Though Experiment:
If all inference tokens in the world today ran on GB300 NVL72 Racks and DeepSeek v4.1 Flash, we would only need 200 Racks.
Let's say, Opus 5.5 needs 15 times this, then we need 3000 racks.
Nvidia is cranking out like 60,000 (equivalent) racks this year.
Based on my grok-assisted napkin math, Anthropic (just the SpaceX deal) will have enough capacity to crank out the equivalent of 25 quadrillion DS v4.1 flash tokens per day or 1.6 quadrillion Opus 5.5 tokens per day.
That's 55x the current rate, or ~4x if all Opus tokens.
Overconfidence is the killer.
Radical optionality over truth.
Your only edge is the speed at which you can change your mind.
My latest scenario: we are entering a period of permanent (decades) ultra-high uncertainty, in AI and all around. kevinkelly.substack.com/p/ou…
The Bitcoin Braaibroodjie Index
2018: 1 BTC = 11,500 Braaibroodjies
2020: 1 BTC = 15,200 Braaibroodjies
2022: 1 BTC = 24,100 Braaibroodjies
2024: 1 BTC = 54,000 Braaibroodjies
2026: 1 BTC = 48,830 Braaibroodjies
2030: 1 BTC = 10 ... show more
@JohannBiermann1
Recent South African Election Poll*:
Rassie will get 4.9 million votes
Siya will get 6.2 million votes
If we do a Rassie-Siya Kolisi and throw in Jesse Kriel as Minister of Defense we have enough votes to change the constitution.
*ChatGPT estimate
Let's do a thought experiment
Without this one skill, Buffett wouldn’t be worth $151B. He’d be worth $14B. Still rich, but no legend.
What set him apart was not only knowing how to invest...
Buffett engineered his ownership to make compounding sacred.
His wealth did not grow in a vacuum.
It grew on tax-aware engineering.
He did not just compound returns.
He compounded the compounding process.



