The Thinking Machines Exodus: Why Mira Murati’s Top Talent is Leaving After Stock Vesting

 

 The Thinking Machines Exodus: Why Mira Murati’s Top Talent is Leaving After Stock Vesting

**Subheading:** *A third of the founding team is gone. Meta, OpenAI, and xAI are picking off the rest. With nine-figure offers on the table and "one-year cliffs" unlocking, is this the new reality of the AI talent war?*

**Estimated Read Time:** 15 minutes

**Target Keywords:** *Thinking Machines Lab talent exodus, Mira Murati team departures, AI talent war 2026, one-year cliff vesting, Meta poaching AI researchers, OpenAI hiring spree, xAI recruitment, stock options startup retention, AI compensation packages, Tinker AI model.*

## Part 1: The Human Touch – The $1.5 Million LinkedIn Message

Let me tell you about the most expensive subject line in recruiting history.

Sam Agre is a recruiter. Not the kind who sends generic "Let's connect" messages. The kind who knows that in the current AI talent market, subtlety is a waste of time.

When he wants to poach an engineer from Thinking Machines Lab, he gets straight to the point. The subject line of his LinkedIn message reads:

**"$1.5 million cash 'and up'"** 

That's not a typo. That's the annual cash compensation being offered to top AI researchers. It's more than **three times** the maximum salary Thinking Machines Lab advertises on its careers page, which tops out at $475,000 .

And here's the kicker: that's just the cash. The total packages—including stock, bonuses, and guarantees—are reaching **nine figures**. Hundreds of millions of dollars over several years .

One researcher was reportedly offered **$1 billion** to jump ship .

Let that sink in. We're not talking about CEOs or hedge fund managers. We're talking about engineers. Coders. The people who build the models that power your ChatGPT queries.

This is the reality of the AI talent war in 2026. And the epicenter of that war right now is a one-year-old startup called **Thinking Machines Lab**.

Founded by Mira Murati—the former OpenAI CTO who briefly ran the company after Sam Altman's shocking 2023 firing—Thinking Machines Lab launched with a bang in early 2025. It raised **$2 billion** in a seed round at a **$12 billion valuation** before it even had a product . It assembled a "dream team" of 42 founding members, many of whom had built ChatGPT together at OpenAI.

Investors were ecstatic. The media was breathless. This was going to be the next great AI lab.

Fast forward to May 2026. And the dream is cracking.

Nearly a **third of the founding team**—13 people, including three of the six co-founders—have left . Meta has poached seven founding members. OpenAI has taken five. Even Elon Musk's xAI has grabbed one .

The exodus was triggered by something mundane yet utterly predictable: **the one-year cliff**.

In startup compensation, equity typically vests over four years with a **one-year cliff**—meaning you get nothing if you leave before 12 months, but after that, you unlock your first chunk of shares. For the founding team, that cliff just hit .

And as soon as those shares were in hand, the floodgates opened.

"Why stay," the thinking goes, "when Meta is offering me $200 million to leave?"

This is the story of how one of the most promising AI startups in the world is being picked apart by the very giants it hoped to challenge. It's a story about loyalty, leverage, and the brutal math of the AI arms race.

Let me walk you through what happened, why it matters, and whether Thinking Machines Lab can survive the exodus.

## Part 2: The Professional – The Numbers Behind the Exodus

Let's put on our analyst hats. No drama. Just the facts.

### The Scorecard: Who Left and Where They Went

Here is the current tally of founding team departures from Thinking Machines Lab, based on a Business Insider review of LinkedIn profiles and conversations with sources :

| Name | Role at TML | Destination | Timing |

|------|-------------|-------------|--------|

| Andrew Tulloch | Co-founder | Meta | October 2025 |

| Barret Zoph | Co-founder / CTO | OpenAI | January 2026 |

| Luke Metz | Co-founder | OpenAI | January 2026 |

| Sam Schoenholz | Researcher | OpenAI | January 2026 |

| Christian Gibson | Founding member | Meta | February 2026 |

| Noah Shpak | Founding member | Meta | February 2026 |

| Ian O'Connell | Researcher | Meta | January 2026 |

| Joshua Gross | Founding engineer | Meta | March 2026 |

| Jolene Parish | Top executive | OpenAI | February 2026 |

| Lia Guy | Researcher | — | Recent |

| +3 others | Various | Various | Various |

In total: **13 of 42 founding team members** have departed. That's **31%** of the original team .

Meta has been the most aggressive poacher, securing at least seven founding members. OpenAI has taken five. xAI has taken one .

### The Compensation Math: Why They're Leaving

To understand why people are leaving, you have to understand the numbers being thrown around.

| Compensation Element | TML Standard | Poaching Offer |

|---------------------|--------------|----------------|

| **Base Salary** | $350k - $475k | $1.5M+ cash |

| **Total Annual Package** | ~$500k - $700k | $10M - $50M+ |

| **Multi-Year Guarantee** | Standard 4-year vest | $200M - $1B |

One defector told Business Insider: *"I got an opportunity that I couldn't turn down"* .

The $1 billion offer—reportedly made by Meta to a single researcher—was denied by Meta's communications director, Andy Stone, who said the figures being reported "aren't accurate" . But even if the $1 billion number is inflated, multiple sources confirm that **nine-figure packages** are being offered to top talent .

Recruiter Sam Agre put it bluntly: *"You don't want to fall behind in that type of arms race"* .

### The One-Year Cliff: A Design Flaw?

Here's the structural issue that made this exodus possible.

In standard startup compensation, equity grants vest over four years with a **one-year cliff**. That means:

- **Months 0-12:** No equity vests. If you leave before one year, you get nothing.

- **Month 12:** You vest 25% of your total grant.

- **Months 13-48:** You vest the remaining 75% monthly.

The cliff is designed to **retain** talent—to make employees think twice before leaving in the first year.

But in practice, the one-year cliff has become what one compensation consultant calls *"an Achilles' heel for startups ever since the AI boom began"* .

Why? Because once that first chunk of equity vests, employees can leave **without walking away empty-handed**. And with competitors offering nine-figure packages, the calculation becomes simple:

> *"I already have my first 25%. Why wait four years for the rest when I can get five times as much tomorrow?"*

Some compensation experts are now arguing that equity cliffs should be extended to **five years** to lock in early team members . But that's a hard sell when your competitors have no such restrictions.

### The Retention Math: What TML Is Doing About It

Thinking Machines Lab is not taking this lying down.

The company has more than quadrupled its headcount from 42 to **over 150 employees** . It has made high-profile hires of its own, including:

- **Soumith Chintala** as CTO—the creator of PyTorch, who left Meta to join Murati 

- **Kenny Yu** from Meta's elite TBD lab 

- **Neal Wu**, a three-time Olympic gold medalist programmer 

And in March, the company posted a job listing for a **"fresh framework for doling out equity"** —a senior hire dedicated to implementing systems that *"attract and retain highly sought-after talent"* . The base salary for that role: $250,000 to $425,000.

But can better equity design compete with nine-figure offers from Meta and OpenAI? That's the $200 million question.

## Part 3: The Creative – The "Office Romance" That Broke the Camel's Back

Now let me tell you the part of the story that sounds like a Netflix drama—because it's impossible to understand the Thinking Machines Lab exodus without understanding the **Barret Zoph affair**.

### The Summer of Secrets

It was summer 2025. Thinking Machines Lab was just a few months old. The team was still riding the high of that $2 billion seed round.

Then Mira Murati discovered something that would unravel everything.

Her CTO and co-founder, Barret Zoph—a veteran AI researcher who had spent six years at Google and held a VP role at OpenAI—was having a secret relationship with a junior employee .

The employee had since left the company, returning to OpenAI. But the damage was done.

When confronted, Zoph didn't just apologize. According to sources, he claimed he was **"manipulated"** into the relationship—painting himself as the victim .

Murati didn't fire him immediately. Instead, she stripped him of management authority, demoting him to a "technical contributor" role while allowing him to keep his co-founder title .

It was a compromise designed to preserve the team. It backfired spectacularly.

### The "Pizza Meeting" Betrayal

By January 2026, Zoph had had enough of his demotion. He was frequently absent, citing illness and family emergencies. His Slack status was perpetually gray .

Then came the ambush.

Zoph arrived at a meeting with two other co-founders, Luke Metz and Sam Schoenholz. They presented Murati with an ultimatum: **hand over all technical decision-making power to Zoph, or they would walk** .

Murati refused.

Two days later, Zoph was spotted at a **pizza restaurant**—not drowning his sorrows, but meeting with Meta executives to auction himself to the highest bidder .

On January 14, Murati announced Zoph's termination on X (formerly Twitter), citing *"lack of trust, poor performance, and unethical conduct"* .

Within **58 minutes**, OpenAI's applications CEO Fidji Simo announced that Zoph, Metz, and Schoenholz were rejoining OpenAI .

Zoph was given a new role: head of enterprise sales for OpenAI's business products. Metz and Schoenholz would report to him .

The betrayal was complete. And it was broadcast to the world in less than an hour.

### The Creative Hook: A Shakespearean Tragedy for the AI Age

If you're a screenwriter, you couldn't make this up.

- **The protagonist:** Mira Murati, the brilliant former CTO who briefly ran OpenAI, striking out on her own to build a more ethical, transparent AI lab.

- **The antagonist:** Barret Zoph, her trusted lieutenant, undone by an office romance and a hunger for power.

- **The twist:** The betrayal was orchestrated in secret with the very company Murati left—OpenAI—which welcomed the defectors back with open arms and lucrative roles.

- **The irony:** The "ethical" AI lab was undermined by the very human flaws of jealousy, ambition, and revenge.

As one commentator put it, this is *"the Silicon Valley AI soap opera"* .

And it's not over.

## Part 4: Viral Spread – The "Great AI Heist" and the Memes

A story with this much drama—betrayal, nine-figure paydays, office scandals—is made for viral spread.

### The Meme Angle

**Meme #1: "The 58-Minute Revenge"**

A timeline graphic: 12:00 PM - Zoph fired. 12:58 PM - Zoph hired by OpenAI. Caption: *"Fastest turnaround in AI history."*

**Meme #2: "The Pizza Meeting"**

A cartoon of Zoph eating pizza with Meta executives while a speech bubble says: *"So, about that $200 million..."* Caption: *"Not a great look for the 'ethical' AI lab."*

**Meme #3: "The One-Year Cliff"**

A cliff edge with a sign: "You must be this tall to get your equity." A line of engineers holding Meta offer letters jump off. Caption: *"The cliff that became a launchpad."*

### The Viral Headlines

Expect these headlines across social media:

- *"Mira Murati's AI dream team got their stock options. Now a third of them are gone."*

- *"The $1 billion offer: How Meta is buying the AI talent war one engineer at a time."*

- *"From co-founder to fired to rehired in 58 minutes: The Barret Zoph story."*

### The TikTok Angle

For the TikTok generation, the story needs simple framing:

- **"The 'Office Romance' That Destroyed an AI Startup":** *"A CTO had a secret relationship with a junior employee. When he got caught, he tried to stage a coup. Then he joined the competition in under an hour. This is not a TV show."*

- **"Your $1.5 Million LinkedIn DM":** *"Recruiters are sending messages with subject lines that just say '$1.5 million cash.' That's how crazy the AI talent war has gotten."*

- **"Why Everyone Is Leaving Thinking Machines Lab":** *"One-year cliff. That's it. That's the video."*

### The LinkedIn Angle

For professionals, the hook is strategic:

**"The Thinking Machines Lab exodus reveals a structural flaw in startup compensation. The one-year cliff, designed to retain talent, has become an expiration date. Once employees unlock their first equity tranche, the incentive to stay collapses. With competitors offering nine-figure packages, the math is brutal. For founders: consider extended cliffs or creative retention structures. For employees: the window for life-changing offers may not stay open forever."**

## Part 5: Pattern Recognition – The New Reality of AI Talent

Let me step back and show you the patterns emerging from this saga.

### Pattern One: The "Ex-OpenAI Mafia" Is Fracturing

Murati isn't the only former OpenAI executive trying to build a competing lab. There's a whole ecosystem of "ex-OpenAI" startups:

- **Anthropic** (founded by Dario Amodei and other former OpenAI researchers)

- **Thinking Machines Lab** (Murati)

- **Safe Superintelligence** (Ilya Sutskever)

Each of these companies is staffed by people who worked together at OpenAI. And each is now competing for the same limited pool of talent.

The result is a **circular talent flow**: OpenAI → Thinking Machines Lab → back to OpenAI. The same people are moving in and out of the same orbit, each time with a higher price tag.

### Pattern Two: The "Compute as a Counterweight"

Here's what might save Thinking Machines Lab from irrelevance.

On March 10, 2026—in the midst of the talent exodus—the company announced a massive partnership with **Nvidia** .

The deal: at least **one gigawatt** of next-generation Vera Rubin systems, with deployment targeted for early 2027. Nvidia is also making a "significant investment" in the company .

For context: **one gigawatt** is the kind of compute capacity that only the most elite AI labs can access. It's the infrastructure required to train frontier models.

This is the counter-narrative: Yes, Thinking Machines Lab lost talent. But it gained **credibility**—the kind that comes only from a Jensen Huang handshake.

As one analyst put it: *"Compute is credibility. Whoever secures chips, power, datacenter capacity, and the right systems talent gets a seat at the adults' table. Everyone else is doing a very expensive cosplay of a Frontier Lab"* .

### Pattern Three: The "Product Pivot" Narrative

Talent leaves. Compute arrives. But what about the actual product?

In October 2025, Thinking Machines Lab launched **Tinker**, a flexible API for fine-tuning language models . It was... fine. Not revolutionary. Not a ChatGPT-killer.

But on May 11, 2026—just two days ago—the company announced a **new type of AI model** that interacts seamlessly, handles interruptions, and translates languages in real time .

This is the pattern: the "dream team" narrative got them funding. The compute partnership got them credibility. The product announcements are what will determine if any of it matters.

### The Three Scenarios for Thinking Machines Lab

| Scenario | Probability | Description |

|----------|-------------|-------------|

| **The "Survival" Scenario** | 50% | The company stabilizes around 150-200 employees. The Nvidia partnership enables frontier model development. Tinker and the new model gain traction. The departures are remembered as a rough patch, not a death knell. |

| **The "Talent Drain" Scenario** | 35% | More key people leave as competitors continue to outbid. The company struggles to ship products that differentiate from OpenAI and Meta. The Nvidia partnership becomes a footnote. |

| **The "Acquisition" Scenario** | 15% | Murati eventually sells to a tech giant. (Meta tried to buy once already; she refused.) The talent is absorbed, and the brand disappears. |

## CONCLUSION: What the Thinking Machines Exodus Teaches Us

Let me give you the bottom line.

Thinking Machines Lab was supposed to be different. It was founded by the woman who briefly ran OpenAI. It raised $2 billion before shipping a product. It assembled a team of 42 founding members who had built ChatGPT together.

And now, a third of that team is gone.

The reasons are instructive for anyone who follows the AI industry:

**1. The one-year cliff is broken.** Designed to retain talent, it has become an expiration date. Once employees unlock their first equity tranche, the incentive to stay collapses.

**2. Nine-figure offers are real.** Whether it's $200 million or $1 billion, the numbers being thrown around are staggering. No amount of "mission alignment" competes with generational wealth.

**3. Loyalty is a liability.** The same people who built ChatGPT together are now being paid to tear each other's companies apart. The "ex-OpenAI" network is not a family. It's a talent pool with a price tag.

**4. Compute is the great equalizer.** Yes, Thinking Machines Lab lost talent. But it also locked in a Nvidia partnership that gives it access to compute capacity that most companies can only dream of. That alone keeps them in the game.

**What this means for you:**

- **If you're an AI engineer:** Your leverage has never been higher. The offers are real. But remember: today's nine-figure offer may not be available tomorrow. The market is hot, but markets cool.

- **If you're a founder:** Review your equity structure. The one-year cliff is a liability. Consider longer cliffs, creative retention bonuses, or back-loaded vesting schedules.

- **If you're an investor:** Bet on teams, but also bet on infrastructure. The companies that survive the talent wars will be the ones that own their compute capacity.

- **If you're just a user:** None of this changes your ChatGPT experience today. But in the long run, the consolidation of AI talent into three or four mega-labs means less competition, less innovation, and higher prices.

The Thinking Machines exodus is not an anomaly. It is the new normal.

The AI talent war is only getting started. And the next battle is already being planned in a pizza parlor somewhere in San Francisco.

## FREQUENTLY ASKING QUESTIONS (FAQ)

**Q1: What is Thinking Machines Lab?**

**A:** Thinking Machines Lab is an AI startup founded in early 2025 by Mira Murati, former CTO of OpenAI. The company raised $2 billion in seed funding at a $12 billion valuation before launching any product. It focuses on "agentic AI" systems that can take autonomous action .

**Q2: Who is Mira Murati?**

**A:** Mira Murati was OpenAI's Chief Technology Officer and briefly served as interim CEO after Sam Altman's shocking firing in November 2023. She left OpenAI in September 2024 to found Thinking Machines Lab .

**Q3: How many people have left Thinking Machines Lab?**

**A:** According to a Business Insider review, nearly a third of the 42-person founding team—13 people, including three of six co-founders—have left since the company launched .

**Q4: Where are they going?**

**A:** Meta has poached seven founding members, OpenAI has taken five, and Elon Musk's xAI has taken one . Notable departures include co-founder Andrew Tulloch (Meta), co-founder Barret Zoph (OpenAI), co-founder Luke Metz (OpenAI), and top executive Jolene Parish (OpenAI) .

**Q5: Why are people leaving?**

**A:** Two main reasons: (1) The "one-year cliff" has passed, allowing employees to unlock their first equity tranche and leave without losing everything. (2) Competitors like Meta and OpenAI are offering nine-figure compensation packages—hundreds of millions of dollars over several years .

**Q6: What happened with Barret Zoph?**

**A:** Zoph, the CTO and co-founder, was fired by Murati in January 2026 after a power struggle. The conflict began when Murati discovered Zoph had a secret relationship with a junior employee. Zoph was demoted, then attempted a "coup" by demanding more power. After being fired, he joined OpenAI within 58 minutes, along with two other co-founders .

**Q7: Is Thinking Machines Lab in trouble?**

**A:** The talent losses are significant, but the company has also grown to over 150 employees and made key hires, including Soumith Chintala (creator of PyTorch) as CTO. In March 2026, the company announced a major partnership with Nvidia for compute capacity .

**Q8: What products does Thinking Machines Lab have?**

**A:** The company launched its first product, Tinker—an API for fine-tuning language models—in October 2025. On May 11, 2026, it announced a new type of AI model that handles interruptions and translates languages in real time .

**Q9: What is a "one-year cliff"?**

**A:** In startup compensation, equity typically vests over four years with a one-year cliff. That means you get no equity if you leave before 12 months, but at month 12 you vest 25% of your grant. The cliff is designed to retain talent, but it has become a "natural moment" for employees to entertain outside offers .

**Q10: Could Thinking Machines Lab be acquired?**

**A:** Possibly. Meta reportedly approached Murati about acquiring the company in mid-2025, but she refused. With the talent exodus and valuation concerns, acquisition is one possible outcome, though Murati has publicly stated her desire to remain independent .

**Disclaimer:** This article is for informational and educational purposes only. The information presented is based on publicly available sources as of May 13, 2026. Talent movements, compensation figures, and corporate strategies are subject to rapid change. Please consult with appropriate professionals for advice specific to your situation.