ok this Claude Code curve is f*cking insane
Anthropic shipped a coding agent that lives in the terminal as a research preview in February 2025
The run-rate revenue Anthropic disclosed since:
Sep 2025 → over $500M
Dec 2025 → $1B, six months after general availability
Feb 2026 → over $2.5B
That's 5× in about five months
And the S-1 shows who pays. Claude Enterprise users without Claude Code spent under $75 a month on average. The top 5% of users with it spent about 20 times that
When a software company grows this fast, most people credit the model
The real shift is the pricing:
Seat → usage → work done → revenue
Old way: charge per seat, same price whether the user works or not
New way: bill every task, so the heaviest users pay the most
The whole guide fits in one line: when the product does the work, revenue follows the work, not the headcount
Here's the part I think matters: run-rate is one month of revenue times twelve. It measures speed, not a year of sales, and Anthropic hasn't broken out Claude Code since February
Is per-seat software pricing dying, or is this just one breakout product?
your CLAUDE.md is a wish list
the agent reads it, weighs it, and skips it at step 40
hooks don't ask. they check
6 events, 3 exit codes and the Stop hook that makes "done" mean the tests passed https://t.co/SaBSktRbsC
This is f*cking insane
The NYT says Jensen Huang ran Nvidia with humiliation and public scoldings. Almost nobody has watched him explain the method himself, on camera:
"It's not physical torture. It's torture the same way that Taiwanese parents torture people
Nothing is ever good enough
You can't show me something without me giving you some criticism"
In 6 minutes, he walks through exactly where his management style comes from:
00:16 - His father: precise, perfect handwriting, does everything perfectly
00:40 - His mom: an obsession for details, can't let anything go
01:43 - His parents went to the United States with nothing
03:08 - The question: you'd rather torture people to greatness than fire them?
03:57 - His feedback comes always, immediately, never saved for later
04:34 - "Nvidia's retention is the best in the world," with people there for 33 years
When a CEO gets called abusive, most people blame his ego
Jensen describes a system instead:
Criticism → immediately → in the open → back to loving the person
Old way: save feedback for a private 1:1 and soften it
His way: say it right now, then move on
The board says it won't bring him back
Is this the culture that built Nvidia, or the reason it's looking for a new CEO?
This is f*cking insane.
Paweł Huryn measured what a $100 AI subscription is really worth:
"$100 of Claude Max buys about $4,640 of Opus at API prices"
He burned tokens on default settings, watched the weekly usage meter tick 1% at a time, then priced every token at API rates
→ Claude Max: 46.4x on Opus 5.5, 27.3x on Fable 5.2
→ Google AI Ultra: 25.8x
→ X Premium: 20.3x
→ ChatGPT Pro: 10.25x (October)
When the AI bill feels high, most people blame the price
The real lever is the multiplier:
Flat fee → heavy use → cheap tokens → score per dollar
The catch: a cheap token is not a cheap result. Even at 46x, an Opus 5.5 max run costs $1.26 and finds 41.7 of 105 bugs. GPT-6.2 Sol max finds 54.7 for $0.68
Old way: pay per token and ration every prompt
New way: pay a flat fee, then pick the model by what each dollar finishes
One tester, one week of meters, and limits can change any day
Are you on a subscription or still paying the API?
This is f*cking insane.
Paweł Huryn measured what a $100 AI subscription is really worth:
"$100 of Claude Max buys about $4,640 of Opus at API prices"
He burned tokens on default settings, watched the weekly usage meter tick 1% at a time, then priced every token at API rates
→ Claude Max: 46.4x on Opus 5.5, 27.3x on Fable 5.2
→ Google AI Ultra: 25.8x
→ X Premium: 20.3x
→ ChatGPT Pro: 10.25x (October)
When the AI bill feels high, most people blame the price
The real lever is the multiplier:
Flat fee → heavy use → cheap tokens → score per dollar
The catch: a cheap token is not a cheap result. Even at 46x, an Opus 5.5 max run costs $1.26 and finds 41.7 of 105 bugs. GPT-6.2 Sol max finds 54.7 for $0.68
Old way: pay per token and ration every prompt
New way: pay a flat fee, then pick the model by what each dollar finishes
One tester, one week of meters, and limits can change any day
Are you on a subscription or still paying the API?
ok this Claude Code curve is f*cking insane
Anthropic shipped a coding agent that lives in the terminal as a research preview in February 2025
The run-rate revenue Anthropic disclosed since:
Sep 2025 → over $500M
Dec 2025 → $1B, six months after general availability
Feb 2026 → over $2.5B
That's 5× in about five months
And the S-1 shows who pays. Claude Enterprise users without Claude Code spent under $75 a month on average. The top 5% of users with it spent about 20 times that
When a software company grows this fast, most people credit the model
The real shift is the pricing:
Seat → usage → work done → revenue
Old way: charge per seat, same price whether the user works or not
New way: bill every task, so the heaviest users pay the most
The whole guide fits in one line: when the product does the work, revenue follows the work, not the headcount
Here's the part I think matters: run-rate is one month of revenue times twelve. It measures speed, not a year of sales, and Anthropic hasn't broken out Claude Code since February
Is per-seat software pricing dying, or is this just one breakout product?
This is f*cking insane.
Everyone is talking about Trump asking Nvidia to give Jensen Huang a "second chance"
Almost nobody is talking about what Jensen told Lex Fridman about TSMC:
"Three decades. I don't know how many tens, hundreds of billions of dollars of business we've done through them.
And we don't have a contract"
In 3 minutes, he explains how Nvidia's supply chain runs on trust instead of paper, and why he turned down the chance to run TSMC in 2013
Today the WSJ called his handshake deals a danger for Nvidia
When a founder leaves, most people count his shares.
Jensen is describing the asset that doesn't transfer:
Promise → delivery → trust → billions in capex on his word
Old way: lawyers sign the contract, then the partner builds
New way: the founder explains it from first principles, and partners commit "several billion dollars of capital investments each"
The board can hire a new CEO. It can't hire 30 years of trust
This 3-minute clip explains Nvidia's next problem better than any analyst note
This is f*cking insane.
Everyone is talking about the SEC probing Nvidia and the board shutting down Jensen Huang's comeback
Almost nobody is talking about what Jensen told Lex Fridman about Nvidia's ceiling:
"Is it possible for NVIDIA to be a three trillion dollar revenues company in the near future?
The answer is of course yes"
In 5 minutes, he explains why computers stopped being warehouses and became factories, and why "$1,000 per million tokens is just around the corner"
When people value Nvidia, most of them count market share.
Jensen says there is no share to take:
Warehouse → factory → tokens → revenue
Old way: "it's theoretically impossible for a fabless semiconductor company to exceed a billion dollars"
New way: "that number is just a number"
The board can block his return. It can't block how he thinks about the size of the market
This 5-minute clip beats most $500 investing courses
This is f*cking insane.
Everyone is talking about the SEC probing Nvidia and the board shutting down Jensen Huang's comeback
Almost nobody is talking about what Jensen told Lex Fridman about Nvidia's ceiling:
"Is it possible for NVIDIA to be a three trillion dollar revenues company in the near future?
The answer is of course yes"
In 5 minutes, he explains why computers stopped being warehouses and became factories, and why "$1,000 per million tokens is just around the corner"
When people value Nvidia, most of them count market share.
Jensen says there is no share to take:
Warehouse → factory → tokens → revenue
Old way: "it's theoretically impossible for a fabless semiconductor company to exceed a billion dollars"
New way: "that number is just a number"
The board can block his return. It can't block how he thinks about the size of the market
This 5-minute clip beats most $500 investing courses
holy sh*t, this claude code + mcp + a sandbox = f*cking cheaper agent
one page turns anthropic's "code execution with mcp" into a graph
43 nodes, 76 links
one ls of ./servers reaches all 42 in 3 hops
how it saves your limits:
> every mcp server becomes a folder, every tool a .ts file
> the agent opens only what it needs: 150,000 tokens → 2,000
> a 10,000-row sheet gets filtered in code, the model sees 5 rows
> emails and phones turn into [EMAIL_1], [PHONE_1] before the model sees them
> working code goes to ./skills, the next run starts with it
your tools stop eating your context
your agent writes code instead of calling tools one by one
your limits last longer
the dev who sets this up once runs the same agent for less
the stack:
📁 code execution with mcp
↳ https://t.co/6JdAbLLT95
📁 code mode
↳ https://t.co/gCXHjiyR6r
📁 mcp servers
↳ https://t.co/zZpHfNSbNa
📁 agent skills
↳ https://t.co/wc9KVjuLiy
This is f*cking insane
Nvidia's board is hunting for a new CEO. This week Jensen signaled he'd be willing to come back
In January, 10 months before he resigned, he explained on a podcast exactly how a CEO gets removed:
"The CEO was always designed to report to the board of directors, and the board of directors report to shareholders"
"If the CEO doesn't do his job, according to 12, 13, 15 board members... the CEO is let go"
Asked whether there can really be a next CEO of Nvidia:
"Well, there will never be one like me"
In 4.5 minutes he lays out how he thinks about the chair he just left:
→ it's "an employment in an institution", "not like a church" and "not like a prison"
→ "I'm the best person for the job" and "I have to earn being the best person for that job every day"
→ "Nvidia has 61 CEOs": almost 60 direct reports who watch him reason through every decision
→ "An empty chair is better than a chair filled with the wrong person"
→ "Whether it's a missing CEO or missing VP of anything, the company will keep moving on"
The host notes he interviewed 22 CFOs before hiring Colette Kress. She is now Nvidia's interim CEO
When a founder gets pushed out, most people see the end of a story
Jensen described a job you re-earn every day: board → shareholders → earn it daily → keep the chair empty until the right person comes
Old way: the founder is the company
New way: the founder is an employee the board can let go
The board is keeping the chair empty for now. Should it wait for the right person, or hand it back to the man who wrote the rule?
This is f*cking insane
Nvidia's board says it won't reinstate Jensen Huang. Jensen says he'd be "honored" to come back
The whole fight fits in one formula:
CEO = Board + Shareholders
Only the board picks the CEO. Shareholders pick the board. So every road back runs through shareholders
The playbook has five routes:
→ persuade: rally investors now, so the board moves before a 2027 contest. One adviser calls it "a big if"
→ referendum: a director with more "against" than "for" votes must offer to resign
→ proxy access: 3% held for 3 years puts up to 2 nominees on Nvidia's own ballot. Window: Feb 24 to Mar 26
→ proxy fight: his own slate, his own money, and he has to solicit holders of 67% of the vote
→ special meeting: needs 15% held for a year. He has 3.58%, so he needs 11.42% from allies
The math almost nobody is doing: BlackRock 7.43% + Vanguard 7.31% + Jensen 3.58% = 18.32%. That clears the bar. Neither fund has said a word
Meanwhile 200,000+ people signed a petition, 3 directors pushed for his return, and Nvidia has shed more than $1 trillion in value since he left
When a founder gets pushed out, most people think it's over
Governance says it's a vote: stake → allies → special meeting → board
Old way: the board fires the founder and the story ends
New way: the founder starts counting shares
If you held NVDA, would you vote him back in?
This is f*cking insane
Anthropic closed its debut up 17%. The same day the Dow lost 1,101 points, and on Friday ANT gave back 11%
Wall Street can't agree on one question: is this the top
The whole split fits in one line: three desks, three answers
→ the drop: Dow −2.19% to 49,097, its first close under 50,000 since June. S&P −2%, Nasdaq −2.2%
→ the split: since the S&P's Oct 6 record, Nvidia is down 39% and Intel 50%. Anthropic still priced at $1.5T
→ the bear: Capital Economics says the falls "have further to run"
→ the middle: UBS cut global stocks to Neutral, "AI monetization will need to ramp quite rapidly"
→ the bull: Goldman sees the S&P 500 at 8,700 by end-2027, +17%. That call came two days before Thursday's drop
When the market drops, most people ask if AI is over
Look closer: AI capex produced about half of S&P 500 earnings growth this year, per Goldman. If you own an index fund, you own the data-center bet
Old way: AI stocks move as one
New way: chips sell off while IPO buyers still pay up
The S&P is on track for 7 straight down days, its longest streak since 2020. Top, or the beginning of the middle?