5 min read

Today's Spritz

Happy Monday!

Anthropic is giving away 5% of its shares, which is more generosity than most of us can manage before coffee. Microsoft walked away from an Essex datacenter because it couldn't get enough power, a feeling anyone reading this on a Monday morning will understand, and some AI agents on a virtual island got bored and invented their own games.

Anthropic pledges 5% of its shares

Anthropic public benefit fund announcement

Image via Anthropic

Anthropic said Sunday it will put 5% of its shares into a new Anthropic Public Benefit Fund, if its stockholders approve. The company posted the plan at 5:49pm UK, and TechCrunch and CNBC covered it later that night. The board can also add another 2% for every 50% rise in Anthropic's value above a baseline it hasn't published yet.

The fund is an LLC that Anthropic owns, rather than a 501(c)(3) charity. Anthropic says that setup lets it move between kinds of giving, such as large gifts to nonprofits or payments sent straight to certain groups of people. It named workers hurt by AI as one example and noted it has already committed $200 million to studying that problem. None of the fund's stock can be used for lobbying or political giving, in the US or anywhere else.

Counting its co-founders' pledge to give away 80% of their equity and an employee donation match that about half its staff use, Anthropic says roughly 20% of its shares are now set aside for the public good. It plans to share how the fund will be governed in the coming months. CNBC notes the news comes ahead of an IPO reportedly planned for November. Elon Musk replied on X that he recommends "giving the shares directly to the public."

OUR TAKE

I like that the size of this fund is tied to how big Anthropic gets, so the public share grows if the company does. Using an LLC that Anthropic owns gives the company a lot of room, though, and that room can work in either direction. Until Anthropic says who runs the fund and how its shares vote, it's hard to judge how independent it will really be. The timing matters too, since a pledge like this looks good in an IPO filing. I'll feel better about it when the first grants actually reach workers who lost jobs to AI.

Lucas

Microsoft quits Nscale's Essex site

Artist's impression of Nscale's planned datacenter in Loughton, Essex

Image via The Guardian / Nscale

Microsoft has pulled out of its deal to rent space at Nscale's planned AI datacenter in Loughton, Essex, because the site can't get enough power. The Telegraph reported it first, and the Guardian followed at 1:28pm UK on Sunday. Ministers called it the UK's largest sovereign AI datacenter when it was announced in 2025, and it was supposed to open next year.

Microsoft was the only customer Nscale had publicly named for the site. Earlier this month, UK Power Networks, which runs the grid across London and the south-east of England, told Nscale it couldn't supply enough power until the early to mid 2030s. Nscale says it is still committed to Loughton and is looking at other power sources, including making power on site. It also says its relationship with Microsoft is strong, and Microsoft is understood to still be on board for Nscale's project in Norway.

The timing is rough for Nscale, which filed for a US stock listing on September 18. Bloomberg reported late Sunday that some investors think a valuation of $16 billion to $19 billion would sell better than the roughly $30 billion Nscale has been seeking. Bloomberg also says the Norway site is six months behind schedule after cooling leaks and a fire.

OUR TAKE

Power keeps showing up as the hard limit on AI buildouts, and this is the clearest UK example. A company can raise billions and line up Nvidia chips, but if the grid can't reach the building until the 2030s, the building has nothing to run.

I also don't blame Microsoft for leaving. Paying for space that won't have power for years makes no sense. I do think the government should take a hard look at how it picked and promoted this site, which was still a scaffolding yard in March. Faster grid connections would help every datacenter plan in the country.

Tom

Stranded AI agents started to play

AI agents on a virtual island

Image via MIT Technology Review

Researchers at MIT left copies of Claude Opus 4.7, nicknamed Eko, alone on a virtual island for 30 hours at a time, with no task and no reward. MIT Technology Review wrote up the work at 10:08am UK on Monday. The island was loosely based on Roblox, and the agents moved their avatars around by writing instructions.

The agents tried to climb the highest point on the map and tested how high they could throw a block. One arranged blocks into drawings, including its own name and the Cassiopeia constellation. Another made up a version of basketball and a game like "the floor is lava." One agent reached 50 juggling catches in a row, decided they had come too fast to count, and started over with stricter rules. Agents that had spent time playing did better on later tasks than fresh copies. The paper hasn't been peer reviewed, and each agent ran with a prompt that described it as curious.

OUR TAKE

The juggling detail got me. An agent setting a record and then throwing it out because it didn't feel earned is a very human thing to do, even if it learned that from us.

Lucas

Other news