Founder, Social Innovation Circle | LC Philanthropy | Adjunct Professor of Finance, Hong Kong University | impact investor supporting social entrepreneurs
This is one of the more insightful analyses I’ve seen yet of the current AI bubble.
The author argues that we’re in a classic leveraged real estate boom. The turning point is when growth stops accelerating (the second derivative), not when growth stops.
We’ve already crossed that point and no one noticed.
The real comparison is with the subprime debacle of 2007-8, not the dot com bust of 2001.
open.substack.com/pub/ground…
The proportion of savings that Americans hold in shares has more than doubled in the past 20 years.
The average saver now holds nearly half their assets in stocks. This is way higher than even just before the dot com bubble of 2001 (~40%) or the mania which preceded the great financial crisis of 2008 (~32%).
Many people borrowed to buy into this market, which is held up by only a handful of tech stocks.
The pain is going to be ugly when the bubble bursts.
I’m not sure why this isn’t bigger news, but seems the Saudi backed Yemen forces are now in full retreat on the Red Sea coast, falling back to defend Aden, and have basically ceded control of the Bab-Al-Mandeb strait to the Houthis.
So there goes the Red Sea export route.
Meanwhile the IRGC attacked the US radar systems at their military base in Bahrain over the weekend with drones. No-one really saying how much damage done.
And Trump is saying he can’t be bothered to negotiate with Tehran any more. Some people think this is a negotiation tactic. I think he’s genuinely lost interest and can’t be bothered anymore.
It’s rare to see a more transparent hit job done on a major public figure than that which we saw on Jensen Huang this week.
@NuryVittachi here does a good job explaining what really happened and why:
THINK ABOUT IT.
- Bill Gates had lots of illicit girlfriends—no problem.
- Elon Musk has multiple children with multiple women—no problem.
- Dario Amodei dated a Chinese woman, Jade Wang, before marrying Cami Clark, who pitched a pornography business to none other than Jeffrey Epstein—no problem.
- But Jensen Huang commits old-fashioned infidelity—and this is such a big problem that he has to resign?
Doesn’t make sense.
Here’s the real story.
.
CHAPTER ONE: BIDEN’S CHIP WAR
Jensen Huang of Nvidia was puzzled. It was 2022, and his video-graphics processors had become globally popular as the heart of high-end computing across the planet.
But that autumn, Biden violated the World Trade Organization’s most fundamental principles. He ordered the world’s makers of advanced semiconductor chips and chipmaking equipment (in the Netherlands, Taiwan and Japan) to stop sales to China.
Biden’s administration implied that the chips were for China’s military gear, but this was a lie. The US’s own security establishment admitted that 99.9 per cent were used for civilian purposes, like medical technology and AI computing.
Huang was dismayed. He could no longer sell Nvidia’s A100 and H100 to China.
.
CHAPTER TWO: SOLVING A PROBLEM
Jensen Huang was a problem-solver. In February of 2024, Nvidia launched the H20, a made-for-China AI chip less powerful than its US-only top model.
That way he could keep the US government happy but also have something to sell to the Chinese.
Not everyone in the US was happy about this, and export rules fluctuated.
.
CHAPTER THREE: THE DEEPSEEK MOMENT
In January 2025, a nerdy young man from the Chinese finance industry announced that his firm Deepseek had quickly and cheaply created a high-efficiency open-weight AI.
The implication was huge: the Chinese were finding ways to develop top level AI functionality without massive clusters of top-tier Nvidia GPUs.
Jensen Huang knew he had to respond urgently to keep his place in the China market. He flew to meet trade officials in Beijing in the middle of April and vowed to make "every effort" to serve the Chinese.
The US anti-China hawks had been quietly mobilizing—and made an announcement. In April of 2025, the US banned all export of Jensen Huang’s H20s to China.
Unbelievable! The Nvidia CEO was furious. "Anybody who thought that one chess move to ban China from H20s would cut off their AI ability is deeply uninformed," Huang told an interviewer at a trade meeting in Taipei.
.
CHAPTER FOUR: HUANG’S SPECIAL SKILL
Then came some rare good news. The Nvidia CEO started lobbying Trump behind the scenes. Huang was a master negotiator – and his efforts proved successful.
In the middle of July, 2025, Huang was back in China—his third trip in 12 months.
Huang appeared without his signature black leather jacket. Instead, he wore a traditional Chinese tang suit, and made some of his speech in Mandarin.
Washington had cleared the way for Nvidia to export the H20 to China, he said. We’re back!
Trump had demanded a cash cut of H20 sales in return for approving the exports—US government policy appeared to be literally for sale.
.
CHAPTER FIVE: SPOOKS WERE NOT HAPPY
Huang was moving chips to China again, and the anti-China hawks in the US government were unhappy—and so were the spooks.
There’s a widespread tendency to visualize the US intelligence community as right-leaning hard men, Trumpites in black suits.
This was the opposite of the truth. The CIA leadership greatly preferred the simple, blanket negativity towards “the enemy” characteristic of the Democratic Party. With Russians, Chinese, and North Koreans portrayed as cartoon villains, black ops were justifiable.
But for all his bluster, Trump had a propensity towards treating the leaders of Russia, China and even North Korea, as actual humans who could be negotiated with.
This transactional approach was enormously problematic for the CIA, which needed cartoon villains to justify their annual budget of more than US$15 billion.
And Trump’s habit of parachuting unsuitable new leaders in to command the intelligence community—such as the widely disliked Mike Pompeo—crystalized the negativity, as CIA chroniclers have written.
.
CHAPTER SIX: A TWIST IN THE TALE
Jensen Huang’s celebrations were short-lived. In an unexpected twist, China in September 2025 said it did not want Nvidia chips and would no longer buy them.
- US$15 billion in sales projections vanished.
- Nvidia had to write off a reported $5.5 billion in inventory losses.
- The company had now completely lost access to China's AI market, estimated at US$50 billion.
- And hard evidence was emerging that Chinese companies were building alternatives even for CUDA (a unique element in Nvidia chips).
Huang wanted more positivity from the US. Nvidia was a US firm, and it went from having a 95% market share in China to 0%. “I can’t imagine any policymaker thinking that that’s a good idea,” he told reporters.
.
CHAPTER SEVEN: AGAIN, HE SOLVES THE PROBLEM
But Huang’s extraordinary skill in dealing with Trump paid off again, and by mid-2026, the Nvidia business found a complex equilibrium had emerged in its China business.
Huang had (in December 2025) persuaded the US President to allow Nvidia to sell even the high-powered H200 chip to China, as long as the US government got 25% of the take. So a range of Nvidia chips could again be sold in China.
- Other Nvidia chips were arriving in China via the black market.
- There were suspicions that companies in Singapore and Malaysia, such as Megaspeed, were flouting export restrictions, also delivering chips.
- Some Nvidia chips were stuck in Hong Kong, trapped by the confusing, ever-changing rules, but were expected to be released eventually.
- On the downside (from Huang’s point of view), new China-origin chips, notably Huawei’s Ascend 950, were catching up with advanced capabilities that mimicked those of Nvidia.
In short, China was back in the game.
.
CHAPTER EIGHT: TAKING HIM DOWN
Something had to be done. China had to be dealt with. But first, Huang had to be taken down.
The US anti-China hawks saw Huang’s success in making deals with Trump as a key part of the problem. It was an element which had to be surgically removed.
The CIA and their partners in the western mainstream media sprang into action.
In October 2026, the US intelligence community told New York Times reporters that there were suspicions that Huang’s mistress, Alice Huang Le, was a Chinese spy.
It provided no evidence.
The allegation was plainly ridiculous—the flow of legitimate and illegitimate chips across the border was no secret, and the idea that discussions about the detailed minutiae of nano-level semiconductor engineering could be exchanged during pillow talk was ridiculous, even comedic.
But when the US intelligence community and the mainstream media join forces, they’re virtually invincible.
At this moment in the process, the complicit media are playing its role. A typical headline was that in the Economist yesterday:
“Jensen Huang must resign.”
.
EPILOGUE
Jensen Huang resigned.
Trump has been weakened.
The world’s public has been fooled with the “honeypot” story.
China’s Huawei is pressing ahead—with new, powerful versions of its Ascend chips slated to come out in 2027 and 2028.
.
.
To be truly sovereign, a nation or regional bloc must be independent in four areas: defence, food, energy and tech.
By this definition, there’s only one truly sovereign power in the world: the United States.
China’s solid on defence and tech but still imports food and oil.
The EU is sovereign in none of these areas. It has allowed itself to be entirely captured by the US on every key sector that matters for independence.
That’s why the EU, for all its market size, wealth and population, still punches far below its weight on the world stage.
It’s abject helplessness in the face of the Iran war, which directly impacted its energy and food costs, illustrates that.
I was wondering why shares would fall on a Dem victory. Normally shares rise as the markets see a divided Congress as being a check on an overreaching President.
Spoiler- it’s about public support for data centres
BREAKING: 🇺🇸 Nasdaq futures crashes -2% as Democrat Gina Hinojosa takes an early lead in the Texas governor election.
Hinojosa wants to halt data center expansion.
Texas accounts for roughly one-third of all data centers currently under construction in the US.
I remember during the early dot com era, when Facebook was the hot new app , there were thousands of startups that were all claiming to be niche versions of Facebook - the ‘Facebook for entrepreneurs’, the ‘Facebook for coders’, the ‘Facebook for Moms’ etc. Every pitch book literally started with that tagline.
Turned out that the next Facebook for all of these niches was just… Facebook. (at least until WhatsApp & Instagram turned up).
I think the same thing will happen with AI.
Right now we’re seeing every start-up trying to build itself as an AI wrapper around a classic market role: the AI lawyer, the AI doctor and so on.
And I think the general LLM‘s are going to eat them all.
You don’t need a specialist AI lawyer. Gemini or Claude will do a great job for you.
Same for medicine. And for pretty much any area of niche knowledge.
And the generalists like McKinsey? They’re all toast.
That’s why my team hasn’t made any investments in pure AI software plays up to now.
Not because we’re not seeing amazing applications but simply because it’s hard to see how any of these survive the increasing abilities of the general LLMs.
I think to be a successful AI start up today you have to have some real world moat.
Either access to proprietary data that no one else can read or being the interface to something physical or real that can’t be duplicated digitally.
Pure software is too easily eaten by general AI.
The market is bracing for a renewed assault on Iran by US & Israeli forces that could come as soon as this weekend (talk about “after the midterms” could be a feint).
Trump already cut short planned final rallies in Texas and was being briefed on military options in-person yesterday, which usually signifies that we’re in final stages of planning.
If / when the attack happens, they’re talking about going for Iranian civilian infrastructure, including power stations and water treatment plants.
The two leaders appear desperate for a distraction from the electoral woes that they’re both facing, as well as to engineer a ‘rally round the flag’ effect to bolster their flailing popularity at home.
The Iranians will probably retaliate by escalating attacks on US bases, as well as gulf energy infrastructure.
This will spike oil to well above $130, and send interest rates hurtling upwards.
In other words, it will crash the world economy and destabilise the Middle East, with ramifications that are difficult to project and be with us for decades.
I really hope I’m wrong on this but all the signs are pointing towards the escalation of this insanity.
The US tech giants have shed over $900 billion in market cap in the past 24 hours.
This chart shows why.
The enormous spend on AI capex, which shows no sign of slowing, has fully exhausted all available free cash from the big 4: Microsoft, Amazon, Google and Meta.
With free cash flow gone, all capex from here on out will need to be funded by investors, not profits.
But investors need to be repaid, with interest. And the price of that debt is getting way more expensive.
Which is why the shares of these companies are now falling.
60% of the S&P 500 is now in bear market territory, down more than 20% from their highs.
It’s just a handful of stocks that are doing all the lifting, and even those are starting to fray
There it is.
Over 60% of the S&P 500's components are now in bear market territory, down -20% or more from their record highs.
Four of the Magnificent 7 stocks are now in bear market territory, with these seven names erasing a combined -$4 trillion in market cap since their highs.
We are in a "stealth" bear market.
And, we believe it's a huge opportunity as we outlined below on October 15th.
The tech sell off is now underway.
Intel, the company handpicked by the govt to be a national champion to rival TSMC, crashed 20% yesterday.
But that wasn’t even the big story of the day.
Alphabet plummeted 12%, wiping $500 bn off its market cap, its biggest drop ever (and the second largest in stock market history).
The figure that freaked out investors?
$9 billion.
That’s how much cash the tech giant will have to spend even after using up all of its lucrative ad search revenues (which are still growing 12%) on data center buildout.
And the losses are just getting started.
“These are rookie numbers”.
Alphabet are committed to spending a further $200 billion over the coming months. They won’t be cash flow positive again for years.
They’re not alone.
Total AI capex spend from all hyperscalers for 2027 is projected to be around $1 trillion. There isn’t enough free cash flow from customers to meet that demand.
Which means that most of this capex will need to be funded from debt, which is at decades high levels, or by issuing insane amounts of equity.
Get ready for the great dilution.