This fortnight the news is about what AI costs and how it runs, not what it can do. Factory electricity bills are climbing near data centres. Most enterprise AI pilots in India still can't show a return. The standards that let AI agents work together merged under one body, and Washington moved to control even the compute AI rents. Each one affects a real number in your business.
1. The Standards Under AI Agents Just Merged
On 20 August, Google gave its A2A protocol, which lets AI agents talk to each other, to the Linux Foundation's Agentic AI Foundation. The same body runs Anthropic's Model Context Protocol. Members include AWS, Anthropic, Cloudflare, Google, Microsoft and OpenAI. The two main standards for connecting agents are now with one neutral group, not competing vendors. Tools built on shared standards can be swapped or combined without rebuilding your integrations. Less lock-in to any single supplier, and multi-step automation becomes practical for a smaller firm.
What this means for you:
- Tools built on shared standards can be swapped or combined without rebuilding your integrations.
- Less lock-in to any single supplier, and multi-step automation becomes practical for a smaller firm.
Action tip:
Ask any AI tool vendor whether they support MCP and A2A. Prefer the ones that do, so you can change provider later without redoing the work.
2. The AI Boom Is Showing Up on Factory Electricity Bills
In the grid region covering much of America's manufacturing heartland, the wholesale price of power capacity has risen more than tenfold in two years, largely because of data centres built to run AI. Belden Brick, a 141-year-old maker in Sugarcreek, Ohio, went from $1,600 a month to $12,000. Plaskolite, a plastics manufacturer, went from $200,000 a year to $1.2 million. The Dallas Fed estimates data centres have already pushed wholesale electricity up 2 to 6% nationally, more where they cluster.
What this means for you:
- Power is one of your largest costs, and the price is now partly set by demand you have no part in. Data centres go up in a year or two; power stations take far longer.
Action tip:
Find the capacity charge on your bill. That's where the steepest rises are. Then weigh the usual levers: shifting heavy production off-peak, a fixed-term supply rate, or on-site generation.
3. India Adopts AI Fast, but the Returns Aren't There Yet
A new report drawn from over 300 senior Indian technology leaders found that 60% are still at the pilot stage after two or more years of investment. Only about one in eight can point to a proven return. Adoption itself is strong, with roughly 40% of Indian firms reporting significant AI use, above the global average.
What this means for you:
- The problem is rarely the model. The tool never gets built into how the work runs.
- Speed of deployment, not model power, separates the firms that see a return from the ones that don't.
Action tip:
Take one process, set one financial target, and hold the project to it before widening. Pick a tool that fits an existing workflow over a platform you have to build around. This is the discipline AIFM is built to enforce.
4. The AI Jobs Story for Graduates Is Messier Than the Headlines
US unemployment for recent graduates hit 5.7% in June, against 4.1% for all workers. Stanford research finds early-career workers in AI-exposed roles have lost ground since late 2022. But the New York Fed calls remote work a bigger cause than AI, and a University of Chicago economist notes that firms using the most AI aren't hiring fewer people.
What this means for you:
- Cut junior roles because AI seems to cover them, and you starve your pipeline of skilled staff later.
- The firms handling it well change what junior people do rather than removing them.
Action tip:
Before freezing junior hiring, list which of their tasks AI does well and which still need a person learning the trade. Rebuild the role around the second kind.
5. Washington Moves to Control the Compute AI Rents
A US agency is reviewing how Chinese AI firms use Nvidia chips they can't buy, by renting the computing power through data centres in Thailand, Malaysia and Japan. Current export controls cover owning the chips, not renting access to them. A proposed Remote Access Security Act would close that.
Cloud compute is becoming something governments track and restrict, not just hardware. Rules aimed at one country can change where compute is available and what it costs for everyone renting it.
What this means for you:
- Rules aimed at one country can change where compute is available and what it costs for everyone renting it.
Action tip:
For any plan that depends on rented cloud compute, keep a second provider ready and ask your vendor where its capacity physically sits.
That’s all for this fortnight. The theme this fortnight is the cost and the wiring of AI, not the models. Bills rise near the compute. Pilots still miss their return. The standards and the supply underneath are being reset by industry and governments at once. The edge goes to leaders who watch these as closely as the launches.
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.