Last Week in AI: Spend caps, SpaceX, and jobs

Hey there,

It’s week two of Last Week in AI and wow this week was packed. Here's what made the cut:

• Anthropic confidentially submitted a draft S-1. Obviously the biggest news of the week. Lots of memes about Anthropic tweeting that they "confidentially" filed. Lots of talk about joining the permanent underclass if you don't own shares yet.

• Two podcasts from this week: on All-In, OpenAI CFO Sarah Friar, the thread was IPO optionality, the compute crunch, and OpenAI's economics. On Lenny's Podcast, Benedict Evans, the frame was 1997: clearly huge, early, and unresolved on products, workflows, and winners.

• Google is paying SpaceX for compute which was revealed in a filing with the SEC. $920 million per month for ~110,000 NVIDIA GPUs. Yes, there are a lot of interesting parts here: Google owns a stake in SpaceX, Google is generally a TPU shop (so this might be for Google Cloud, not Gemini), and this now means SpaceX has similar revenues in their data center business as Starlink.

---

Two updates from me:

1. Are spend caps bullish?

tl;dr: yes. A cap moves AI spend from experimental budget to run-rate budget (probably with multi-year contracts if the Anthropic & Cursor reps had a say in the matter).

Uber reportedly set a $1,500 monthly cap per employee, per AI coding tool. Once a public company has a visible anchor, every CFO gets a cleaner way to approve AI spend. The enterprise is memetic.

2. Will AI decrease employment?

tl;dr: no. AI will kill some task-shaped work, make many roles more productive, and push more people toward smaller companies and new firms. Payrolls are too blunt to catch that in real time, especially when narrowed to just one country.

As for next week, Oracle reports Wednesday but SpaceX's IPO on Friday is the big item to watch. I was wondering which of the AI era companies would go public first, and Cursor might get to claim that via this Schrodinger's IPO?

LFG,

Drew