Now Even Google's Buying GPUs From SpaceX?
TL;DR
Google may be paying SpaceX nearly $1 billion a month for GPUs: The hosts cite roughly 110,000 Blackwell GPUs at about $11.60 per hour, a deal they frame as both a compute scramble and a signal that even Google still wants Nvidia and CUDA despite its TPU story.
Cloudflare buying Void Zero looks like a direct answer to its biggest weakness: Void Zero was not just a tooling company but a would-be Cloudflare-based neo-cloud, and the acquisition gives Cloudflare a path to make Vite-native deployment much less painful.
Anthropic is finally saying the quiet part out loud about recursive AI progress: Their paper argues that AI already accelerates AI development through human-mediated loops, and even floats the idea that a global pause on frontier AI could be good if coordination were possible.
Claude is writing most of Anthropic's code, but the hard part is still 'taste': The hosts think code generation is no longer the bottleneck. The real unknown is whether models can develop the judgment to choose the right research directions, not just execute more experiments.
Cloudflare still feels more locked-in than Vercel, not less: Despite Vercel's reputation for vendor lock-in, the hosts argue that Cloudflare apps often require much more platform-specific wiring through Wrangler, Durable Objects, KV, and routing quirks.
Hermes agents feel like a glimpse of the next interface layer for AI: Running Hermes in Discord with many built-in skills gives a more capable, assistant-like system than simple chat agents, but it also creates a bigger black box that constantly self-modifies through skill updates and memory.
The Breakdown
Google is reportedly spending about $920 million a month on SpaceX Blackwell compute, putting it within 10% of Anthropic's xAI bill and puncturing the story that TPUs are enough. The episode also hits Cloudflare's acquisition of Void Zero, Anthropic's unusually candid paper on AI accelerating AI research, and why both hosts are weirdly obsessed with Discord-based Hermes agents.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.

Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
The Art of Tasteful Prompting
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.