OpenAI's Executive Exodus: What's Going On?

Chris Malone left OpenAI last week.

His title was head of data centers. In the summer of 2026, this is probably the last position an AI lab should leave vacant—more than 500 towns across the U.S. are restricting data center construction, Texas is reviewing all grid connection applications, New York State has proposed a one-year moratorium, and every AI company is competing for the same batch of power access permits.

Now, the person in charge of this has left.

Who is he, and why does his departure stand out?

Malone joined OpenAI in March 2025, shortly after the announcement of the Stargate project—the ambitious self-built data center plan that OpenAI, Oracle, and SoftBank were working on together.

His resume was tailor-made for this: before OpenAI, he spent nearly five years at Meta, rising to distinguished engineer and leading the company's data center strategy; before that, he spent over a decade at Google, also as a distinguished engineer and senior director, focusing on data center technology.

The problem is, Stargate got off to a rough start. OpenAI subsequently pivoted, switching to signing contracts with cloud providers for compute rather than building its own facilities. Now it is restarting some self-build efforts—no longer just renting chips, but renting entire facilities. But someone else is leading this, not Malone.

The organizational changes are even more telling. Malone originally reported directly to President Greg Brockman. Earlier this year, the infrastructure division was reorganized, and Vice President Sachin Katti took over the larger group, reporting to Brockman, while Malone and another leader, Adrian Caulfield, became co-heads of a "data center technology engineering and design" team.

In July, OpenAI promoted Uday Ruddarraju to "Chief Technology Officer for Compute Capacity," reporting directly to Brockman; Brent Mayo, poached from Musk's xAI this year, reports to Ruddarraju and is responsible for ensuring compute projects are completed on schedule. Both previously built Musk's Colossus supercomputer in Memphis.

In other words, before he left, the weight of this position had already been diluted.

OpenAI's official response sidestepped this: "Earlier this year, we reorganized our infrastructure organization to support the scale and pace of our work. We have a strong, deeply experienced data center team with clear leadership and the technical expertise needed to execute our plans."

Why are so many people leaving?

Malone is not an isolated case. He is the latest in a string of departures this year.

The Wall Street Journal counted at least 12 high-profile employees, Business Insider counted 13, several of whom left in the past month. And these are not junior staff—they are among the company's top executives: chief revenue officer, chief operating officer, chief marketing officer, chief product officer.

The list looks like this:

Product lead Kevin Weil left in April;

Fidji Simo—Sam Altman's number two, overseeing much of OpenAI's core business—resigned last month after a medical leave;

Longtime executive and COO Brad Lightcap announced this month he is leaving to pursue new projects;

CRO Denise Dresser, poached from the CEO role at Slack just last December, left after eight months.

Two weeks ago, OpenAI appointed its second CRO within a year—Dali Rajic, president and COO of Alphabet-owned Wiz.

Typically, startup employees' stock can only be cashed out after the company is acquired or goes public, possibly with a lock-up period in between. This is the so-called "golden handcuffs": if you want that money, you have to stay.

But OpenAI gave employees a financial sweetener in a funding round last year, allowing them to sell a total of $6.6 billion worth of shares.

Once the money is in hand, the handcuffs loosen. Leaving suddenly becomes much easier.

Google is also losing people—why isn't anyone panicking?

OpenAI is now preparing for an IPO, expected in 2027, with executives hoping for a valuation above $1 trillion.

At this critical moment, any perception that "top people are fleeing" will unsettle investors. Even if Brockman is right that the operational impact is minimal, perception itself is an asset.

And the talent drain is colliding with two other ongoing issues: questions about the business model, and competitive positioning. In the second quarter, OpenAI's revenue grew 18% year-over-year—in the same quarter, rival Anthropic's revenue more than doubled. It is now playing catch-up with Anthropic in enterprise sales.

On the other side, the people leaving Google are no less significant.

Chief Scientist Jeff Dean left this month after 27 years at the company to start his own venture, taking three other senior engineers with him. In June, Nobel laureate John Jumper and AI pioneer Noam Shazeer also left—Shazeer's departure stings especially, because Alphabet spent billions just two years ago to bring him back.

But Google has a deep bench. It bought UK-based DeepMind 12 years ago, and that division is now the cornerstone of its AI business. It even invested in Dean's new company.

Succession was also quick: Demis Hassabis moved from DeepMind into Dean's role, and Koray Kavukcuoglu took over Gemini models and AI research—the latter was at DeepMind before Google acquired it. Analysts' take: Hassabis is a big-picture thinker, while Kavukcuoglu will be more operational, focused on ensuring revenue-generating AI technology is delivered on schedule.

For Google, which has recently fallen behind in the AI race, this shake-up might even be healthy.

Same departures, but the difference lies in who's left behind.

The poacher is their common supplier

There was another personnel story the same week: Nvidia took more than 100 employees from startup Poolside in a $6 billion licensing deal.

These people are expected to join Nvidia's Nemotron open-weight model project, with a clear goal—to create an American counterweight to the powerful open-source models recently emerging from China.

This has two sides. One is the national security narrative: models recently released by Chinese companies like Moonshot AI and Zhipu have indeed made Washington nervous. The other is the business relationship: Nvidia supplies chips to closed-source developers like OpenAI and Anthropic, and now it is using its own chips to build a cheaper open-source model that competes with its customers.

And this road runs both ways. On the same day Malone's departure was reported, OpenAI announced benchmark results for its in-house chip Jalapeño at the Hot Chips conference, claiming it beats Nvidia's GB300 on both throughput per kilowatt and response speed—700 watts versus 1400 watts, up to 1.9x throughput per kilowatt and 3.6x lower latency. The chip, developed with Broadcom, is inference-only, not for training, with small-scale deployment planned for the end of this year and a broader rollout in 2027.

The supplier is moving into the customer's territory, and the customer is moving into the supplier's territory.

But there's one thing both sides still have to solve on their own: those chips eventually need to go into data centers, and data centers need power. And at OpenAI, the person in charge of that just left.