United States 'Data Brokers' have sold $500B+ worth of business data to china.
A recent ChinaTalk investigation identified at least nine US firms (@mercor , @HelloSurgeAI ,@AfterQuery, and others) selling roughly half a billion dollars worth of reinforcement learning training and data to Chinese AI labs.
Non-public data is becoming a global trade, with US intermediaries as the middlemen.
The AI providers that have access to the most EXCLUSIVE data will gain a competitive edge- you will see more and more announcements from providers on dollars committed towards licensing data.
National Security wise?
Well, security wise we are technically safe. Data is de-identified so no one is being spied on, no data details are being lost, no direct risk on the security front.
Compliance wise?
Not sure, I'm not a international legal expert.
America first wise?
Not great- if China based AI providers have access to the same data our US frontier labs have access to, where can the US win?
Let China find their own data in country- let America win on AI and we will all get richer because of it.
Thanks.
1
189
The AI Bubble- Who will Excel, Who will be Left Behind:
the employees that use AI correctly will excel, and the employees that don't will be left behind.
the companies that use AI correctly will excel, and the companies that don't will be left behind.
the providers that have the most exclusive data sets will excel, and the providers that only train on public data will be left behind.
Am I missing anything?
🤖 Made with AI
1
144
Breakdown of the Stanford's AI100 2026 Report:
"AI Goes Mainstream: The Scaling Era and the Rise of Generative AI"
Released November 10th, 2026
My top 3 takeaways (and where I disagree):
1. The panel's central claim is that the "loads of public data + loads of compute" approach that powered the last five years of AI is hitting a wall. High-volume, high-quality data is becoming scarce. They're right.
2. The infrastructure build out is massive: current AI data center investment is roughly triple the late 1990s internet boom (inflation-adjusted). McKinsey estimates another ~$6.7 trillion in data-center capital will be needed to keep pace.
3. The real competitive advantage AI proveers are chasing is exclusive access to high-quality datasets. GPT-3 was trained on roughly 45 terabytes of web text. The next generation of models will require large volumes of licensed, proprietary data that others can't freely scrape.
Where I disagree with the panel:
Explicit panel statement: "AI is not yet paying its way."
While a clean, company-by-company ROI is hard to isolate, one data point stands out to me:
From 2022 to today, Microsoft has nearly doubled its revenue while adding only about 2,000 employees. Over the same period, its internal use of AI has increased roughly 10x.
Coincidence? Maybe. But it's one of the clearer early signals that AI already showing up in operating leverage.
2
3
181
100,000,000 emails
500,000,000 team chats
$10M winning bid - @Google
$7.5M alternate - @mercor
The spirit Airlines data sale hearing is next week (Nov 17th).
That's ~1.7 cents per message, from an airline that went bankrupt.
Your company's slack and inbox aren't clutter. They're an asset AI labs are bidding on.
2
108






