The AI buildout just got financed with debt, not just equity
Good morning ๐ SoftBank just borrowed $11.87 billion to keep funding OpenAI, and Anthropic locked in $13.7B of compute from the company behind Rumble and Truth Social. The money behind AI is starting to look like project finance, not venture capital.
In today's issue:
- ๐ฐ SoftBank borrows $11.87B to keep funding OpenAI, Anthropic locks in a $13.7B RUM Group compute deal
- ๐ ๏ธ Try this: a prompt-defined PII gate that lets you check in less, for real
- ๐ Quick links: the deals in full, plus ChatGPT's billion-user storage layer
Get tomorrow's issue in your inbox.
One concise AI brief, sent after the signal clears the noise.
๐ญ THE ONE THING
๐ฐ The AI buildout just got financed with debt, not just equity
Two deals this week show where AI money is actually coming from. SoftBank locked in an $11.87B two-year loan from roughly 20 banks for its OpenAI bet, above the original $10B target and part of a ~$65B commitment due by October. Anthropic signed a $13.7B, six-year compute deal with RUM Group (the company behind Rumble and Truth Social), plus warrants on 51 million RUM shares at a penny each.
Neither of those is equity. Banks are underwriting compute like a hard asset with a payback schedule, not funding a story on faith. That's the real signal: credit markets now treat AI compute as collateral, not a bet. Watch the loan terms next time, not the valuation. Those are the ones that get tested when a quarter goes badly.
๐ ๏ธ TRY THIS
Give the agent you're supervising less a PII tripwire, not more check-ins
1. Pull AWS's pattern for a model-agnostic PII detector: the entity list lives in the prompt, not the weights, so you add a new entity type by editing text. It already beats an off-the-shelf detector across five public benchmarks.
2. Point it at whatever Bedrock model you're already running and wire it as a gate on your agent's output, before anything reaches a ticket, a Slack channel, or a customer.
3. Build the review surface as an MCP App on Bedrock AgentCore. It's host-agnostic, so the same interactive widget shows up whether you're checking in from Claude or ChatGPT.
4. Only let the gate ping you on a flag. Silence means it ran clean, which is the whole point of checking in less.
Prompt: List the PII entity types that matter for [your domain] (e.g. SSN, medical record number, account ID). For each, write a one-line detection instruction an LLM can follow with no examples. Output as a numbered list I can paste straight into a system prompt.๐ QUICK LINKS
- SoftBank Locked in an $11.87B two-year loan from roughly 20 banks for its OpenAI bet, above the original $10B target and part of a ~$65B commitment due by October. Bloomberg
- Anthropic Is the GPU customer behind RUM Group's $13.7B, six-year compute deal (yes, the Rumble/Truth Social company), plus warrants on 51 million RUM shares at a penny each. The Information
- OpenAI Its storage platform Habitat now handles more than 70 million requests per second across 1 billion ChatGPT users. OpenAI
That's the one worth sitting with today.
Pradeep Perugu