{{first_name | Leader}}, welcome back.
Three developments that are changing AI infrastructure, model economics, and governance.
These are today's updates.
See why teams are switching to Kimi K3 on Fireworks*
China's biggest AI chip milestone yet
Claude Opus 5 cuts the cost of enterprise AI
Meta's latest AI moderation controversy
Funding, jobs, and enterprise AI resources
Plus: an AI Bottleneck Finder to uncover productivity gaps ⬇️
. Together with Fireworks.
Fireworks is an inference and training platform for turning open models into production-ready AI. Pay per token with APIs compatible with OpenAI and Anthropic, without managing GPUs. You can now access Kimi K3 on Fireworks with 2.8T parameters, a 1M-token context window, and native image and video support.
Trusted by teams at Cursor, Notion, and Cognition, and performance that rivals leading closed models on independent benchmarks.
With Kimi K3 on Fireworks, you can:
- Ship and train from day zero: No waitlist, served on Fireworks' GPUs with support for serverless training to specialize the model for your unique workloads.
- Opus quality at Sonnet prices: 40% less than Opus-class pricing at $3 per million input tokens and $15 output.
- Built for regulated teams: with US-only serverless endpoints and zero data retention, teams in finance, healthcare, and other regulated industries can run K3 without reservation.
. Top News.
Chinese memory chip maker CXMT surged more than 470% after its Shanghai debut, giving it a market value of about $488 billion. The $8.6 billion IPO was the largest chip listing ever on mainland China. While CXMT remains smaller than Samsung, SK Hynix, and Micron in the global DRAM market, it now has the backing and capital to expand quickly.
Memory chips are one of the biggest AI infrastructure bottlenecks. A well-funded Chinese competitor could affect future chip supply, pricing, and the cost of building AI systems.
Anthropic released Claude Opus 5, its new flagship model for everyday coding, knowledge work, and AI agents. The company says it delivers performance close to Claude Fable 5 while costing half. It also uses fewer tokens for similar results, helping developers lower inference costs without giving up much performance.
If Opus 5 delivers on its claims, businesses can build AI agents and developer workflows at a much lower operating cost.
Facebook and Instagram displayed more than 7,600 ads for AI "nudify" apps promoted through a China-based advertising partner. The apps used AI to generate fake nude images and videos without consent, even though Meta's policies ban this type of content. Another incident that highlights a familiar question for AI platforms: when moderation conflicts with revenue, how consistently are safety policies enforced?
. Signals.
Tools
Links
. Poll.
What’s stopping your company from getting more value from AI?
. Market.
Funding
$120M raised by Candid Health to automate medical billing
$400M: Meshy AI bets big on AI-powered 3D
Roles In AI
Lead Cloud Platform Engineer at i.AI
Head of AI at AssetWatch
Socials
. Prompt of the Day.
AI Bottleneck Finder
When to use this?
When you want to identify hidden inefficiencies where AI can save the most time.
You are an AI operations consultant.
Based on the company information below, analyze each department (Marketing, Sales, Customer Success, HR, Finance, Operations, and Product).
For each team, identify:
* The biggest productivity bottlenecks
* Repetitive tasks consuming the most time
* Areas where information gets stuck
* The highest-impact AI opportunity
* Estimated business impact (High / Medium / Low)
* The best process to automate first
Then summarize:
* The top 5 productivity gaps across the company
* 3 AI quick wins that can be implemented in under 30 days
* The recommended order of implementation
Present the output as a concise executive report with tables.
Company information:
P.S. Get more such prompts in the Prompting Playbook (free for you)
Stay curious, {{first_name | leaders}}
PS. If you missed yesterday’s issue, you can find it here.