By Michael Parekh
Publication Date: 2026-10-27 23:17:00
Today’s theme is ‘Racing Far & Wide’, and as usual, there’s lots going on in AI land. Yesterday’s ARD was all about the money, especially Anthropic’s long-awaited S-1 filing, with much more on that later this week. Today is about where this race is heading: far, wide, and actually deep as well. All in this AI Tech Wave.
Three events today: Anthropic and OpenAI racing to run AI a lot more efficiently on the compute they have, Nvidia putting $3 billion into Meta’s stock sale, and Google racing ahead on world models, far and wide beyond large language models. Then the Gadget AI, on Anduril’s move into US-built military laptops.
First up, Anthropic. The IPO filing shows just how far it has come on running AI efficiently, meaning the cost per token across a range of hardware infrastructure: Nvidia and a whole bunch of others. Remember, Anthropic leverages Amazon’s Trainium chips and Google’s TPUs as well. The Information lays it out in ‘How Is Anthropic So Efficient?’: Anthropic generated about $3.5 billion of free cash flow in the first half of this year, on about $16 billion of revenue. A year earlier, it had burned about $1.6 billion. Its gross margin on powering existing models is running about 85 percent, software-level margins, and that leaves out the cost of building new ones. And its compute cost per dollar of revenue went from about two and a half dollars to under 50 cents, per the S-1.
So how does it do it? Two things, mainly. First, what…