- Almost all the money in AI goes into hardware: chips, memory, data centres and power. Very little of it has come back as revenue yet.
- A lot of that hardware is paid for with debt. Interest rates are high, so if AI revenue grows slowly, the lending will slow down first.
- About 10 AI stocks are now 42% of the US market. Only one bubble in history went further: railroads, around 1900.
- Learn the warning lights (bond yields, junk-bond spreads, funds blocking withdrawals). Just don't use them to time the market.
- The investor lessons that last aren't about any stock: be disciplined, pay a low price, and avoid what can hurt you.
"Venture Capital" (@venture) is a Telegram channel with about 2.3 million subscribers. I read roughly 80 of its posts from June to October 2026 and kept only what will still matter next year.
One honest note first. Despite the name, the channel is barely about venture capital. It's mostly US stock-market news, investor quotes and trade stories, and it makes money from ads. So I've treated its numbers as "as reported". Every figure below links to the post it came from, so you can check it. The lessons and the plain-English explanations are mine.
I'm not a financial advisor. This is a learning summary, not a recommendation to buy or sell anything.
1. Most AI money goes into hardware, not software
When people say "AI spending" they picture apps and chatbots. The money actually goes into physical things. Goldman Sachs, as the channel reported, expects about $7.6 trillion of AI investment over six years, rising from $765 billion in 2026 to $1.6 trillion a year by 2031 (post).
Inside "compute", the channel kept returning to three bottlenecks:
- The chips. Goldman's estimate assumes Nvidia takes about 75% of compute spending.
- The wiring between chips. An AI system is thousands of chips that must constantly talk to each other. Companies that make the networking parts, like Broadcom (AI chip revenue up 143% in a year), sell into every one of those systems (post).
- Memory. An AI server needs 8–10 times the memory of a normal server. Three companies (Samsung, SK Hynix and Micron) make over 90% of memory chips (post).
Because AI data centres are buying up memory, Apple raised prices on 8 products. The MacBook Air went up $200. Tim Cook called the memory shortage a "hundred-year flood" (post). The AI boom now affects people who've never used an AI tool.
Another sign of how tight supply is: older Nvidia A100 chips still rent at high prices. New chips aren't pushing out the old ones yet (post).
2. The money going in is far bigger than the money coming out
This is the most important chart in the guide. Big cloud companies are expected to spend about $650 billion on AI in 2026. Compare that with what AI earns them. Microsoft's AI revenue is running at about $37 billion a year, and Amazon Web Services' at about $15 billion (post).
A gap like that is normal early in a big technology wave. The railways and the internet both looked like this at the start. What makes this one risky is how it's paid for:
- JPMorgan expects $5.5 trillion of AI spending to 2030, with about $4.1 trillion of it borrowed (post).
- Morgan Stanley estimates that private credit will fund over half of the $1.5 trillion needed for data centres by 2028. Private credit means loans made by investment funds instead of banks.
- US government 10-year borrowing costs are above 5%, the highest in over 20 years (post). Cheap money is gone.
The channel put it well: the money is not gone, but cheap debt is. If AI revenue doesn't grow fast enough to cover the interest, lenders will fund fewer projects. That's the point where the buildout would slow.
Real use is still early, too. About 30% of the biggest US companies say AI has a measurable effect on them, but only about 2% track a specific number. And only about 2% of US households paid for an AI service this spring (post).
3. Ten stocks are now almost half the US market
The "AI Big 10" (Nvidia, Apple, Microsoft, Alphabet, Amazon, Meta, Tesla, Broadcom, AMD and Micron) are now a record 42% of the whole US stock market's value (post). Here's how that compares with past peaks:
What this means in practice: if you own an S&P 500 index fund, about 41 cents of every dollar sits in its top 10 stocks. The other 402 companies share about 23 cents (post). An "index fund" used to mean owning a bit of everything. Today it's mostly a bet on AI.
Since 2024, 42 AI-linked stocks made about two-thirds of the S&P 500's gain. Only 29% of stocks beat the index over three years, close to the dot-com low of about 25% (post).
Concentration tells you how much rides on one story. It doesn't tell you when, or whether, that story breaks.
4. Six warning lights worth knowing (they don't tell time)
Most of the channel's daily posts were about these. Each one gets a one-line explanation, because the terms sound harder than they are.
| Warning light | What it means | Reported level |
|---|---|---|
| Long-term bond yields | What the US government pays to borrow for 10–30 years. When it rises, all borrowing gets dearer. | 10-year 5.23%, 30-year 5.6%, 20+ year highs (post) |
| Junk-bond spreads | The extra interest weak companies pay above the government. Rising = lenders getting nervous. | 12.0% for CCC bonds, widest since 2022 (post) |
| Withdrawal limits | A fund lets investors take out only part of their money. People want out and can't. | Morgan Stanley paying ~43% of requests; Apollo 5% a quarter (post) |
| Long-term unemployment | People out of work 6+ months. Shows pain the headline rate hides. | 27% of the unemployed (post) |
| Interest-rate direction | The central bank raising rates makes money tighter for everyone. | Fed raised to 3.75–4.00% (post) |
| Consumer mood | Surveys of how well-off people feel. Spending usually follows. | 48.1, second-lowest in 74 years (post) |
Here's the odd part. In the same week those numbers were posted, the S&P 500 hit its 28th record high of 2026 (post). Markets and the economy can point in opposite directions for a long time.
Burry's "grieving cycle" of a market top
Michael Burry, the investor made famous by The Big Short, describes every market top as going through the stages of grief (post):
The same channel also posted that the S&P 500 has risen more than 90% since Burry told investors to "Sell" (post). He may be right about the cycle and still badly wrong on timing. Use warning lights to decide how much risk to carry, not when to jump in or out.
5. Patterns that keep repeating
Big IPOs: excitement first, fundamentals later
SpaceX's shares jumped from the IPO price to about $225, briefly valuing it near $3 trillion, more than Microsoft. Within days they fell to $150 (post). Part of the reason: only about 5% of the shares trade. When a small slice of a company sets the price for all of it, the price swings hard.
| Revenue | Profit | |
|---|---|---|
| Microsoft | $318B | $125B |
| Amazon | $743B | $91B |
| SpaceX (at a ~$3T peak) | $19B | −$9B |
Short squeezes: positioning, not news
Wendy's stock rose 42% in one day, from a 13-year low (post). The news (a new finance chief) was small. The cause was that a record 26% of its shares had been bet against. When the price rose, those short sellers had to buy shares back, which pushed it higher still. Short selling means betting a price will fall by selling shares you've borrowed.
Talent spreads, and starts companies
The SpaceX IPO made an estimated 4,400 employees millionaires, and at least 147 companies have already been founded by ex-SpaceX staff (post). It's the PayPal story again: that group went on to build LinkedIn and Palantir and start Founders Fund. If you're a founder, people leaving a company that just went public are worth knowing, as hires, co-founders and early investors.
Losses can be total
$10,000 put into Beyond Meat seven years ago is worth about $20 today, a 99.8% fall (post). Nike is down over 80% from its peak (post). A famous brand doesn't make a stock safe.
6. If you build with AI, here's what it means
This is my own read for founders and freelancers building AI products and services, not the channel's.
- Your raw material keeps getting cheaper. OpenAI cut one model's price by 80%. Leading US model prices fell about 25% in a few weeks, pushed by cheaper Chinese open models (post). Build things that get better when AI gets cheaper, not things that only work while it's expensive.
- The value is in making something easier, not in the model. Zuckerberg says his AI agent took his Instacart orders from twice a year to twice a week (post). Making buying easier made people buy more often. That applies to any business where an agent can take away the friction.
- Real human signal gets more valuable. Reddit now sells "real conversations between real people" to advertisers. Its ad revenue grew from $160M to $625M a quarter in three years (post). As AI-written content floods the internet, first-hand data and genuine reviews are worth more.
- Don't depend on hardware prices. Anything that needs servers, memory or GPUs is exposed to the shortage. Rent capacity when you can; avoid long commitments.
If you want a way to decide which of these to act on, the four planning questions in my business-planning post work well here.
7. The investor rules that will still hold in 20 years
The most reliable posts on the channel were the quotes, because they're old and easy to check. Here they are, with what each one means for a business owner:
| Who | The idea (in short) | For a founder |
|---|---|---|
| Warren Buffett | Success in investing comes from realism and discipline, not a high IQ (post) | A boring process you follow beats a clever plan you don't |
| Charlie Munger | Inversion: ask what would cause harm, then avoid it (post) | Before planning to win, list the ways you could fail |
| Charlie Munger | If you can't stay calm through a 50% fall a few times in your life, don't own stocks (post) | Only invest money you can leave alone through a crash |
| Peter Lynch | People spend hours saving $50 on a flight, then put $10,000 into a stock they heard about on a bus (post) | Put the most effort into the biggest decisions |
| Howard Marks | Buying below value is the most dependable way to profit (post) | The price you pay decides your return, whether for a stock, a business or a hire |
| Stanley Druckenmiller | Wealth comes from a few big bets with small downside, not many mediocre ones (post) | Focus. Three things done well beat ten done halfway |
| Jeff Bezos | Create more than you consume (post) | Applies to time and attention as much as money |
| Market saying | The riskiest time to invest is when the market looks safest (post) | Feeling safe is not evidence |
Seven investors the channel says to study
The channel listed seven names without explaining them (post). Here's the one idea each is best known for, in my words:
- Warren Buffett: buy good businesses at fair prices and hold them for decades.
- Charlie Munger: think with a toolkit of ideas from many fields, not just finance.
- Peter Lynch: invest in what you understand from daily life, then do the homework.
- Ray Dalio: write down your rules and learn from every mistake. His book is Principles.
- Nick Sleep: back companies that pass their savings from growing bigger on to customers, like Costco and Amazon.
- Terry Smith: "buy good companies, don't overpay, do nothing."
- François Rochon: owner-run companies held for the long term. His annual letters are worth reading.
8. How to read a finance channel without getting played
Channels like this one are useful for spotting what to look into. They're risky if you act on them directly. Before believing a post, ask:
- Who pays for this? This channel sells ads. Its business is attention, not accuracy.
- Is it hindsight maths? "$100,000 invested on that day would now be $402,000" (post) teaches nothing. Looking back, you can always find a winner.
- Are only the winners shown? A post about a fund making 47% in two weeks on options (post) never comes with the posts about trades that lost.
- Is there a source link? Most figures here had none. Find the original (Reuters, the FT, the company's own filing) before you repeat it.
- Is a target dressed up as a prediction? Even the channel admits "price targets aren't predictions" (post).
One thing to do today
Pick the three warning lights from section 4 that matter most to you, and note today's numbers. Check them once a month. You'll understand the news far better than by reading every daily post, and you won't feel pressure to act on any one of them.
If you'd rather learn to pull and chart numbers like these yourself, my Data Analyst Field Guide covers the tools.