Academic research shows that how much people talk about a coin predicts price better than what they say.
Sophisticated AI sentiment tools underperform simple volume metrics like Google Trends.
Volume beats vibes.
Hyperscaler AI capex is projected at $4.8 trillion by 2030. To justify that at a 10% hurdle rate, AI revenue needs to hit $1T annually. That's 1% of global GDP.
The bet is massive. So are the stakes if it doesn't pay off.
The contrarian's paradox: using narratives to invest isn't the problem. Unfounded narratives are.
The difference between conviction and delusion is whether you understood the value before the price moved.
When you read bullish analysis after a stock pumps, ask yourself: would this person have written this six months ago?
Usually the answer is no. They reverse-engineered the thesis from the price action. That's not insight. That's consensus in disguise.
If your altcoins can't clear the Bitcoin hurdle rate, you're essentially taking extra risk for free.
Every position needs to justify its place relative to simply holding BTC.
That's the brutal math of crypto portfolio management.
Bitcoin down 30%. Gold's worst day since the '80s
S&P 500?
Up nearly 2%
This wasn't a crash. It was a margin call cascade that broke every asset class except the one everyone expected
New Evergreen Wealth breaks down the mechanism
anthoniemeijer.substack.com/p/when-margi...
#Stocks #Crypto
Recent research shows AI sentiment analysis has weak correlation with stock prices
Simpler metrics like comment volume, Google Trends often predict better
The real edge isn't knowing when to buy. It's knowing when everyone else is too bullish
anthoniemeijer.substack.com/p/volume-bea...
We have all been too quick to make up our minds and too slow to change them. And if we don’t examine how we make these mistakes, we will keep making them. This stagnation can go on for years. Or a lifetime.
Opportunity cost is the silent portfolio killer
Crowded trades underperform. Unloved assets mean-revert. Staying out guarantees a 0% return.
Reduce opportunity cost. Size the positions.
Some examples:
Something significant is happening in AI. Growth rates are unprecedented.
Adoption is real.
But will value concentrate in 5-10 winners while most fail?
Forget the bubble talk. But don't forget to think for yourself.
Hyperscaler AI capex is projected to hit $4.8 trillion by 2030.
To justify that investment at a 10% hurdle rate, AI revenue needs to reach $1T annually.
That's 1% of global GDP (ex-China).
The bet is massive. So are the stakes.
a16z is heavily weighted toward AI applications ($11.8B).
But if OpenAI keeps copying app features, or Microsoft bundles Cursor-like tools into VS Code...
The application layer could get squeezed.
Where does value actually accrue?
a16z says AI isn't a bubble. Their data is compelling.
But they own 2/3 of the market they're analyzing.
This isn't just market analysis, it's also a fundraising document.
Doesn't make them wrong. But worth holding in mind.
The private AI market in three layers:
- Infrastructure: $15.2B (Databricks, Replit)
- Foundation Models: $11.5B (OpenAI, xAI)
- Applications: $13.9B (Cursor, Harvey)
Combined: ~$40B in annual revenue.
Being an optimist this year will make you more money than taking the fear-mongering rage bait.
Most likely scenario: the dollar doesn't debase, we don't go to war, and the stock market finishes higher over the next few years even with another dip.
Pessimism sells clicks. Optimism compounds wealth.
AI companies grew revenue 106% YoY in 2025.
Non-AI companies? 39%.
Top decile AI companies hit 693% growth.
Cursor, ElevenLabs, and Deel are reaching $100M ARR faster than Slack or Shopify ever did.
The gap is widening.
The growth is real, enterprise adoption is real, infra buildout is committed. Not debating that.
The open question is whether value stays broad or concentrates in a handful of winners while most fail. That's not bubble vs no bubble. That's a distribution question. Wrote about this today actually
a16z says AI isn't a bubble. Their data is striking: 106% median growth, $4.8T in committed capex.
But they also own 2/3 of the market they're analyzing.
New newsletter: the bull case, and why you should stress-test it
anthoniemeijer.substack.com/p/forget-the...
Pets and their food aren't getting disrupted by AI anytime soon. Two solid days in a row, even when the rest of the market bled red yesterday.
Revenue quality tiers:
- Tier 1: Organic fees (users pay because they want to)
- Tier 2: Subsidized fees (offset by token incentives)
- Tier 3: No value accrual (fees don't reach holders)
Most tokens are Tier 3. The winners are Tier 1.
Institutions aren't buying narratives.
- They're buying fee revenue.
- They're buying value accrual.
- They're buying cash flow equivalents.
2026 is the "Dawn of the Institutional Era."
Fundamentals finally matter.
The crypto valuation playbook, 2026 edition:
- Can you DCF it? (Does it generate revenue?)
- What's the fee quality? (Organic or subsidized?)
- Does value accrue to holders? (Buybacks, burns, distributions?)
Three questions. Most tokens fail all three.
Investment filter for AI infrastructure:
Don't ask "what chips do they use?"
Ask "what's their power situation?"
The grid is the new moat.
A single hyperscale AI site needs 30,000-40,000 tons of copper.
GE Vernova: +103% return in 2025. Backlog: $135B+
The chip trade was 2023.
The power trade is 2025.
The copper trade is AI exposure without model risk.
AI's real bottleneck isn't chips—it's power.
- Data center power demand: +165% by 2030
- Grid interconnection wait: 4-7 years
- Transformer lead times: 4+ years
You can buy all the GPUs you want. Good luck plugging them in.
🧵
New newsletter: Apophenia, the pattern-seeking trap that costs investors money.
Also: why Gen Z spending $2,600/yr on pets matters, a World Cup crypto play, and the forgotten Magnificent Seven stock.
substack.com/@anthoniemei...
Aim to balance being conservative and reactive.
All good investors have a Bayesian model in their head and update their forecasts as new information arrives but you also shouldn't update too aggressively.
Fundamentals didn't matter in early crypto.
They matter more now.
They'll matter a whole lot more in the future.
Which protocols pass your quality filter?
Protocols that actually pass:
- Aave: $157M revenue, $50M/yr buybacks, $25B loan book
- Hyperliquid: 97% of fees → automated buybacks ($645M YTD)
- GMX: 12.9% of supply retired, 30% fees to stakers (at ATL now)
Real cash flows. Not narratives.
The crypto quality filter:
✓ Network effects or switching costs
✓ Protocol revenue > token emissions
✓ Fee capture (buybacks, burns, yield)
✓ Aligned, active team
Sound familiar? It's just ROIC + moat + management, translated to on-chain.
