Archiving legendary Founder stories, insight & advice | Learn from the greatest minds and be inspired.
Joined October 2022
- Tweets13.3K
- Following85
- Followers369K
- Likes75.9K
Founder Archive retweeted
The biggest opportunity for companies right now lies in data.
But before signing a data deal, ask who holds the raw files, and for how long.
Companies are starting to license their Slack, tickets and CRM to train AI models. Every support ticket has a customer in it. Every Slack thread has your team in it.
It's possible to use 15 demographic details to correctly re-identify 99.98% of Americans in a dataset. Blacking out names also strips what makes the data useful: who handed what to whom, and how the work moved through the team.
Spirit Airlines' data sale is stuck on exactly this. The privacy ombudsman cleared it on customer data. The flight attendants' union objected that employee records weren't covered.
The questions you should ask before talking price:
1/ Where are the raw files held, and when are they deleted?
2/ Who does the de-identification, and how has the method been tested?
3/ How are indirect details in free text handled (health, location, family)?
4/ Does the method keep the workflow intact, or just black out names?
5/ Who keeps ownership of the data?
6/ What can it be used for, and by whom?
And check your customer contracts first. Slack's API terms have barred third-party apps from using Slack data to train LLMs since May 2025.
This is what the likes of @micro1_ai built for. Privacy is built in from the start through its own PII de-identification model, which replaces each real person with a consistent synthetic identity across every file, so the record of how work moved stays intact. Personal information is de-identified, data is handled in isolated pipelines, and the company keeps ownership of its data.
Check out what your data is worth below!
Founder Archive retweeted
Google just signed up to keep a 1970s nuclear plant running until 2050.
In Finland, the Loviisa plant supplies about a tenth of the country's electricity. Without new investment, its operating life was set to end around 2030.
On September 9, @Google signed a 22-year agreement to buy up to half of its output, which supports extending the plant's life through 2050.
It comes alongside €13 billion of Google data center and related investment in Finland across 2027 and 2028.
Twelve days later, Georgia Power announced a Google agreement tied to upgrades at two existing American nuclear plants, Vogtle and Hatch. New turbines, pumps, motors and cooling equipment, squeezing about 96 more megawatts out of reactors that are already built.
It still needs approval from Georgia's Public Service Commission.
Both deals buy more life, or more output, from reactors that already exist.
The fastest megawatt available to the AI industry right now is one that is already standing.
Maintaining what we built used to be the unglamorous part of infrastructure. It has just become one of the most valuable.
ALT Loviisa Nuclear Power Plant, Finland, seen across the water. Photo: Thomas Gartz / Wikimedia Commons, CC BY-SA 4.0.
Founder Archive retweeted
Sat down with Lyndsay Malchuk of the International Investment Forum in New York.
We covered the Lion resource, drilling at depth, the PEA with BBA, metallurgy, nickel, and why Quebec.
28 minutes. Worth a watch.
youtube.com/watch?v=hPbf15re…
$PNPN.V | $PNPNF | $IVV1
Founder Archive retweeted
Sam Altman doesn't want one superintelligence.
"I would much rather have a trillion AIs of slightly superhuman intelligence join society than one AI of 1 trillion IQ join society."
His reasoning: society holds up because lots of people keep each other in check. Teams of AI agents can work the same way. A single giant brain loses that.
From @collision's Cheeky Pint with @sama.
Founder Archive retweeted
OpenAI doesn't reward its own people for burning compute.
Sam Altman: "We try never to reward people for using a lot of compute. No token leaderboards. It's actually slightly bad to be high on the token leaderboard."
But they did let one math push run wild, on purpose.
"We did let it get out of control for one week on the math."
Collison: "And it worked!"
From @collision's Cheeky Pint with @sama.
Founder Archive retweeted
Kylie Minogue is 58.
Pop star since 1987. Last year she played 66 shows on a world tour.
She says she's never as fit as when she's on tour.
6 things she does to stay fit:
Founder Archive retweeted
In 1760, two Philadelphia underwriters refused to pay an insurance claim.
It took five years and an appeal to London to settle it.
The claimant was John Brown of Providence, with his late uncle Obadiah. Their ship had sailed, they said, to exchange prisoners of war with the French, under an official flag of truce. A British privateer captured it and found molasses and sugar bought from the French at Saint Domingue.
The underwriters, David and William McMurterie, had covered only a small part of the risk. £100, for a premium of £23. They argued the capture was "not a Loss within the Meaning of the Policy," because the voyage had been "illicit and contraband Trade."
The Supreme Court of Pennsylvania held they had to pay. The McMurteries appealed all the way to the Privy Council in London. On July 16, 1765, it upheld the decision.
Hannah Farber tells this story in her work on early American insurance, and her conclusion is pointed. Insurance gave merchants "a framework for sidestepping questions of legality altogether."
The underwriters wrote the policy. The court held them to the words they wrote.
ALT An East Prospect of the City of Philadelphia, after Scull and Heap, 1768. U.S. National Archives. Public domain.
Founder Archive retweeted
Success is the progressive realization of a worthy ideal of your choosing
Success is going on a 10am run and listening to the birds and the rhythm of your own breath
Success is long dinners with friends from different backgrounds
Success is living on your own terms, making peace with the constraints life gives you, and going after the rest with everything you've got
Success is seeing the world in all its beauty and darkness
Success is building something meaningful with people you care about
Success is waking up next to the person you look forward to talking to the most
Founder Archive retweeted
Sunith de Fonseka played basketball for Sri Lanka.
His role on court is to get the best out of everyone else.
In 12 years as a Royal Air Force officer, one rule was drilled into him.
Founder Archive retweeted
Our concentrate came back grading over 25% copper.
That figure gets far less attention than the recovery number, and it matters more to the people who will eventually buy the material.
Recovery tells you how much of the copper you managed to get out of the rock. Concentrate grade tells you what is actually in the bag you ship. A smelter is buying the bag.
Weak concentrate means you pay freight on material the smelter then charges you to take out again. It shows up as a deduction on every settlement for the life of the operation.
There is a second thing in those results that I think matters more.
SGS ran two locked-cycle tests. One on a blend representing run of mine feed, one on the low grade disseminated material by itself. Both produced a single copper concentrate, and the platinum, palladium, gold, silver and nickel all reported into it.
One concentrate. One product to sell, one set of terms to negotiate.
The reason sits in the mineralogy. The copper occurs as coarse grained chalcopyrite and cubanite, and most of the platinum group minerals sit inside those copper sulphides or attached to them. They travel together because they formed together.
Hydrometallurgical work on a direct to metal route is still running.
None of this is an economic study. We have not completed one. It is the part of the test work that decides how complicated this would be to sell.
ALT Table 4 from Power Metallic's 8 September 2026 resource release: SGS locked-cycle flotation results for the Lion zone. Both the blended composite and the low-grade composite produced a copper concentrate grading over 25% Cu, with 98.9% and 98.3% copper recovery.
Founder Archive retweeted
my life changed when i realized most of my days run on a few carefully chosen if-then statements:
• if i have negative thoughts, i go for a long walk
• if i feel uninspired, i read about someone great. it puts my problems back in proportion
• if i feel low energy, i go to the gym.
•if i feel disconnected, i message my wife (or a friend)
decide your if-thens on a good day, so a bad day doesn't have to
Founder Archive retweeted
At 48, Pedro Pascal tore a back muscle and dislocated his shoulder filming Gladiator II.
At 50, he was Mister Fantastic.
Now 51, he looks fitter than ever.
4 things that got him back:
Founder Archive retweeted
Thousands of AI chips can change their power draw in the same instant.
They do it hundreds to thousands of times a day. The electric grid was never designed for a customer like that.
When a large AI model is being trained, the chips work in lockstep. They draw hard, pause while the results are exchanged, then draw hard again. In the words of one July announcement, "many thousands of GPUs ramp up and down in synchronized bursts."
A neighborhood does not behave that way. Ten thousand air conditioners do not switch on in the same second and off again a moment later.
That July announcement came from Crusoe and ON.energy. They plan to install 5 gigawatts of specialized battery-backed power equipment across several hyperscale campuses, with commissioning starting this year and running into 2027. Its job is to absorb the swings on site so the grid sees a steadier customer.
In Texas, the system was validated against ERCOT's large-load interconnection requirements and its ride-through protocols, the rules for how a big load has to behave when the grid wobbles. Under Senate Bill 6, passed in 2025, large new loads in Texas must be equipped so ERCOT can curtail them in an emergency.
Every structure I have litigated was designed around an assumption about the load it would carry. When the load changes character, the design has to change with it.
AI changed the character of electrical load. The industry is now building the equipment to catch up.
ALT Crusoe and ON.energy press release, July 21, 2026: 'Crusoe and ON.energy to Deploy 5 GW of AI UPS at Multiple Hyperscale Campuses', with the line 'many thousands of GPUs ramp up and down in synchronized bursts' highlighted.
Founder Archive retweeted
aristotle had a word, zelos, for the pain you feel when someone you feel is similar to you has achieved something you could have but haven't yet.
he was careful to keep it separate from envy. envy wants the other person to lose it. zeal wants you to go and earn it.
i think about this a lot because X and IG show you everyone at once. every founder who raised, every friend living the dream, all in the same few scrolls. so you get the pain all day and almost none of the signal. the work is noticing which ones sting in a way that makes you want to move, and checking whether the race they point at is actually yours, and letting the rest scroll past
Founder Archive retweeted
Natasha's mum was a teacher. She told her that when she trained, learning about autism had been optional.
That one detail turned a mum of 4 into a lobbyist.
From 2016, every registered teacher in Victoria had to train in teaching students with disability.
Here is her story:
Founder Archive retweeted
Tom Welling played Clark Kent for 10 seasons of Smallville.
He says he "never had the time to work out."
6 things that kept him in Superman shape anyway:
Founder Archive retweeted
Caleb Simmonds keeps a log of every film he watches. He says it is addictive. You keep watching to build up your stats.
He has still never seen a single Twilight film.
He is also one of the most productive people on our customer experience team.
Here is his story:
Founder Archive retweeted
An investor asked me what a smelter actually pays for. Most people guess wrong.
You don't sell metal. You sell concentrate. And the smelter doesn't credit you for everything in it.
Every metal in that concentrate has a payability, the share of contained metal you actually get paid for. Copper usually carries a high one. Precious metals get paid at a lower rate, often only above a minimum threshold.
Then come the deductions. A treatment charge for processing. A refining charge per unit of metal. Freight to get it there.
What survives all that is your net smelter return.
It's why two projects with the same headline grade can have completely different economics.
Think of a single-metal deposit as one income stream. One payability, one price, and when that price turns you feel all of it. A polymetallic deposit is several streams sharing the same cost base, and they rarely fall together.
At Nisk we've got copper, nickel, palladium, platinum, gold and silver in one system. That's not a marketing list. Each one's a separate line on a smelter settlement.
That's the lens I'd read any deposit through. How much is there, and then how much of it actually gets paid for.
Keep the questions coming. I'd rather answer them here than have people guess.
$PNPN.V | $PNPNF | $IVV1
Founder Archive retweeted
In 1935, only about 2% of Texas farms had electricity.
Thirty years later, only 2% did not.
Nationally the figure was about 1 in 10. In Texas it was worse. Rural families lit their homes with kerosene lanterns and pumped water with windmills, because private utilities would not string lines across miles of land for a handful of customers.
On May 11, 1935, President Roosevelt created the Rural Electrification Administration by executive order. It lent money to cooperatives, owned by the customers themselves, to build the lines the utilities would not.
In the Texas Hill Country, a newly elected congressman named Lyndon Johnson lobbied hard for an exception to the rules on how many customers a line needed per mile. In 1938 the Pedernales Electric Cooperative was formed. Its loan covered 1,718 miles of line and $1,332,000, at the time the longest mileage and largest sum ever approved in a single REA grant.
The first line was energized on November 1, 1939, serving 422 members in Burnet County.
By 1965, 98% of Texas farms had power.
It was one of the clearest times America decided electricity had to reach places the market would not reach on its own. The next decision is being made now, around data centers, and Texas is at the center of it again.
ALT 'Here It Comes', Rural Electrification Administration poster by Lester Beall, about 1937. Library of Congress. Public domain.