31 October, 09:31
An economic theory is worth something only if it can be tested against facts that arrive after the book is written. A theory that explains only the past is history. A theory that recognizes itself in next quarter's earnings reports is a tool.
In "The Economics of Losses" I described a stage that sets in during the autumn of a hegemon, once capital has exhausted both production and financial games. At that stage, income comes neither from creating wealth nor from claims on other people's income. It comes from loss: from spending that brings no commensurate gain in the capacity to create, and from risk shifted onto those who decided nothing. Among the sectors where this construction was already showing through, I listed information technology alongside construction, space and war. At the time, that may have sounded like an overstatement. Technology, after all, was supposed to be the opposite of autumn: a new material expansion, the spring from which, in Giovanni Arrighi's scheme, the next cycle of accumulation begins.
So the question is simple. If the most creative industry of the modern economy, the one promising the world a new industrial revolution, is growing by the rules of creation, then my claim about tech is wrong. If it is living by the rules of autumn, the markers confirming the Economics of Losses keep piling up. On October 29, 2026, the quarterly reports of America's largest companies answered that question, and Wall Street delivered the verdict itself.
That day, shares of Alphabet, Google's parent company, plunged more than 12%, erasing roughly $500 billion in market value, Reuters reported. For scale: the entire American bailout of the financial system in October 2008 cost $700 billion. Here, close to three quarters of that sum vanished in a single trading session, from a single company. Amazon lost more than 6% after hours. Microsoft rose more than 3% at first, then finished the day almost exactly where it started. Oracle, the day before, jumped nearly 11%, and not for profits or a new product, but for a promise to borrow less. Investors suddenly stopped asking how fast cloud services were growing and started asking something else entirely: can a company pay for its artificial intelligence with its own money, instead of going back to the market with its hand out every quarter?
The striking thing is that Alphabet is not losing money. It booked $26.2 billion in net profit for the quarter. But its free cash flow, the money a company keeps after paying for all its capital investment, went negative by $9.07 billion. A quarter earlier it stood at minus $5.9 billion, and even that, the first negative figure in the company's entire history as a listed firm, had already spooked the market. Now the hole is deeper. Capital spending for July through September more than doubled: $51.4 billion against $24 billion a year earlier. Operations brought in $42.3 billion. In other words, the company spent 121.4% of the cash it earned on infrastructure. The gap was filled with other people's money: in that quarter alone, as Alphabet's filing with the US Securities and Exchange Commission shows, it raised $12.4 billion by selling shares and $28.7 billion in bonds.
And AI does sell. Google Cloud doubled its revenue to $30.71 billion. But look at the proportion: $30.7 billion in quarterly cloud revenue against $51.4 billion in quarterly capital spending. Meanwhile the old cash cow is slowing. Google Search, which fed the whole company for years, grew just 12% instead of the expected 14.7%. The engine that produced real cash is losing power, and the new engine is being fueled on credit. Alphabet's chief financial officer, Anat Ashkenazi, is not promising to ease off: the capital spending plan for 2026 stays at up to $205 billion, and in 2027, she said, it will rise substantially.
At Meta, negative free cash flow is already a fact: minus $2.12 billion, because with $32.79 billion in operating inflows, capital spending reached $34.91 billion, according to Meta's third-quarter report. Its annual investment plan is $135-145 billion. Amazon spent $77.26 billion on infrastructure in a single quarter, and its trailing 12-month free cash flow, per Amazon's report, fell to minus $42.4 billion from plus $14.8 billion a year earlier. Its cloud unit, AWS, grew 39%, to $46 billion in quarterly revenue. The same proportion again: spending on hardware runs at almost twice the revenue of the business the hardware is bought for. And the same promise again: $220 billion in cash capital spending in 2026, and more in 2027.
Microsoft looks like the honor student of the quarter. It put $45.2 billion into computing infrastructure and still kept $18.6 billion in free cash flow, nearly $10 billion more than analysts expected. Azure revenue grew 48%. But read closely, and even the honor student shows the same trend. First, of that $45.2 billion, only $39.4 billion was paid in cash; the rest was booked as finance leases, obligations whose bill arrives in later quarters. Second, free cash flow was 27% lower than a year ago. Third, and most important, about 75% of the infrastructure spending went to processors, graphics chips, memory and storage with a relatively short service life. And the day after its report, the company began preparing a bond sale of at least $25 billion. For Microsoft, it is the first trip to the dollar bond market since 2017.
Add it up. According to Reuters, Google has raised more than $140 billion in debt and equity in 2026, Meta $62.5 billion, Amazon $98 billion through bonds. Together that is over $300 billion in less than a year, and Microsoft is now in line for another $25 billion. Just three companies, Alphabet, Amazon and Meta, plan at least $560 billion in capital spending this year. That is four fifths of the 2008 bailout that once frightened the entire world, except that this is not a one-time rescue. It is an annual norm, and one that promises to grow.
Honesty is required here, because my own book rests on a method: the mundane explanation first, everything else after. Debt by itself is not a sign of autumn, and capital investment by itself is not a loss. The railroads of the 19th century were also built on credit, and also through bankruptcies and stock market panics. But after those panics, the rails, embankments and bridges remained, serving the grandchildren of the people who built them. So the question is not whether the tech giants are borrowing. It is what they are borrowing for, and who will pay if the bet fails. And this is where the parallels with what I described in the book stop being coincidences.
The first parallel concerns the nature of the spending. In the chapter on the financial autumn, I separated three kinds of outlays: the creation of new productive capital, the maintenance of existing capital, and expenditure formally booked as capital investment that brings no commensurate gain in productive capacity. In the same place I wrote about the repair after which the asset needs repair again within a few years. A data center three quarters full of short-lived chips works in a similar way: the need for the next big outlay is built into the purchase itself. The building will stand for decades; the hardware inside will have to be replaced long before the building ages. The economy is not building a railroad that will serve for a century. It is building an asset that will demand fresh money constantly. I do not claim that AI creates nothing: cloud revenue is growing by tens of percent. I claim something else: spending is growing faster than revenue, and the gap is being closed with the future.
The second parallel concerns why no one hits the brakes. In the book's introduction, I wrote that in this race no one slows down, not because everyone around is a villain, but because whoever refuses to row will not stop the shared boat: someone else instantly takes his seat, and he alone loses. That is exactly what we are seeing, in its purest form. None of the four companies can cut spending, because the one that does loses the race to the other three. So all four promise, in unison, that 2027 will be even more expensive. The market punishes spending, but it will punish falling behind even harder, and management chooses the punishment that comes later and lands on someone else.
The third parallel concerns who carries the risk. Management makes the decisions, and its pay is tied to growth. The money that pays for those decisions comes from buyers of bonds and shares. Among them are pension funds, insurers and index funds: the savings of millions of people who never voted for a single data center. If the chips lose value faster than revenue grows, the loss will land on them. This is not yet the Economics of Losses in the full sense, because the common fund, as I define it, is not the savings of individuals but what is taken from everyone in the name of the common good. But only one step remains.
I examined that step in detail using 2008. To claim that what happened was the Economics of Losses, and not simply a rescue from catastrophe, three links have to be shown. Who earned the income before the crisis, and who carried the risk. How the risk was moved onto the common fund. And why those who earned the income kept their gains, while the causes of a repeat were never removed. The AI race has already passed the first link: the income is private, the risk is scattered among people who decided nothing. The second link is still ahead. But the four companies in question weigh so heavily in American stock indexes and American retirement savings that one day the question "too big to fail" will be asked not about banks, but about them. And I will say it plainly: if this race ends in a crash, the state will bail out not the bankers but the builders of artificial intelligence, and it will do so under the banner of technological sovereignty and the race against China. The language of care will give way to the language of security, and the bill, as in 2008, will go to the taxpayer.
That is the main thing this week revealed. For the hegemon, artificial intelligence is its biggest bet on a new material expansion, on the very spring that was supposed to lead it out of autumn. But that spring is being financed with the methods of autumn: debt, future income and other people's savings. In that case, it risks not opening a new cycle but becoming the most expensive perpetually ailing patient in history: infrastructure that has to be bought again and refinanced endlessly, because its health does not pay enough, and its death would mean admitting the loss. This is precisely how the Economics of Losses prolongs the autumn. Winter never comes, because every crisis is flooded with new money that will have to be repaid by those not yet born.
The most telling moment in this story is not Alphabet's crash but Oracle's rally. A company that forecasts positive free cash flow only in the second half of fiscal 2028 gained nearly 11% for nothing more than a promise to borrow less. The market paid not for what it had built, but for its promise to burn less. When a promise to shrink the loss is worth more than what has been created, this is no longer a bubble of optimism but its opposite: the market has started counting what the reckoning will cost. Testing my claim is now simple, and it will be tested not by my arguments but by the 2027 reports. If AI revenue catches up with spending, this was a material expansion, and I was wrong about tech. If not, the bill goes first to shareholders and creditors, and then, following a well-known script, to the state.