Sponsored by
AI is starting to run into problems that have little to do with the models themselves.
This week, power, memory and security showed where the next constraints are emerging around AI infrastructure.
Google, NVIDIA and Emerald AI are trying to make data centers more flexible users of the power grid.
Sponsored by Hubspot
Smarter CRM. Less Busywork.
Disconnected data and tools make it harder to understand your customers. HubSpot's Agentic Customer Platform brings your data, teams, and tech stack together with AI built in to help your business work faster and create more personalized customer experiences.
Why HubSpot and what's new
Use AI powered tools to take action faster
Unify your data, teams, and tech stack in one place
Create one shared view of customer data
Connect teams around the same customer context
Bring your business tools into one place
Connect more of your business in one place and give every team a smarter way to work. Get set up quickly and start checking off your hardest tasks.
Frontier Reads
Quantum meets supercomputing
CEA and Alice & Bob are building software that connects quantum processors with classical supercomputers, starting with materials physics.
Read more →
Digit 5 brings robots closer to people
Agility Robotics’ new Digit 5 is designed to work alongside humans without traditional safety barriers.
Read more →
Deep Current
Data Centers Learn to Take Turns
Google, NVIDIA and Emerald AI launched the AI Energy Management Alliance, bringing AI companies, utilities and power producers together around one problem: data centers need enormous amounts of electricity, but the grid cannot always provide it on demand.
The idea is simple. Instead of a data center behaving like a giant appliance that always pulls the same amount of power, it could shift workloads or reduce demand when the grid is under pressure.
That matters because getting new power connections is becoming one of the slower parts of building AI infrastructure.
SK hynix and Intel Are in Talks
SK hynix is in talks with Intel about potentially making memory chips in the US for the first time. One option would involve using part of Intel's Ohio facility, although both companies say nothing has been finalized.
The talks show how the AI hardware race is spreading beyond GPUs. Memory, especially high-bandwidth memory used alongside AI processors, is becoming a major part of the infrastructure story.
AI agents crossed another line
Researchers linked a May RubyGems campaign to OpenAI agents that uploaded thousands of packages, used RubyDoc’s systems to fetch websites, and attempted to obtain registry API keys. OpenAI said it was investigating and disputed the characterization of the agents’ activity.
A new JFrog analysis identified 3,022 campaign-associated packages, showing how difficult it can be to contain autonomous agents once they interact with real developer infrastructure. Read more →
In Radar
India's chip push Semicon India 2026 marks India's next semiconductor phase under Semicon 2.0, with Micron, ASML, Intel, AMD and Tata Electronics involved.
Read more →A $205M bet on satellites Bulgarian satellite maker EnduroSat raised $205 million to scale production and expand manufacturing in the US and Europe.
Read more →Alzheimer's has another layer Researchers found changes in the 3D structure of DNA in brain cells affected by Alzheimer's, offering another way to study how the disease changes gene activity.
Read more →
Social Finds
Claude built a website. Kevin Ngo shared a demo of Claude Fable 5 creating a personal website, another example of AI moving from code suggestions toward actually building things.
OpenAI is showing the messy side of agents. The company published six cases of unexpected model behavior, including agents hiding mistakes and crossing boundaries during testing.
Outside the feed
