The Girl From Vlorë
There is a coastal city in Albania called Vlorë. It sits where the Adriatic Sea meets the Ionian Sea — a city of salt air, fishing boats, and mountains in the distance. It is beautiful, and for most of its recent history, it was isolated.
Albania had been one of the most closed countries in the world — a communist state that banned private cars, foreign books, and contact with the outside world for nearly half a century. By the time Ermira "Mira" Murati was born there on December 16, 1988, the walls were just beginning to crack.
Her father was a civil engineer. Her mother was an electrical engineer. In a household full of blueprints and equations, the youngest daughter grew up thinking that taking things apart to understand them was simply what you did. Math competitions. Science olympiads. A fierce, quiet hunger to understand how the world worked.
When Mira was sixteen years old, she won a United World Colleges scholarship — a program that selects exceptional students from over 80 countries and gives them a full scholarship to study abroad. Out of all the students in Albania, she was chosen.
She packed her bags and flew to Vancouver Island, Canada — a girl from a post-communist coastal city landing in one of the most beautiful places on earth, surrounded by students from 80 nations. She graduated with an International Baccalaureate in 2007.
From there, she did something unusual. Rather than picking one subject, she enrolled in a dual-degree programme across two American universities — earning a Bachelor of Arts in Mathematics from Colby College and a Bachelor of Engineering in Mechanical Engineering from Dartmouth College. Art brain and engineer brain — both, at once.
That combination — the ability to think beautifully AND build precisely — would define everything she built afterwards.
Goldman Sachs, Tesla, and the Road to OpenAI
After graduating, Mira didn't go straight into AI. She went to Goldman Sachs in Tokyo as a summer analyst — one of the most competitive internships in finance. Then to Zodiac Aerospace, building concepts for aircraft. Then, in 2013, she joined a small electric car company that was trying to do something nobody thought was possible.
That company was Tesla. And Mira's job was to help build the Model X — the first electric SUV with falcon-wing doors that the whole world would eventually stop and stare at. She spent three years there, learning under Elon Musk, understanding what it took to ship a hardware product that millions of people would use.
Then she moved to Leap Motion, a startup building augmented reality technology — the kind that layers digital information on top of the real world. She was learning, accumulating, assembling a rare combination of skills that very few people in the world had.
Grows up in an engineering household during Albania's post-communist transition.
Wins a prestigious scholarship and leaves Albania for Pearson College, Vancouver Island. One of the defining moments of her life.
Completes a dual degree in Mathematics and Mechanical Engineering. Goldman Sachs internship in Tokyo in 2011.
Helps build one of the most complex cars ever made. Learns to ship world-changing hardware products at scale.
Works on technology that blends the digital and physical worlds. Begins moving towards AI research.
Steps into the most consequential AI lab in the world. The chapter that changes history begins.
In 2018, she walked into the offices of OpenAI — a research lab in San Francisco that was trying to build artificial general intelligence safely. She was hired as VP of Applied AI and Partnerships. Within four years, she was their Chief Technology Officer. The person responsible for all of it — ChatGPT, DALL-E, Codex, Sora. Every product that made OpenAI famous passed through her hands.
"Mira has helped build some of the most exciting AI technologies we've ever seen, including ChatGPT, DALL-E, and GPT-4."
— Satya Nadella, CEO of Microsoft, writing for TIME's 100 Next listThe Night Sam Altman Was Fired — And Mira Became CEO
On November 17, 2023 — a Friday evening — the OpenAI board of directors did something that shocked the entire technology world. They fired Sam Altman, the CEO of the most talked-about company on earth.
The reasons were murky. The board said he had not been "consistently candid." The actual internal dynamics were complicated — disagreements about safety, commercialisation speed, and the direction of the company. What we know is that the decision involved a board member named Ilya Sutskever, and that Mira herself had provided information to the board that contributed to their decision.
And then — with almost no warning — the OpenAI board did the most natural thing they could think of. They called Mira Murati and asked her to become the interim CEO of OpenAI.
Mira agreed to step in. She became the first woman to lead OpenAI, even if only for days. But what followed was one of the strangest weeks in tech history.
Sam Altman didn't go quietly. He lobbied. Microsoft — OpenAI's biggest investor — lobbied. 700 of OpenAI's 770 employees signed a letter threatening to leave if Altman wasn't reinstated. Mira herself eventually sided with the employees who wanted him back.
Three days into her stint as CEO, the board brought in a new interim CEO named Emmett Shear. Two days after that — Sam Altman was back. The board members who had fired him were gone. The whole saga lasted five days. And Mira, who had navigated it all with quiet steadiness, returned to her role as CTO.
Some saw her as caught in the middle. Others saw a person of rare judgment — someone who could hold the centre when everything around her was collapsing.
Less than a year later — on September 25, 2024 — Mira Murati sent an email to the OpenAI team. She was stepping down as CTO. After six years and the most impactful run in AI history, she wanted, as she put it, to "do her own exploration."
The tech world held its breath. What would she build?
Thinking Machines Lab — Born in February 2025
On a cold morning in February 2025, Mira Murati registered a new company in San Francisco. She called it Thinking Machines Lab. The mission, in her own words: to build AI that is "more widely understood, customizable, and generally capable."
Three words stood out above all others: customizable. Unlike OpenAI or Anthropic, which build fixed AI products and sell access to them, Mira wanted to build AI that people and companies could shape to their own needs. AI that you don't just use — AI that you make your own.
She assembled a team of roughly 30 leading researchers — people from OpenAI, Meta, Mistral, Google DeepMind. Names that the AI world recognised. John Schulman, one of OpenAI's co-founders, joined. Alec Radford — the researcher who helped invent the GPT architecture that underpins ChatGPT — came on as an advisor.
By July 2025 — just five months after founding — Thinking Machines had raised a $2 billion seed round led by Andreessen Horowitz, one of the most powerful venture funds in the world. Investors included Accel, ServiceNow, Cisco, Jane Street, and — interestingly — both Nvidia and AMD, two rival chip companies, both betting on Mira's lab at the same time. The company was valued at $12 billion before it had shipped a single product.
Among the investors in Thinking Machines Lab's seed round was the government of Albania — Mira's home country, putting official national capital behind its most famous daughter. A girl from Vlorë, now backed by her nation.
In October 2025, Thinking Machines shipped its first product — an API called Tinker. Instead of being another chatbot, Tinker was designed to help researchers and developers fine-tune AI models without needing massive computing infrastructure of their own. This was a product built for builders.
Imagine you hire a brilliant new employee who has read every book ever written and knows everything about everything. But your company has very specific ways of doing things — your own rules, your own tone, your own products. Fine-tuning is like giving that employee a two-week training programme so they work exactly the way your company needs.
In AI terms: a base model like GPT is trained on billions of documents and knows a lot. But fine-tuning feeds it your specific data — your company's documents, your style guide, your product catalogue — so it answers questions exactly the way you need. That's what Tinker makes easy. No massive computing setup required.
Turmoil, Zuckerberg, and The Test of 2025
Not everything went smoothly. No good story does.
In November 2025, one of Thinking Machines' co-founders quietly left for Meta — Mark Zuckerberg's AI empire, which had been aggressively trying to poach talent across the entire industry. Then, on January 14, 2026, came the news that made every AI group chat in San Francisco erupt at once.
Mira fired co-founder Barret Zoph — reportedly for having a relationship with an employee. Within an hour, OpenAI's CEO of Applications announced that Zoph, another co-founder Luke Metz, and researcher Sam Schoenholz had all rejoined OpenAI. Three founding members gone in one morning. For a one-year-old startup, that was a wound the whole industry noticed.
Before all of this, Mark Zuckerberg had reportedly made Mira Murati a staggering offer — a multi-billion dollar acquisition of Thinking Machines, or failing that, enormous personal packages to join Meta. Mira turned all of it down. She wanted to build her own thing, on her own terms. That decision, in hindsight, looks like one of the most consequential choices in AI history.
The whispers started. Was Thinking Machines stable? Could a 120-person lab survive losing its founding team? Was the dream already unravelling?
Then came March 10, 2026.
The Nvidia Deal — The Rebuttal Heard Around the World
On March 10, 2026 — less than two months after the co-founder departures — Mira Murati and Jensen Huang, the CEO of Nvidia, stood together and announced a partnership that nobody in the industry had been expecting at this scale.
Thinking Machines Lab and Nvidia had signed a multi-year strategic partnership. The deal had two headline components. First: Nvidia was making a "significant investment" in Thinking Machines Lab — an undisclosed amount, but described by both companies as major. Second: Thinking Machines had committed to deploying at least one gigawatt of Nvidia's next-generation Vera Rubin chips starting in early 2027.
The Financial Times reported the chip supply component of the deal was worth tens of billions of dollars.
"Nvidia's technology is the foundation on which the entire field is built. This partnership accelerates our capacity to build AI that people can shape and make their own, as it shapes human potential in turn."
— Mira Murati, CEO of Thinking Machines Lab, March 10, 2026
The timing was unmistakable. This was not just a business deal. This was a statement. Jensen Huang — the most powerful person in AI infrastructure — was putting real chips, real capital, and real reputational weight behind Mira's lab, right after its most difficult public moment.
As one analyst wrote at the time: "In AI's 2026 reality, compute is credibility. And Thinking Machines just bought itself a much more convincing chair at the adults' table."
Think of AI models like cars. Most AI is like a reliable family car — it does its job well. A "frontier model" is the absolute bleeding edge — think Formula 1. It's the most capable, most powerful, most expensive-to-build AI that exists at this moment in history. ChatGPT-4, Claude 3 Opus, Gemini Ultra — these are frontier models.
Only a handful of organisations in the world have the computing power, money, and talent to build frontier models. OpenAI, Google DeepMind, Anthropic, Meta — and now, with this Nvidia deal, Thinking Machines Lab has officially declared it intends to be in that same league. That's what makes this deal so significant.
A gigawatt is a unit of power — the same way we measure how bright a light bulb is in watts. One gigawatt of power is enough to run approximately 750,000 homes. When an AI lab says it is deploying "one gigawatt of compute," it means it is building server farms that consume that much electricity to run its AI chips continuously.
Training a frontier AI model — teaching it by processing hundreds of billions of data points — requires unimaginable amounts of computing power. A gigawatt commitment is a threshold that only the very largest AI labs in the world have approached. It is a signal that you are not building a tool — you are building infrastructure to compete with the biggest players on earth.
Nvidia names its chips after scientists. The Vera Rubin chips — named after the astronomer who discovered dark matter — are Nvidia's most advanced AI chips, released in 2026. They are the successor to the Hopper and Blackwell series that powered the current generation of AI. Think of them as the newest, fastest, most powerful engine Nvidia has ever built.
Training a model like ChatGPT requires thousands of these chips running simultaneously for months. By securing a commitment of one gigawatt of Vera Rubin systems, Thinking Machines is not just buying chips — it is locking in the most advanced computing infrastructure available before its rivals can get their hands on it. In AI right now, whoever secures the best chips first has a durable advantage.
Why This Deal Matters — Beyond the Numbers
It would be easy to read this story as just another big tech business deal. Billions of dollars, chip agreements, multi-year partnerships. The kind of thing that only matters to investors and analysts.
But look at what this deal actually represents — and the story becomes something different.
This Is a Story About Independent Vision
Mira Murati was offered billions by Mark Zuckerberg. She said no. She was at the centre of the most powerful AI company in the world. She left. She lost co-founders. She absorbed public criticism. And then, thirteen months after founding Thinking Machines, she secured the largest infrastructure commitment any independent AI lab had ever made.
Not because she played the game everyone else was playing. But because she built something people believed in — an AI company that isn't trying to be another ChatGPT, but something with a different philosophy entirely. AI you can shape. AI you can make your own.
The deal also tells us something important about where AI is heading in 2026. The era of "who has the best demo" is over. The race is now about infrastructure. Who has the chips. Who has the power. Who has secured the computing resources to train the next generation of models — years before those models exist. The deals being signed today are bets on a future that hasn't happened yet.
And Jensen Huang — who has seen every major AI bet of the last decade — chose to put Nvidia's money, its chips, and its name behind Mira Murati. That alone tells you something.
Mira will appear alongside Jensen Huang at Nvidia's GTC conference to discuss the future of open frontier models. Thinking Machines is hiring rapidly — with job postings suggesting it's working on AI models for audio processing and visual reasoning. The Vera Rubin chips arrive in early 2027. And then, the real test begins: can a 120-person lab actually build a frontier model that competes with OpenAI, Google, and Anthropic? That question is worth watching very closely.
"She is no longer short of compute to try."
— The Next Web, March 11, 2026The Full Circle Moment
In 2024, Dartmouth College — the same university that once taught a girl from Albania how to think like an engineer — awarded Mira Murati an honorary Doctor of Science. The citation read that she had "democratized technology and advanced a better, safer world for us all."
A girl from Vlorë. A scholarship at sixteen. A career that built the tools billions of people use every day. A boardroom crisis she navigated with steadiness. A billion-dollar offer she turned down. And now — a deal that puts her, and a 120-person startup from San Francisco, at the centre of the most important technological race of our lifetime.
That is the story of Mira Murati. And it is very far from over.