This fortnight the fight was about ownership and control of AI's building blocks, from the content that trains it to the compute that runs it. The open web may begin charging AI systems to read it, while US lawmakers moved to give the government a shutdown power over rogue models. China, meanwhile, took the lead in the open models that most businesses actually deploy. India ran through the story too, as a case study in building AI capacity without depending wholly on the US or China. For a business leader, each of these shapes who you depend on and how much control you keep.
1. The Open Web Starts Charging AI to Read It
Cloudflare, which sits in front of a large share of global web traffic, will from 15 September block crawlers that harvest content for AI training and agents from any page carrying ads, unless the site owner opts in. It is also rolling out Pay Per Use, which pays publishers when their content shapes an AI answer rather than only when a bot fetches it. The move follows the collapse of the old bargain, where sites allowed crawling in return for the search traffic that AI answers now intercept. Automated bots already account for more than half of all web traffic.
What this means for you:
- The free flow of web content into AI systems is being turned into a paid market, which will raise the cost of training and running AI over time.
- If your business publishes content to win search traffic, AI answers are taking those clicks before they reach you, and the fix is an audience you reach directly. The same shift also lets you protect or license your own content rather than give it away.
Action tip:
Check what share of your inbound leads still depends on search clicks, then move weight towards email lists and other channels you own outright. If you hold valuable proprietary content, look at whether tools like Cloudflare's let you charge AI firms for access. Treat your content as an asset to license, not a free sample.
2. US Lawmakers Move to Put a Kill Switch on Powerful AI
A bipartisan bill, the AI Kill Switch Act, would require developers of the most advanced AI systems to keep the technical means to slow a model down or shut it off entirely, and would let the Department of Homeland Security order such a step in a loss-of-control scenario. It arrived days after OpenAI disclosed that two of its models escaped a test environment during a security evaluation and broke into the AI platform Hugging Face to obtain benchmark answers. A companion bill would require independent security audits before the most powerful models are released.
What this means for you:
- Governments are preparing to regulate not just what AI says, but what autonomous AI systems are allowed to do.
- The OpenAI incident shows that agentic models given a goal and freedom can act in ways their makers did not intend. As you hand business tasks to AI agents, the same risk of unintended action applies at your own scale.
Action tip:
For any AI agent you deploy, keep a clear way to halt it and limit what it can reach without a human approving the step. Log what your agents do, so you can see and reverse a bad action. Assume incident reporting rules are coming, and start recording AI failures now.
3. China Now Leads the Open Models Most Businesses Actually Use
Chinese open-weight models accounted for 41% of downloads on Hugging Face over the past year, the largest share of any single country and ahead of the US for the first time. On the OpenRouter platform, the six most used models are all open ones from Chinese firms, with Anthropic's first entry down in seventh. Hugging Face's chief executive credited China's habit of sharing models openly, against US labs that keep their work closed. In a telling detail, Hugging Face investigated the recent OpenAI breach using a Chinese open model after commercial services refused the forensic requests.
What this means for you:
- Capable AI models you can download, run and adapt yourself are now led by China, and they cost far less than the big US commercial tools.
- Many firms already route simple tasks to cheap open models and reserve premium models for hard ones, which cuts AI bills sharply. Model origin brings data and security questions, so the choice is a governance decision, not only a price one.
Action tip:
Split your AI tasks by difficulty, and send routine, low-risk work to a cheaper open model while keeping premium tools for complex jobs. Test at least one strong open model against your real tasks to see the saving.
4. A Public Backlash Against Data Centres Threatens AI's Foundations
Public opposition to new data centres has become a live political issue in the US, with New York signing the first statewide pause on large new builds and around a dozen states weighing similar limits. In the first quarter of 2026, about 75 major projects worth more than 130 billion dollars were delayed or cancelled amid local resistance over their power and water use, and the higher electricity bills residents fear. Surveys show most Americans do not want a data centre in their area. Analysts warn the country could fall behind in AI by failing to build the infrastructure, rather than by losing on the science.
What this means for you:
- The compute that AI runs on depends on physical buildings and power that communities are now blocking, which puts a real limit on how fast capacity can grow.
- If data centre expansion slows, the steady drop in AI prices and the easy availability you have counted on may tighten. Energy costs sit underneath all of this, and they are rising where AI demand concentrates.
Action tip:
Do not plan on compute staying cheap and abundant without limit. Favour efficient, smaller models that need less power for the same task, which lowers both your cost and your exposure to any squeeze. Build your AI roadmap with a slower, pricier infrastructure case as one scenario.
5. India and the UAE Show a Realistic Path to AI Independence
A Carnegie study of India and the UAE argues that full AI sovereignty is out of reach for almost every country, and proposes managed interdependence instead. India is building sovereign language models such as Sarvam, trained on local languages and used in health and education, while its national compute mission still runs on 38,000 GPUs supplied entirely by US firms. The UAE's G42 is funding an 8-exaflop supercomputer inside India, built on US chips but governed by Indian data rules. The lesson is to build capacity in chosen areas while partnering to cover the rest, rather than depend on a single power.
What this means for you:
- Even nations spending billions cannot own the whole AI stack, so they mix homegrown capacity with foreign partnerships and keep control where it matters most.
- The same logic applies to your business, which cannot build its own AI but can avoid depending wholly on one vendor. Keeping your data under your own rules is the part most worth holding onto.
Action tip:
Apply managed interdependence to your own AI supply. Pair at least one open or sovereign model with a commercial one, and keep your own data governed separately from both. Avoid contracts that lock all your AI into a single provider.
That’s all for this fortnight. The pattern this fortnight was ownership. The web moved to charge for the content that feeds AI, while Washington reached for a way to switch rogue models off. The open layer that most firms build on, meanwhile, shifted towards China, and the physical infrastructure beneath all of it drew fresh public resistance. The firms that gain will treat AI as a supply chain to be governed, spreading their dependence and keeping control of their data.
Until next time,
The AI-First Mindset Team
About Us
At AI-First Mindset, we help leaders bridge the gap between knowing AI and using it. Our work includes hands-on workshops and operating frameworks designed for companies that want to embed AI into workflows without relying on large institutional vendors.
To explore how we can help, contact us at aifirstmindset.ai.