For business owners
The AI race is now about power and agents
November 9, 2026 in summary: Anthropic is buying 1.7 gigawatts of gas-fired generation to power its data centres, Apple is pushing to turn Siri into a real agent, and DeepSeek's agent app passed 1.5 million daily users. The model race is becoming an infrastructure and agent race.
- Daily brief
- AI news
- Automation
November 9 was the day the AI race stopped looking like a contest over models and started looking like a contest over everything around them. The three stories that mattered today were about electricity, an assistant trying to become an agent, and an agent app growing faster than anyone expected. Read together, they say the same thing: the bottleneck has moved. It is no longer just who can train the smartest model. It is who can power it, and who can turn it into something that does work on its own.
The power story is the biggest one, literally. Anthropic has reportedly struck a $3.4 billion deal for 1.7 gigawatts of gas-fired generators to power its AI data centres, a purchase the Financial Times calls the largest single power equipment buy by an AI lab, according to The Neuron. That is not a normal procurement announcement. When a software company starts buying power generation at the scale of a small city, it is telling you where the real constraint sits. Chips get the headlines, but the machines have to run somewhere, and somewhere now means a power plant. The labs that can secure energy at this scale will be the ones that can keep training and serving bigger models, and the ones that cannot will be renting capacity from someone who can. Energy is becoming a competitive advantage in AI the same way it is in manufacturing, and the bill for it will eventually show up in the price of the tools everyone else uses.
The second story is about what the models are for. Apple's new chief executive John Ternus is reportedly pushing software chief Craig Federighi to move faster on AI and turn Siri into an agent that keeps working on goals on its own, rather than just answering questions, according to PYMNTS reporting on Bloomberg's Mark Gurman. That is a telling shift in ambition. Siri has spent years as a voice interface for simple commands. An agent that keeps working on a goal is a different product entirely, one that plans, acts, and follows through without being asked at every step. Apple arriving late to this idea is itself the news. The company with the most devices in people's pockets is now racing to make its assistant behave the way the newest AI products already do, and the push from the top suggests the timeline has shortened.
The third story shows how fast that behaviour is catching on. DeepSeek says its open-source agent app Harness has hit 1.54 million daily users in under three months, burning about 92 trillion tokens a day, with a plugin marketplace on the way, according to TradingView summarizing Bloomberg. Those numbers are worth pausing on. An open-source agent app reaching a million and a half daily users that quickly means the appetite for software that does things, not just chats, is already mainstream. It also means the agent market is not going to belong to one company. An open option that anyone can run and extend puts pressure on every closed product to justify its price, and the plugin marketplace is the part that makes it sticky: once developers build on top of it, the app becomes a platform.
What this means for a small business
Three practical takeaways. First, expect AI costs to keep climbing with energy prices, and budget accordingly. The labs are now buying power plants, and that spend has to come from somewhere, so lock in pricing where you can and keep an eye on usage-based bills. Second, start thinking in agents, not chatbots. The products winning today are the ones that complete a task, whether that is following up with a lead, booking a job, or chasing an invoice, and the gap between a chatbot and an agent is where the real time savings live. Third, keep an open option warm. DeepSeek's Harness shows that capable open-source agents are arriving fast, and a business that can switch between providers, or run its own, is not hostage to one vendor's roadmap or price increases. Power and agents are the two fronts that will decide what AI costs and what it can do for you next year, and both moved today.