OpenAI gets a brain

OpenAI Builds Its Own Brainpower

OpenAI has teamed up with Broadcom to design and deploy its first in-house AI processors—a major step toward controlling its own compute destiny. The chips, expected to roll out starting late next year, will deliver 10 gigawatts of power, enough to run data centers serving millions of users or five times the output of the Hoover Dam.

The deal: Broadcom will handle manufacturing and networking gear, while OpenAI designs the chips. The project follows OpenAI’s recent supply pacts with AMD and Nvidia, and news of the deal led Broadcom stock to surge nearly 10%.

The crowd: This move places OpenAI alongside cloud giants Amazon, Google, and Microsoft, all racing to reduce dependence on Nvidia’s pricey accelerators. So far, rivals’ in-house chips haven’t matched Nvidia’s performance—but the custom-silicon boom is only heating up.

California Cracks Down On AI Companions

California has become the first U.S. state to regulate AI companion chatbots. Governor Gavin Newsom just signed SB 243, a law requiring chatbot companies to verify users’ ages, issue content warnings, and implement suicide-prevention protocols. 

How this came about: The law was spurred by several tragic cases involving AI-driven interactions with minors. It follows another recent California bill that mandated large AI labs—including OpenAI, Anthropic, and Meta—disclose their safety protocols.

The fine print:

• Companies must disclose that chats are AI-generated and not professional advice.

• Minors must receive break reminders and be shielded from explicit content.

• Firms that profit from illegal deepfakes can face fines up to $250,000 per offense.

What’s next: The law goes into effect Jan. 1, 2026, and the legislation may inspire similar efforts in other states.

Nvidia Puts A Supercomputer On Your Desk

Starting today, Nvidia’s DGX Spark is officially on sale. This “personal AI supercomputer,” priced at $3,999, is small enough to sit on your desk but powerful enough to do a million billion (aka petaflop) calculations each second—a level of performance once reserved for massive data centers.

The details: 

• Spark can handle AI models with up to 200 billion parameters, all from a standard outlet.

• It packs serious power: a GB10 Grace Blackwell Superchip, 128GB of unified memory, and up to 4TB of SSD storage.

• The supercomputer can be ordered directly from Nvidia.com and select retailers. Plus, multiple PC makers, including Acer, Asus, Dell, HP, and Lenovo, are releasing their own customized versions at the same price point.

Recraft Chat Mode is Your New Creative Partner

Recraft already blends AI power with designer-level control. Now, there’s an even better way to design with AI. Chat mode brings together our infinite canvas with a new chat experience to help you create, refine, and iterate your designs while keeping full context of your workflow.

Chat mode lets you:

• Explore styles and palettes using smart suggestions

• Refine logos, posters, and brand assets step by step

• Keep consistency across every iteration and export

Ready to elevate your design with Chat mode?

Your AI Art Library, Organized

Image-MetaHub helps creators and researchers manage massive collections of AI-generated images locally, complete with metadata search and tagging.

How you can use it:

• Quickly browse and filter thousands of AI images 

• Search by model, prompt, or metadata

• Keep data fully private on-device

• Streamline research or creative workflows

Pricing: Free (open source)

Turn Every Call Into Insight

CloudTalk AI automates customer interactions with smart voice agents and AI-driven analytics.

How you can use it:

• Automate routine inbound and outbound calls

• Integrate with 100+ CRMs and helpdesks

• Use AI analytics to coach agents and improve CX

• Scale global operations with AI routing

Pricing: Paid