{{first_name | Hey}}, welcome back.
Today’s issue is about the growing divide over who should govern AI, the US chip industry’s struggle, and the widening gap between US and Chinese AI models.
These are today's updates.
Build and scale an AI governance program with Vanta*
Jensen Huang says AI safety should be left to the industry
The US faces a major shortage of chip workers
OpenAI and Anthropic earn 10x more than China’s AI models
Tools, jobs, resources, and last issue’s winning prompt ⬇️
Top News
At Salesforce's Dreamforce conference, Jensen Huang pushed back on the idea that AI is some kind of “alien mind,” a phrase an OpenAI safety researcher used earlier this month. He wants the industry to regulate proven harms, not hypothetical ones, and argues that industry is better positioned than government to keep AI safe. Huang has stuck to this position for months, most recently advocating that companies should own the safety burden while regulators stay out of the way.
The people building these systems do understand the failure modes better than any outside body could. But it also puts a lot of weight on trust, since it asks security and risk leaders to accept that self-regulation is working, often before there’s a clear way to verify it.
That’s the tension AI governance is facing right now. AI adoption is moving faster than regulation, leaving companies to work out what AI governance should look like inside their existing security programs. Cybersecurity leader Jane Frankland joins Vanta's GRC experts for a session on exactly this: building AI governance into your security program, assessing agents against your risk tolerance, and preparing for the EU AI Act, ISO 42001, and NIST AI RMF.
It’s happening on September 29th. Save your seat*
A new McKinsey and SEMI Foundation report projects the US will be short between 127,000 and 157,000 semiconductor workers by 2030. Samsung and SK Hynix are flying in engineers from South Korea just to get their new US fabs running, and sending American staff overseas for months of hands-on training since the expertise mostly lives in Asia.
Micron now runs expedited campus hiring sprees at South Korean universities, offering permanent jobs the same day. Universities are also racing to catch up. Purdue and Arizona State are building new chip-focused programs, with ASU converting an old Motorola fabrication facility into a training cleanroom funded by Applied Materials.
Billions of dollars from the CHIPS Act have helped fund chip factories in the US. But building a skilled workforce takes much longer. At Micron, training a single factory technician can take nearly two years. Meanwhile, demand for AI chips is growing every month. The SEMI Foundation has now formed an industry advisory committee to build a national strategy for training the workforce.
For companies counting on US-made chips to shorten supply chains or reduce their dependence on Asia, this creates a major gap, with benefits still years away.
Rhodium Group estimates OpenAI's annual recurring revenue at $40 billion and Anthropic's at $65 billion. All of China's AI companies combined generate around $10.7 billion, with ByteDance leading at $4 billion and DeepSeek trailing at just $500 million. Yet investors are pricing Chinese AI startups as if the growth story already happened. Notably, Moonshot trades at 50 times revenue and DeepSeek at 163 times, compared to 21 times for Anthropic and 34 times for OpenAI.
Source: Rhodium - Comparison of US vs China AI models’ annual recurring revenues, 2026 latest available
The mismatch is attributed to how these companies make money. Chinese models are mostly open-source and cheap to run. This got them users fast but left little revenue for the labs themselves. That’s because anyone with the hardware can download and run the model without paying the developer anything.
According to Rhodium, Chinese AI labs depend heavily on strong stock markets to keep raising money and expanding but its stock market has not been very reliable at supporting that kind of growth. While China’s AI progress may look strong based on model performance, the funding needed to keep that progress going could be harder to sustain.
Signals
Tools
Resources
Poll
Choose the prompt you want us to publish next
Market
Funding
Crusoe raises $3.9B to build modular AI data centers
Factory raises $200M to scale AI coding agents for enterprises
Roles in AI
Head of Product Design at Glean
Trending
The debate over who should govern AI, and how, is dividing the industry into two camps: self-governance and stronger government oversight. Anthropic has released an article outlining three ways it is tracking the pace of AI development. The article gives a peek into what self-governance could look like.
How Anthropic is measuring the pace of AI development
Prompt of the Day
AI Training Needs Analyzer
When to use this?
Use this before an important leadership meeting. Write everything you know and have the AI find the argument, gaps, trade-offs, and decision you actually need to communicate.
You are an AI Training Needs Analyzer for an enterprise team.
I’ll give you information about a team, department, or group of employees. Analyze where they need AI training and turn it into a practical training plan.
Assess:
1. Current AI use cases and workflows
2. AI skill gaps by role
3. Tasks that could benefit from AI but currently don’t
4. Risks caused by low AI literacy, poor prompting, or unsafe AI use
5. Training priorities based on business impact and urgency
6. The type of training needed: AI fundamentals, prompting, workflow automation, tool-specific skills, data/privacy, governance, or advanced AI use
7. Who needs training first and why
8. Recommended training format and duration
9. How to measure whether the training worked
Return the analysis in this format:
- **Training Priority:** High / Medium / Low
- **Audience:** Role or team
- **Skill Gap:** What employees currently struggle with
- **Business Impact:** What this gap affects
- **Recommended Training:** Specific skills or topics to teach
- **Suggested Format:** Workshop, hands-on lab, self-paced course, coaching, etc.
- **Time Needed:** Estimated training time
- **Success Metric:** How we should measure improvement
- **Next Step:** The most practical action to take
Then create a 30-60-90 day AI training roadmap, starting with the highest-impact gaps.
Base recommendations on the information I provide. Clearly flag assumptions where information is missing.
Here is the team information:
[Paste team structure, roles, AI tools used, current workflows, skill levels, and business goals]
Get more such prompts in the Prompting Playbook (free for you)
Stay curious, {{first_name | reader}}
PS. If you missed yesterday’s issue, you can find it here.
