On Wednesday, Kalley and Cory broke the news that former GitHub CEO Nat Friedman and Safe Superintelligence co-founder Daniel Gross were in advanced talks to join Meta Platforms. Along with the hires, Meta is also discussing buying out a portion of the duo’s fund, which raised more than $1 billion two years ago.
Jun 19, 2025
Dealmaker
By Miles Kruppa
Sponsored by
Over 300 organizations rely on The Information for exclusive insights into public and private companies, including in-depth analysis of the ways in which tech moves markets. Click here to contact our corporate and enterprise team to learn more.
Welcome back!
On Wednesday, Kalley and Cory broke the news that former GitHub CEO Nat Friedman and Safe Superintelligence co-founder Daniel Gross were in advanced talks to join Meta Platforms. Along with the hires, Meta is also discussing buying out a portion of the duo’s fund, which raised more than $1 billion two years ago.
This is an extraordinary turn of events for the fund. On the plus side, limited partners should be thrilled to get quick cash returns on some of the firm’s high-valued but illiquid investments, including AI search engine Perplexity, coding copilot Cursor and, of course, Safe Superintelligence, the startup formed by former OpenAI chief scientist Ilya Sutskever. Harvard University’s endowment is one of the fund’s LPs, I’m told.
On the other hand, Meta is slated to become an indirect stakeholder in the startups—a fact that may worry founders competing with the internet giant.
Meta will only be taking a minority stake and will not have information rights, which usually include financial statements and board minutes. That may mollify those founders. Limited partners, meanwhile, will retain a portion of their original stakes in the NFDG fund as well as get some cash back.
My bigger question is around how Safe Superintelligence stakeholders feel about the deal. From what I’m hearing, the announcement and Gross’ imminent departure caught some of them off guard. Stay tuned!—Natasha Mascarenhas
Now, over to Miles for the rest of the column...
AI Investors’ New Favorite Metric
During the dot-com boom, investors often used metrics such as page views to justify soaring valuations for companies like Pets.com and eToys. Has the artificial intelligence boom found its alternative data point of choice?
Investors and tech executives are beginning to point to the rising consumption of AI building blocks known as tokens to support their optimistic bets on the technology. Mentions of tokens are popping up everywhere from earnings calls to venture capital presentations, often as a proxy for broader AI adoption.
AI foundation models like OpenAI’s GPT-4.1 digest passages of text by breaking them down into tokens—usually comprising just a few letters instead of a whole word. OpenAI and other makers of AI models generally charge developers on a per-token basis to integrate models into their apps.
“It’s like the atomic unit of AI,” said Philippe Laffont, founder of investment firm Coatue Management, which had $54 billion in assets under management at the end of last year, during a talk last week at the Economic Club of New York.
Laffont said Coatue weighed growth in AI token consumption as a more important factor for the stock market than the U.S. deficit and tariff threats. Or, as Coatue put it in a recent presentation: “Tokens >> tariffs!!”
“People are feeling, what if AI is going to create this kind of incredible productivity boom?” Laffont said during the New York event.
Coatue has used the token thesis to justify some of its big AI investments. At the firm’s annual East Meets West conference in Montecito, Calif., this week, Coatue told attendees token growth is leading to a “new inflection in computer demand,” benefiting shares in companies like Broadcom and Taiwan Semiconductor Manufacturing Co., according to a presentation it uploaded to its website.
In a section on private markets, Coatue also cited a rise in the number of tokens processed by AI cloud services providerTogether AIas evidence that the uptake of reasoning models is leading to growing AI consumption. Together is processing 24 trillion tokens annually as of this month, a six times increase since the third quarter last year, Coatue said. (Together sells developer access to open-source AI models on a per-token basis.)
Coatue has been particularly bullish on AI, placing bets on chipmaker Nvidia, cloud upstart CoreWeave, Together AI and enterprise search startup Glean. It’s also invested in the first $10 billion installment of OpenAI’s current round. But it isn’t the only big investor keeping an eye on token consumption.
Brad Gerstner, CEO of OpenAI backer Altimeter Capital, pointed to “parabolic” token volume at Google as a bullish signal for AI adoption during a recent CNBC appearance. Google CEO Sundar Pichai said last month at the company’s annual I/O conference that it processed more than 480 trillion tokens in April, a 50 times increase from one year prior.
The number of corporate earnings calls and conference presentations mentioning AI tokens has so far more than tripled this year compared to the same period last year, according to AlphaSense AI research. Microsoft CEO Satya Nadella jumped on the trend during the company’s April earnings call, telling investors it had processed more than 100 trillion tokens in the first quarter, a fivefold increase from the same period last year.
“If the market harbored any doubt for the insatiable demand for AI, this statement during Microsoft’s quarterly earnings yesterday quashed it,” Theory Ventures founder Tomasz Tunguz wrote in a blog post the following day.
Many investors and executives seem to be taking cues from Nvidia CEO Jensen Huang, who has likened tokens to a new kind of commodity powering the AI boom, similar to electricity. Huang recently gave the same pitch at a conference hosted by Sequoia Capital.
There are, of course, some limitations to a token-based analysis. For one, not all tokens are created equal. AI developers like OpenAI charge a wide range of prices per million tokens processed for different kinds of models. Overall per-token prices have dropped significantly over the past yearas Nvidia and other specialized AI chips have become cheaper and easier to access. If prices keep falling, it may become increasingly difficult for companies to make a profit simply by selling access to AI models.
AI bulls like Coatue may not care as long as consumption and revenue keeps going up. Many have pointed to an economic concept called Jevons Paradox and predicted AI usage will only go up as prices fall.
That might also help explain why tokens are a key part of the pitch for Humain, the AI startup owned by Saudi Arabia’s more than $900 billion sovereign wealth fund. Humain is betting the country’s significant energy resources will help it become the lowest-cost provider of access to AI chips in the world and help it diversify away from oil.
“Today, Saudi Arabia is leading the world in energy exports through oil,” according to a board presentation viewed by The Information. “In the future, Saudi Arabia will lead the world in energy exports through AI tokens.”
Start your day withApplied AI, the newsletter from The Information that uncovers how leading businesses are leveraging AI to automate tasks across the board.Subscribe nowfor free to get it delivered straight to your inbox twice a week.
Save the date for The Information's next AI summit, AI Agenda Live: The Next Wave. AI Agenda Live will help attendees make sense of what’s coming in the field, from new research breakthroughs to bleeding-edge AI applications.
Reporters Cory Weinberg and Natasha Mascarenhas tell you what’s coming next, who’s winning—and who’s losing—in the high-stakes world of startup investing.
