Michael Reid
Forum Replies Created
-
Posts
-
China Doesn’t Need to Win the AI Race
It only needs to slow it down. 2008 showed us what happens next.Jay MartinAug 04, 2026
It Was Never About the HousesEveryone remembers 2008. Almost nobody remembers 2006.
2008 is the year with the name. Lehman Brothers collapsed. Markets fell apart. Governments rescued banks. But by the time any of that happened, the outcome had already been decided. The year that actually matters is 2006 – the year the whole structure quietly died while every headline said things were fine.
Stay with me, because the details of what happened in 2006 are a timeless lesson that we will inevitably learn again.
In the early 2000s, the most popular mortgage sold to riskier American borrowers was called a 2/28. It worked like this: for the first two years, you paid a low, easy rate. Then, for the remaining twenty-eight years, the rate jumped to something much higher – something most of these borrowers could never afford.
That sounds like a trap, but here’s the thing: nobody expected to pay the higher rate. Not the borrower. Not the bank. The plan – the openly understood plan – was that your house would be worth more in two years. You would take a new loan against the higher value, pay off the old loan, and start a fresh two-year window of cheap payments.
By the end of 2006, nearly four out of five of these loans written in 2003 had already been refinanced; the system worked great.
The loans were never built to be repaid. They were built to be replaced.
Now, here is the part almost everyone remembers backwards. House prices did not crash in 2006. They were at record highs. What changed was the speed at which prices were climbing. Price gains that had been running in the mid-teens started shrinking – still positive, still climbing, just climbing slower.
And in that same year, 2006, with prices near their all-time peak, borrowers started missing payments in growing numbers.
Why?
A borrower whose house went up eight percent instead of fifteen could not pull out enough new value from their home equity to replace the old loan. Remember, the new loan had to be big enough to pay off the entire old loan plus the fees – and banks would only lend against value the house had actually gained over and above the existing loan – so when the gains shrank, the new loan came up short. The loan replacement chain broke.
The crash in prices came a year later, and the panic came two years after that. The real estate crash was the echo, not the boom.
That part is very important to understand: The loans did not fail when prices fell. The loans failed when prices stopped rising fast enough.
Demand for housing was never the problem: people needed houses before 2008, during 2008, and after 2008. The problem was that a financial structure had been built on top of housing that only worked if prices rose faster every single year. The real economy – families needing a place to live – was healthy. The financial economy stacked on top of it needed something no real economy can deliver forever: acceleration.
The Bills Get Paid With the Next Round
Now let me show you where I see this same structure today.
OpenAI, the maker of ChatGPT, is the most valuable startup in history. Here is its price tag over the last two and a half years:
In early 2024, investors valued the company at $86 billion.
By October 2024, $157 billion.
By March 2025, $300 billion.
By October 2025, $500 billion.
And this past March 31, it closed a $122 billion funding round – the largest private raise ever recorded – at a valuation of $852 billion.
Now the other side of the ledger. OpenAI brought in roughly $20 billion in revenue in 2025. That is real money, and it tripled from the year before. But the company spends far more than that – on computing power, on staff, on research – and loses tens of billions of dollars a year. It has never earned a profit.
So ask the obvious question: how does a company that loses tens of billions a year pay its bills?
It raises new money. OpenAI signed its giant computing contracts months and years ago – and as those bills come due, each new funding round is what pays them. Meanwhile, the higher valuation convinces the next group of investors to fund the round after that. Investors keep writing bigger checks for one reason: the price keeps going up. Fast.
Each round from 2024 through 2025 valued the company at roughly 1.7 to 1.9 times the round before it.
For OpenAI, the rising valuation is not a scoreboard. It is the income. The company pays yesterday’s bills with today’s higher price – by raising cash against the gain in asset value, exactly the way a 2/28 borrower paid off the old mortgage with the new appraisal.
And the bills are enormous, because of how AI computing is bought. OpenAI has signed contracts promising to pay for computing power years into the future – hundreds of billions of dollars’ worth – whether it ends up using that power or not. These are called take-or-pay contracts: you take the product, or you pay anyway.
The companies supplying the computing power – Microsoft, Oracle, Google, Amazon – are set to receive those hundreds of billions, and they record them as ‘backlog ‘: guaranteed future revenue, signed and locked in.
Oracle’s backlog now stands at $638 billion, up 363% in a single year. Microsoft’s stands at $625 billion. Across the four big platforms, the total contracted backlog is roughly $2.1 trillion.
Here is a detail that is very important. Analysts who have traced those contracts estimate that about half of that $2.1 trillion is owed by just two companies – OpenAI and Anthropic – neither of which earns a profit.
More than half of Oracle’s entire backlog traces back to OpenAI alone.
And it goes one step further, just like it did in 2006. The tech giants like Microsoft and Oracle are not simply waiting to collect on these promises. They are borrowing against this guaranteed future revenue – raising debt to pour concrete and fill buildings with chips, with the signed contracts serving as proof to lenders that the money is coming…
Their construction spending has gone from $150 billion in 2023, to $226 billion in 2024, to $410 billion in 2025, to roughly $725 billion planned for this year. And this year, for the first time, that construction bill is bigger than all the cash these companies collect from their entire businesses combined. Every additional dollar of building is now funded by borrowing.
Follow the chain slowly, because this is the whole picture:
OpenAI promises to make future payments it can only fulfill by raising new money.
It can only raise that new money if its valuation keeps climbing.
The tech giants count those future payments as guaranteed revenue.
Then they borrow real money against that guarantee.
Wall Street calls this backlog “locked-in future demand.” Traced to its source, it is a $2 trillion loan to borrowers with zero income.
Thirty Years of One-Time Events
So what could slow the climb?
For that, we need to talk about some recent news out of China – and about a pattern that is now thirty years old.
In the 1990s, China took over furniture, textiles, and toys. Analysts called it cheap labour, nothing more. In the 2000s, it took steel and shipbuilding. A one-off, the same analysts said. Then solar panels – today China makes roughly eight out of every ten in the world. Then batteries. Then electric vehicles. Tesla, which once dominated the Chinese EV market, now holds only a single-digit share, while BYD, its former student, sells more electric cars than any company on earth.
Every single time, the American reaction followed the same script: dismiss it as an isolated event, right up until the industry was gone. Nobody connected the dominoes.
On July 16, 2026, a Chinese AI lab called Moonshot released a model named Kimi K3. Within a day it took the number one spot on a widely watched coding leaderboard, beating the best models from Anthropic and OpenAI in blind tests – at roughly forty percent lower cost. Eleven days later, Moonshot gave the model away: anyone, anywhere, can now download it and run it on their own computers, free.
The White House AI czar, David Sacks, called it what it is: “This is concerning.”
The usage numbers say it is more than concerning. On OpenRouter – a marketplace where businesses shop for AI models the way you shop for flights – American models handled about 70% of the traffic a year ago. Today they handle about 30%. The single most-used AI provider on the platform is now Chinese.
Now connect this to the structure we just walked through.
The Chinese models do not need to be better than American ones. They need to be nearly as good and nearly free – and they are. That pulls some customers away entirely, and it forces down the prices American labs can charge the customers who stay. Both forces push on the same number: the speed of American AI revenue growth.
And the speed is the collateral. Remember the funding ladder: every OpenAI round from 2024 to 2025 came in at 1.7 to 1.9 times the round before. The next step the structure is counting on – a public share offering at more than $1 trillion – would be a step of barely 1.2 times. The smallest jump ever, at the exact moment free Chinese models are attacking the growth that justifies it. The offering was expected this year.
It is reportedly slipping.
China does not need to beat American AI. It only needs to slow it down – because a structure financed on acceleration does not break on decline. It breaks on “slower.”
House prices in 2006 didn’t have to crash to kill the machine. They only had to rise eight percent instead of fifteen.
The Catfish Comes Home
Last October I wrote about the catfish effect — the strategy China has run for twenty-five years. Beijing invited the world’s strongest companies into its market on purpose: Google, Facebook, Uber, Tesla. The foreign competition forced Chinese firms to get better, faster. And when the students had learned enough, the rules tightened, and the teachers went home. Baidu, WeChat, Didi, and BYD came out of those waters stronger than anything that swam in.
Now look at what Washington is debating this month: banning Chinese AI models from the American market.
Sit with the symmetry for a second. China used competition to grow strong. America is preparing to use protection to grow weak.
Because a ban does not fix anything. Ban the Chinese models, and millions of American businesses lose access to a nearly-free tool their competitors in the rest of the world keep using — while the protected American labs, guaranteed their home market, lose the pressure that forces improvement. Allow the models in, and the growth curve keeps bending, and the trillion-dollar structure built on that curve keeps straining. There is no third door. When every available move makes things worse, it means the real mistakes were made years earlier — one dismissed domino at a time.
What Makes This Domino Different
There is a third borrower in this story, and it is the biggest one of all.
The United States government also spends more than it earns. Over the last twelve months, the gap was $1.6 trillion. It covers that gap the only way a borrower without profits can: by raising new money (treasuries) from lenders around the world. And here is the part most people never think about. America does not pay off its old debt, either. When an old treasury comes due, the government sells a new treasury to pay back the old one. Roughly $12 trillion of existing debt must be replaced this way before the end of next year.
The debt is never repaid. It is replaced. Where have you heard that before?
So ask about America the exact question we asked about OpenAI: how does a borrower that spends more than it earns, and never repays its old loans, stay operational? The same way. Only as long as people keep lending. And why do people keep lending? Not because they expect the money back – America has not run a meaningful surplus in a quarter century. They lend because they believe there will always be another lender behind them: that American growth will keep the whole structure credible, forever.
The 2/28 borrower ran on rising house prices. OpenAI runs on a rising valuation. The US Treasury runs on the world’s unshaken belief in American growth. Three borrowers, one requirement: the next check must always be bigger than the last.
And these structures are not sitting side by side. They are stacked, each standing on the one below. The belief in American AI holds up the S&P 500. The S&P 500 holds the world’s savings in American markets. And the world’s savings fund the Treasury’s next auction – at a moment when the national debt is $39.8 trillion and the interest alone now costs over $1 trillion a year, more than the entire military.
In 2008, when everything broke, frightened money poured into US government bonds, because the fear was pointed at the banks. Nobody doubted the US government. So when investors yanked their money out of everything else, they needed a safe place to put it – and they lined up to buy government bonds. A huge crowd of eager lenders meant Washington could offer tiny interest rates and still borrow trillions. The panic itself handed the government cheap money for the rescue.
Now run the next crisis. This time the fear would not be pointed at the banks. A crash in the AI trade is a crash in the belief that America owns the future of technology – the very belief that keeps the world lending to Washington in the first place. The scared money doesn’t line up to buy American bonds this time. Some of it walks away to other markets. And a government that spends $1.6 trillion more than it earns cannot stop borrowing while lenders hesitate. It has to keep selling treasuries to a thinner crowd, which means offering higher and higher interest to get the same money.
It’s Not the Economy, Stupid
People needed houses before 2008, during 2008, and after 2008. People will use AI before, during, and after whatever comes next. The technology is real. The demand is real. That was never the question – and anyone arguing about whether AI is “real” is answering a question nobody needed to ask.
In 2006, the only number that mattered was not “are house prices high?” It was “are house prices still rising faster than last year?” The moment the honest answer became no, everything that followed was just arithmetic working itself out – quietly for a year, then loudly for two.
So don’t ask whether American AI is impressive. It is. Ask the 2006 question: what happens to a $2 trillion promise when the growth slows down?
We already know the answer. We just don’t like remembering it.
So What Do I Do, Jay?
Here is how I think about a setup like this.
Start with what you cannot do: you cannot time it. The gap between “the growth slowed” and “the structure broke” lasted almost two years last time. The people who saw the mortgage problem in 2006 looked wrong – publicly, painfully wrong – for month after month while prices kept printing record highs.
Anyone who tells you the date this breaks is guessing. The mechanism is knowable. The calendar is not.
What you can do is watch the right number. This whole essay comes down to one lesson: the headline numbers will look wonderful right up to the end. House prices were at record highs while the loans underneath them were dying. So don’t watch the records. Watch the speed. Does OpenAI’s next raise price above the last one, and by how much? Do the backlogs keep growing, or just stay large? And watch for one moment in particular: the first time a tech giant announces it is cutting its construction spending – and its stock goes up on the news. The day the market rewards a company for leaving the race is the day the race is over.
Next, know what you actually own. Roughly forty percent of the S&P 500 is ten companies. If your retirement sits in an index fund, you are not spread across five hundred businesses – nearly half of your savings is a bet on one single belief, the same belief this entire essay has been about. That’s not a reason to panic. It is a reason to know it. Most people don’t.
Then ask the question this essay has been circling the whole way through. Everything in it – the mortgage, the funding round, the backlog, the bond – is the same object: a promise that only holds if a bigger promise arrives behind it. So look at each thing you own and ask: does this depend on somebody else’s promise staying believed? Some things do. Some things don’t. There’s a reason that in every era where paper promises came under question – the 1970s, 2008, today – the world’s savings drifted toward things that are nobody’s IOU. I’ll let you draw your own conclusion there.
And finally – manage your mind, because this is where most people actually fail. Not in the analysis. In the waiting. If the structure holds for another year of record highs, the crowd will tell you that you were wrong, that this time is different, that the skeptics missed the greatest boom in history. That pressure breaks more investors than any crash does. The people who came through 2008 intact were not the ones who predicted Lehman’s date. They were the ones who understood the machine, positioned themselves so its breaking wouldn’t break them, and then had the discipline to look wrong until they were right.
Be patient. Trust your process. And keep your eye on the only number that has ever mattered: not how high, but how fast.
Honest question – let me know in the comments: what am I missing?
That’s it for today,
Jay Martin
If you appreciate my writing, please share it with someone!
https://jaymartin.substack.com/p/china-doesnt-need-to-win-the-ai-race
It looks like this was removed due to defective sound track so posting again
And I was finding it very interesting
‘Flee KIEV, before it’s too LATE’: Russia issued a Tough Warning to Western Diplomats and Servicemen
D:
My assessment is you are wrong about Trump, Chris Hedges and Iran
Iran is being decent while kicking you all home and if you don’t soon get going home the damage may become apocalyptic.
For your study to improve your understanding
Hala Rharrit …
… of State Department says things we all know now about Minab.It was deliberate, as was an attack on children in Starobelsk–three waves of drones to finish off survivors and rescuers. That’s US “tactics”, that’s how US military “fights” stand-off “shock and awe” against civilians, especially children. That’ll show them damn Iranians, Russkies and others, as long, of course, as nobody shoots back–then it is a different matter.
Brazil, which is facing a presidential election on Oct. 4 in which the speculative vultures of Wall Street and the City of London are doing everything in their power to try to stop President Lula da Silva from being reelected.
Brian Berletic of the New Atlas discusses how Iran’s successful retaliation has led to desperation for Trump & the US empire, and how China and Russia are preparing for an unthinkable war that’s already underway yet no one’s talking about
D this 3 hour interview is worth much reflection with your judgement constrained throughout
A miracle may occur if you still yourself absorb and reflect on it
D writes:
All over we have ex-soldiers, veterans, officers, vowing we should have a war where nothing is blown up and no one is killed. Great idea! Q: HOW?
I write to go home and stay home and post all your military around your borders to defend and not to attack
