← Back to archive

Tuesday, May 5, 2026

OpenAI's $10B PE play (and Musk's confession)

OpenAI just closed a $10B joint venture with 19 PE firms to embed AI across their portfolio companies (bold move), while Musk admitted xAI trained Grok using OpenAI's models via distillation (wild). Meanwhile, Meta's jumping into humanoid robots with an acquisition of ARI for Superintelligence Labs. So here's the question: Would you trust a $10B PE deployment of AI across companies that probably aren't ready for it?

Top Stories

1
OpenAI finalizes $10B "deployment company" JV with 19 PE firms to embed AI in their portfolio companies

Bloomberg

OpenAI has formed a $10 billion joint venture with 19 PE firms to systematically deploy its AI technology across thousands of portfolio companies, marking a significant shift toward enterprise-scale AI adoption through financial industry partnerships.

openaienterprise-ai-adoptionfundingprivate-equity
2
Meta is moving into humanoid robots

Meta acquired humanoid robotics startup ARI to bolster its Superintelligence Labs division, bringing top-tier robotics talent to develop foundation models for whole-body humanoid control as part of its path toward AGI through physical world training.

metaroboticshumanoidacquisition
3
Musk admits Grok was trained using OpenAI's models

Elon Musk confirmed in court that xAI trained Grok using distillation techniques on OpenAI's models, validating assumptions that U.S. AI labs use these methods on each other despite publicly opposing the practice when done by Chinese competitors.

xaiopenaidistillationgrok
4
OpenAI adds animated Pets and config imports to Codex

Testing Catalog

OpenAI updated Codex with playful animated Pets overlays and quietly added auto-import of competitor config files, signaling a strategic shift toward making Codex a sticky, cross-agent desktop super-app beyond pure coding functionality.

openaicodexagentsdeveloper-tools
5
How did 'large' language models get that way? The role of Transformers and Pretraining in GPT

GreaterWrong

The article traces how Transformers' parallel processing architecture and self-supervised pretraining on massive unlabeled datasets enabled LLMs to scale dramatically, with supervised fine-tuning and reinforcement learning added later for alignment and specific behaviors.

llmtransformerspretrainingrlhf

Keep Reading

Industry Voices

Enjoyed this issue?

Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.