Happy Tuesday.
The current thing in tech and business is Anthropic’s announcement that it’s now doing $30B ARR, up from $9B at the end of 2025.
Today’s lineupOpenAI Member of Technical Staff Riley Walz at 12:00 PM
Noon Co-Founder & Co-CEO Aditya Bandi at 12:20 PM
Hermeus President Zach Shore at 12:30 PM
mappedin Co-Founder & CEO Hongwei Liu at 12:40 PM
Lewis-Burke Associates Chair, Tech & Venture Practice Zak Kukoff at 12:50 PM
Coatue Management Co-Founder Thomas Laffont at 1:15 PM
By John Coogan and Brandon Gorrell
Image: AlamyIntel joins the Elon Corp’s Terafab project to “design, fabricate, and package ultra-high-performance chips at scale.” The announcement comes after Elon was at Intel HQ over the weekend. He has said that Terafab ultimately will produce 1 TW/year of compute power. Intel’s stock popped 2% ahead of open on the news.
I’ve been eagerly awaiting a big cornerstone customer to show up for Intel. In fact, back in 2022, I was trying to imagine a world where Elon bought Intel outright. He had just marshaled $44B to buy Twitter, the 2022 CHIPS Act put together $280B (including $53B specifically for U.S. semiconductor manufacturing). Intel closed the year under $110B market cap. It wasn’t that much of a stretch to imagine something happening here four years ago, but we didn’t wind up on that path. Hopefully this is the first step on the long road to generate enough demand for domestic chip manufacturing to really move the needle.
Questionable takes swirl around Meta’s internal effort to increase AI usage among its staff. Meta employees are competing on a leaderboard called “Claudenomics” to use the most tokens, according to The Information. A chart — and viral take — accompanying the story claims Meta staff used 60.2 trillion tokens over 30 days, which would pencil out to a third of Anthropic ARR. Both of these claims are pretty questionable, for reasons we’ll discuss on the show.
Tyler did some back-of-the-envelope math to show that the “one-third” revenue estimate is way too high. Still an important data point in how power-law-driven token consumption can be. Meta has close to 80,000 employees and probably tens of thousands of them writing code or working close enough to engineering disciplines to make use of large token budgets. Still, to spend a billion dollars on tokens in a single month would require 10,000 engineers each spending $83K. Jensen threw out a per-engineer AI budget of $250K per year, which was met with some skepticism, even though it works out to $20K per month. Meta might get to that level of per-employee AI spend eventually, but it’s unlikely that it’s happening today. Regardless, this is yet another reason that the strategy around Meta Superintelligence Labs makes sense, not only with AI improving all the core products through better recommendations and ad targeting, but Meta will consume a ton of tokens internally just running their business.
Anthropic says it’s reached $30B ARR and announces agreement with Google and Broadcom for multiple gigawatts of TPU capacity, which will be coming online next year. This $30B figure is over triple Anthropic’s $9B run rate at the end of last year. Crazy numbers. From the company’s blog post:
When we announced our Series G fundraising in February, we shared that over 500 business customers were each spending over $1 million on an annualized basis. Today that number exceeds 1,000, doubling in less than two months.
Absolutely staggering chart! AI demand, and particularly demand for frontier tokens (especially when properly delivered through a coding agent) is truly showing no signs of slowing down. The only potential drag on growth seems to be potentially marshaling enough compute to fulfill demand, but the industry is quick to reallocate resources to the highest possible value and at $25 per million output tokens of Opus 4.6, hyperscalers are already starting to help on the supply side. Ben Thompson writes that “Google does not see a clear path to using all of that capacity for themselves, either for their own services or, more pertinently, for their Gemini enterprise offerings. That’s ok, though: the beauty of Google Cloud from Google’s perspective is that they can still capture AI growth broadly as long as compute is happening on Google Cloud.”
Is it completely over for “the models will commoditize” narrative at this point? Companies seem eager to pay for the latest and greatest time and time again.
Bloomberg reports that OpenAI, Anthropic, and Google have begun sharing information through the Frontier Model Forum, a nonprofit the three companies set up with Microsoft in 2023, to prevent Chinese competitors from distilling their models. Distillation involves training models on other model outputs — OpenAI et al consider this a national security risk, and a threat to their own ability to price their models.
The final step in any discussion around AI safety always turns to the global AI race. If China is consistently able to catch up through distillation, it makes any discussion around mutual agreement between American AI labs for longer testing procedures pretty moot. As models get bigger and more powerful, hopefully it becomes easier to track distillation efforts. Also, at a time when there is intense competition between every lab, it’s reassuring to see that the opportunity for some limited collaboration still exists.
We send this email every weekday, right as our show goes live. Subscribe to get it in your inbox.
HeadlinesThe Information: Meta Employees Vie for AI ‘Token Legend’ Status
The Deep View: Anthropic passes $30B ARR after Pentagon clash
Hermeus raises $350M at a $1B valuation
Reuters: Intel joins Musk’s Terafab AI chip project to power humanoid, data center goals
WSJ: What to Know About OpenAI’s Ideas for a World With ‘Superintelligence’
Bloomberg: OpenAI, Anthropic, Google Unite to Combat Model Copying in China
WSJ: Inside a Corporate Retreat That Went Very Badly Wrong
NYT: Bill Ackman’s Fund Offers to Buy Universal Music in $64 Billion Deal
NYT: Shots Fired at Indianapolis Councilman’s Home, After Vote Backing Data Center
AP says it will offer buyouts as part of pivot away from newspaper-focused history
Wired: Edward ‘Big Balls’ Coristine Is Helping Out on Viral Fraud Videos Now
Bloomberg: Cerebras Backer Eclipse Raises $1.3 Billion for Robotics, AI Infrastructure
WSJ: Five Amazing Tech Innovations We Should Expect in the Next 25 Years
The Wrap: McClatchy Journalists Revolt Against AI: ‘It’s a Betrayal’ | Exclusive
SF Alexandria: Engineering Feelings
Posts of the Day