and whether the borrower can survive the wait.

AI makes that examination more interesting. Some projects may get cheaper. Some businesses may become possible for the first time. Some investments will disappoint even if the technology succeeds.

Back at the orchard, financing can pay for planting. Knowledge can help us grow more from the same land. The extra apples still have to arrive.

We keep arguing about how much money to print. The harder question is what we can grow with it and how long the harvest takes.