Compute and seats scale while safety and culture get audited
AWS OpenAI partnership and Kimi K2 Thinking, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
OpenAI
AWS OpenAI partnership
DEPLOYif you run OpenAI workloads
WHAT SHIPPEDAWS and OpenAI announced a multi-year, strategic partnership that provides AWS infrastructure to run and scale OpenAI's core AI workloads starting immediately. The $38 billion agreement gives OpenAI access to AWS compute comprising hundreds of thousands of NVIDIA GPUs, with the ability to expand to tens of millions of CPUs.
THE CATCHThe agreement runs over seven years with all capacity targeted to be deployed before the end of 2026. The clusters use both GB200s and GB300s via Amazon EC2 UltraServers, and AWS says clusters already top 500K chips.
IF YOU RUNChatGPT inference or next generation model training and need more compute
SOURCEOpenAI, AWS and OpenAI announce multi-year strategic partnership
Vercel
Kimi K2 Thinking
DEPLOYif you build reasoning agents
WHAT SHIPPEDMoonshot AI's Kimi K2 Thinking and Kimi K2 Thinking Turbo are now available on Vercel's AI Gateway with no other provider accounts required. Kimi K2 Thinking handles up to 200 to 300 sequential tool calls.
THE CATCHKimi K2 Thinking is open source and achieves top results on reasoning and coding benchmarks. Turbo is the high speed version for scenarios requiring both deep reasoning and low latency.
IF YOU RUNagentic workflows that need many sequential tool calls
SOURCEVercel, Moonshot AI Kimi K2 Thinking and Kimi K2 Thinking Turbo are now available
The Deploy Log
Models and APIs
| DEPLOYif you sell AI to businesses 1 million business customers. OpenAI announced more than 1 million business customers around the world are directly using OpenAI, with more than 7 million total ChatGPT for Work seats, up 40% in just 2 months. OpenAI |
| WATCHif you evaluate multilingual models IndQA benchmark. OpenAI released IndQA, a new benchmark that evaluates AI systems on Indian culture and languages, spanning 2,278 questions across 12 languages and 10 cultural domains, created in partnership with 261 domain experts. OpenAI |
| DEPLOYif you run customer support AI CRED Cleo. CRED built Cleo, an AI conversational companion powered by OpenAI models including GPT-4.0, GPT-5, and o3, and in three months since launch Cleo reached a 98% resolution accuracy rate with a 14 percentage point improvement in CSAT scores. OpenAI |
Tools and products
| DEPLOYif you script isolated compute environments Vercel Sandbox CLI. Vercel introduced the Sandbox CLI, a command-line interface for managing isolated compute environments built on the Docker CLI model, supporting Node.js node22 and Python python3.13 workloads. Vercel |
| DEPLOYif you protect AI endpoints from bots BotID Deep Analysis. BotID Deep Analysis, Vercel's advanced bot protection system, is free for all Pro and Enterprise customers from November 5 to January 15, 2026, and uses thousands of telemetry points for real-time client-side checks. Vercel |
| WATCHif you build on Snowflake data v0 Snowflake integration. Vercel announced a Snowflake integration for v0 that lets you connect v0 to Snowflake, ask questions about your data, and build data-driven Next.js applications that deploy directly to Snowflake, with data never leaving your Snowflake environment. Vercel |
| DEPLOYif you roll out new versions gradually Skew Protection lifetime. Vercel Skew Protection max age can now persist for the entire lifetime of your deployments, removing the previous limits of 12 hours on Pro and 7 days on Enterprise. Vercel |
| WATCHif you use Edge Config on Pro Edge Config pricing. Vercel Edge Config Reads and Writes are moving from package-based to per-unit pricing on the Pro plan, with reads at $0.000003 per read and writes at $0.01 per write. Vercel |
Breakthroughs
| WATCHif you automate data science tasks DS-STAR data science agent. Google Research introduced DS-STAR, a data science agent that raised accuracy from 41.0% to 45.2% on DABStep, 39.8% to 44.7% on KramaBench, and 37.0% to 38.5% on DA-Code compared to the best alternative. Google Research |
Left out
- Cognizant Claude rollout: a figure in the draft could not be found on the publisher's own page
- BillionToOne IPO: a figure in the draft could not be found on the publisher's own page
- Nested Learning: the publisher's page does not show it available yet
- Chime marketing AI: it did not carry a number that moved
- Redirects observability: it did not carry a number that moved
- Static IPs for builds: it did not carry a number that moved
- Waymo safety audits: it did not carry a number that moved
- Post-quantum TLS: it did not carry a number that moved
- AI domain search: it did not carry a number that moved
The pattern
The $38 billion AWS OpenAI partnership deploys hundreds of thousands of NVIDIA GPUs before the end of 2026 while OpenAI reports 1 million business customers and 7 million ChatGPT for Work seats, up 40% in 2 months. Vercel answers this scale with per-unit Edge Config pricing at $0.000003 per read and $0.01 per write, a Sandbox CLI for node22 and python3.13, and Skew Protection that persists for the entire deployment lifetime.
The same week, safety and culture become audit surfaces. BotID Deep Analysis is free for Pro and Enterprise customers from November 5 to January 15, 2026, and IndQA spans 2,278 questions across 12 languages and 10 cultural domains built with 261 domain experts. DS-STAR raises data science accuracy from 41.0% to 45.2% on DABStep, while CRED Cleo reaches a 98% resolution accuracy rate with a 14 percentage point improvement in CSAT scores using GPT-4.0, GPT-5, and o3.