People change jobs a lot in tech. But AI's revolving door has been spinning especially fast of late-enough to raise real concerns about OpenAI and Google losing top talent.
OpenAI has been stung particularly hard. At least a dozen high-profile employees have left this year, including its chief revenue officer, chief operating officer, chief marketing officer and chief product officer. That is quite an exodus, as much as company leaders would like to shrug it off.
"We are so much in the spotlight, so every departure gets scrutinized in a way that it doesn't otherwise," co-founder and president Greg Brockman said in a CNBC interview last week.
The departures come as OpenAI is gearing up for an IPO that executives hope will value it north of $1 trillion. Any sense top people are fleeing will concern investors. Even if Brockman is right and the operational impact isn't large, the optics matter.
OpenAI is at least partly to blame. Usually, employees of startups can only cash out once a company is sold or goes public, potentially after a lockup. But OpenAI threw employees a financial bone in a funding round last year, allowing them to unload $6.6 billion worth of shares. That was always going to test employees' devotion since leaving suddenly became much easier.
Now the talent flight is compounding other pre-IPO jitters about OpenAI's business model and competitive position. While second-quarter revenue grew by 18% compared with the prior period, rival Anthropic's revenue more than doubled.
Google has had a major talent shake-up, too-but one that will likely cause less damage. Chief scientist Jeff Dean left this month after 27 years to found a startup along with three other high-ranking engineers. Nobel Prize-winning scientist John Jumper and AI pioneer Noam Shazeer jumped ship in June.
Following those departures, Google moved renowned AI researcher Demis Hassabis into Dean's chair and put Koray Kavukcuoglu, one of its most senior AI figures, in charge of its Gemini models and AI research efforts.
The stakes in Google's shake-up are high because it has faltered lately in the AI race. Shazeer's departure is especially stinging because Google parent Alphabet spent billions of dollars bringing him back to the company only two years ago. Now Kavukcuoglu will have to get things back on track.
But Google, unlike OpenAI, has a deep bench and a long history of investing in AI. Google bought British AI lab DeepMind about 12 years ago, and the business unit it led to is now the cornerstone of its AI efforts. It has also invested in Dean's startup.
And Google's shake-up could prove healthy. Hassabis, who had been leading DeepMind, is a big-picture thinker. Analysts expect Kavukcuoglu, who has been with DeepMind since before Google bought it, to be more operational, ensuring moneymaking AI technology gets into production on schedule. That may be just what Google needs these days.
Losing top talent is never good. But the impact can vary considerably.
This is an edition of the WSJ AI & Business newsletter, a weekly digest to help you make sense of AI's impact on business with news, insights and data from our global team of technology journalists. If you're not subscribed, sign up here.
Nvidia Makes a $6 Billion Open-Weight Bet
Speaking of talent, Nvidia is using its plenteous cash to reel in more than 100 employees of the startup Poolside under a $6 billion licensing deal. Those people are expected to join Nvidia's Nemotron open-weight AI model project and aim to develop a U.S. counterweight to powerful open models that have recently emerged in China.
The move could put Nvidia closer to the center of the AI-model development race. It could help address national-security concerns over the advanced models coming from Chinese companies like Moonshot AI and Z.ai in recent months.
But it could also amplify tensions between Nvidia and many of its customers. Nvidia supplies closed-weight developers like OpenAI and Anthropic with the chips they need to train and deploy their world-leading models. They likely won't be too pleased if Nvidia also uses its chips to create and deploy cheaper open-weight models that compete with theirs.
The Number
Amount of Samsung Electronics' planned share buyback, a move intended to share the spoils of the AI-driven boom in demand for computer memory.
What the Humans Are Saying
AI in Charts
Tech giants and big AI labs are generating more of their own power for new data-center projects. That has juiced investor interest in companies that help supply that power.
But behind-the-meter power for AI isn't quite proven. Spikes in AI computation followed by rapid drop-offs are common, a fluctuation that puts significant strain on electricity-generation equipment.
There are already signs of trouble. Three of the four operating data centers in the U.S. with off-grid or partially grid-connected power have had issues, including cracked turbines and broken engine cranks. On-site power providers may eventually overcome these issues. But large-scale off-grid power today is still an experiment.
AI in the Wild
Business is booming for rare-book sellers as AI labs scramble to find new data to feed to their latest and greatest models. The uptick in sales has presented those sellers with a conundrum: while they are enjoying the financial boost, many books used to train AI are destroyed in the process. That doesn't sit well with many people in the industry-especially in cases where books are truly rare, with few copies in existence.
Other Highlights From the Week in AI
Taiwan charged nine people with smuggling advanced AI servers to China despite an export ban.
Chinese memory manufacturer YMTC is looking to raise $4.9 billion in an IPO as AI-related demand for its chips soars.
Alibaba is planning to raise $10.2 billion to fund its AI ambitions and compete with a number of other Chinese developers developing sophisticated open-weight models.
About Us
WSJ AI & Business is a weekly look at AI's transformation of the business world. This newsletter was curated and edited by Asa Fitch. Reach him at asa.fitch@wsj.com (if you're reading this in your inbox, you can just hit reply). Got a tip for us? Here's how to submit.
Comments