In partnership with
Hey {{first_name | there}},
A Reddit post with 7k upvotes told me the AI bubble is popping.
China wiped out half a trillion dollars "overnight." Anthropic is drowning in losses. Amazon is shutting down its AI division. Musk says companies are failing.
Scary, right? Half of it is wrong.
But the half that's right is a big deal. Especially the China part.
So I went through the actual reports. Here's what really happened, in plain words.
What happened in China
China has a stock index called the STAR 50.
An index is just a basket of stocks you track as one number. This one holds China's biggest AI and chip companies.
Since its July high, it has fallen 40%. That wiped out about $480 billion in less than four months (Reuters).
And it didn't stay inside tech.
The CSI 300, which tracks China's 300 biggest companies, is down about 20% from its June high. Across all Chinese stocks, $2.4 trillion has vanished since a peak in May.
Why it went wrong (5 reasons)
It was a pile-up. Here are the five things that stacked on top of each other.
1. The prices got silly.
At its peak, the STAR 50 traded at nearly 250 times earnings.
In simple words: you were paying 250 years of today's profits to own the company.
The Nasdaq 100 sits at about 30 times. Even after the crash, the STAR 50 is still around 100 times.
2. One theme ate the whole market.
Goldman Sachs says AI stocks make up 42% of the value of Chinese shares in the mainland and Hong Kong. And 57% of all trading.
So when AI fell, everything fell with it.
3. Small investors piled in on hype.
When Unitree went public in August, small investors asked for 8,000 times more shares than were on offer.
MetaX and Moore Threads jumped 700% and 400% on their first trading days.
One small investor said he bought Unitree because the government backed it. He's now sitting on losses. A Beijing fund manager summed it up as a "speculative crazy bull."
4. Chinese AI got too cheap (for its own good).
This one is ironic.
When DeepSeek released its V4 model, it also cut some API prices by 90%. Zhipu and MiniMax shares dropped right after.
In July, Moonshot's Kimi K3 sent chip stocks tumbling again.
Great for you as a user. Terrible for companies trying to charge for the same thing.
5. Washington stepped in.
The US is moving towards banning Zhongji Innolight and Eoptolink from supplying American data centres.
Traders also pointed to tax bills and the risk that other countries follow with their own bans.
Remember. A bubble popping doesn't mean the tech is useless. It means people paid far more for it than the profits can justify.
Musk's warning for "neoclouds"
On October 25, Musk posted that lenders have mostly stopped funding companies with no real income outside AI data centres.
He said SpaceX raised $40 billion in debt to buy GPUs, but only because it has Starlink and rocket revenue to back it.
His prediction for the 200+ neoclouds: "guppy-level death and consolidation." Some survive. Most don't.
Research firm Gartner agrees on the direction. It expects more than 80% of neoclouds to stop operating on their own by 2030.
Neocloud stocks fell the next day.
Amazon shrinks its AI lab
This is the real story behind the "Amazon is shutting down AI" claim.
In July, Amazon cut jobs in its AGI unit and closed an 80-person AI research office in San Francisco.
Most of its Nova models (Premier, Omni, Reel and Canvas) were put on "keep the lights on" mode. Still running, but getting the bare minimum.
The money and people now go to one big new model effort, led by robotics researcher Pieter Abbeel.
Amazon says building large AI models is still one of its most important projects.
My take: Amazon didn't quit AI. It quit trying to build every kind of model at once.
Big Tech's report card this week
Microsoft and Alphabet (Google's parent) both reported results on October 28.
Both clouds grew fast. The market treated them very differently.
✦ Microsoft: Revenue up 19% to $92.3 billion. Azure up 48%. And $18.6 billion in free cash flow. Shares popped after hours.
✦ Alphabet: Google Cloud more than doubled, up 103%. But it burned $9.1 billion in free cash flow, and its long-term debt jumped to $125.1 billion. Shares sank.
Free cash flow is the money left after paying the bills and building the data centres.
See the pattern? It's the same lesson as China.
Investors stopped paying for growth alone. They now want to see the cash.
What this means for you
If you use AI tools: Nothing breaks. The price war is good news for you. Chinese models keep getting better and cheaper.
If you're building an AI startup: Hype is getting expensive. Lenders now ask where your money comes from outside AI. Read our startup funding guide, and stress-test your numbers with the financial projection generator before an investor does.
If you invest: Two red flags from China. One theme taking over the whole market. And IPOs where demand is thousands of times the supply. When you see either, slow down. (I'm not a financial advisor. This is not investment advice.)
Bubble = prices running far ahead of profits.
Pop = prices crashing back down to reality.
AI = still here.
You don't need to watch the markets all day.
Block 10 minutes this week.
Next time a scary AI headline goes viral, find the original report before you share it.
Look at the AI tools you pay for. Check if a cheaper model now does the same job.
If you run a business on AI, write down how you'd survive if your AI costs doubled.
Send this to one person who thinks AI is either dead or magic. It's neither.
PS: The best place to share this is your team's group chat. That's where the "AI is over" posts are landing right now.
See you next week,
Aashish
Your company data could be worth $100k–$3 million
There's a new channel of revenue that has just been unlocked for companies with over 20 people that have been around longer than a year.
Few know it, but AI labs now let you monetize your company’s operational data, and they're spending $15 billion this year on it to train the next frontier of LLMs. Your workflows, documentation, processes, and internal convos like Slack are extremely valuable and they're paying anywhere from $100k to $3 million for it per company.
Before you ask, all the data is anonymized and cleaned before it's sent to a Lab. Bezel (Y combinator backed) helps find a data partner that is the best fit for you.
