Top highlights of the week
OpenAI explains how it made ChatGPT's data storage 12 times faster to read
Why it mattersOpenAI says ChatGPT reads got 12× faster on Azure Cosmos DB with global caching, a design you can copy for high-traffic products.
Seventy: Sunday, 25 October 2026
What happened. OpenAI describes how it rebuilt Habitat, its online storage system, using multi-tenancy, global caching and Azure Cosmos DB. It says reads for ChatGPT are now 12 times faster.
AWS shows how to stop automated updates breaking live AI services that scale automatically
Why it mattersTeams running AI endpoints on AWS can stop hand-fixing failed deployments and update live endpoints through their normal automated release process.
What happened. AWS says that editing a live SageMaker endpoint's setup, such as machine type, machine count or settings, makes automated CloudFormation and CDK deployments fail, because auto scaling has to be switched off first and CloudFormation then loses track of it. An AWS blog post describes a reusable CDK building block that turns auto scaling off and back on automatically during updates.
NVIDIA's weather AI lets energy, insurance and transport firms build local forecasts on their own data
Why it mattersPower, insurance and logistics firms can build local weather forecasts tuned to their own data, but the tools run on NVIDIA hardware.
What happened. NVIDIA says European organisations are using its Earth-2 family of open AI weather models, which developers can customise with their own data. The tools run on NVIDIA computing and cover global forecasts, regional forecasts and environmental research.
EU opens investigation into whether Booking.com does enough against fake listings and scams
Why it mattersIf you list on or advertise through online marketplaces, expect regulators to scrutinise scam controls and paid-ranking labels, and check yours now.
What happened. The European Commission has started a formal investigation into Booking.com under the Digital Services Act. It is examining whether the platform failed to properly assess and reduce the risk of fake listings and phishing scams, and whether it hid that some search rankings were paid for.
A restaurant franchisee used ChatGPT Work to cut price-update time by 95%
Why it mattersMulti-location operators can test ChatGPT Work on repricing and cash reconciliation, two routine back-office jobs, where one franchisee reports 94-95% less time.
What happened. An independent Schlotzsky's and Cinnabon franchisee says ChatGPT Work shortened menu repricing by 95% and cash reconciliation by 94%. The franchisee also set up a process that could release $100K in cash.
EU now has an official checklist for companies selling high-risk AI to follow
Why it mattersCompanies selling in the EU now have a defined standard to follow for the article 17 quality management requirements.
What happened. The EU has listed EN 18286:2026, written by CEN and CENELEC, in its Official Journal. It is the first formal technical standard tied to the AI Act and covers quality management systems under article 17.
07Chips
NVIDIA-powered Windows laptops from $2,399 can run AI on the device itself
Why it mattersStaff who work with sensitive data can now run AI on a $2,399+ laptop instead of sending it to a cloud service.
What happened. NVIDIA's RTX Spark laptops are on sale from $2,399, built by ASUS, Dell, HP, Lenovo, Microsoft and MSI. They pair Blackwell graphics with a memory pool of as much as 128GB shared across the machine, in slim Windows designs, and more models are due.
Early signals
- Fine-tuning & training9 stories from 3 sources this week, about 9× the usual pace.
- Deployments & case studies19 stories from 4 sources this week, about 3× the usual pace.
- Partnerships7 stories from 4 sources this week, about 3× the usual pace.
Counted across everything Seventy read this week, not only what made an issue: at least four stories from three sources, at twice the usual pace or more.
That's the whole issue.
7 stories, about 2 min 43 sec. Nothing was left out to fill the time.
Words of the day
- unified memory
- A single memory pool shared by the CPU and GPU, letting large AI models fit and run without copying data between separate memories.
- multi-tenancy
- A design where one shared system serves many separate customers or users while keeping their data and workloads isolated.
- CloudFormation
- An AWS service that creates and updates cloud resources automatically from template files describing the desired setup.
“I'm convinced that about half of what separates the successful entrepreneurs from the non-successful ones is pure perseverance.”