Infrastructure quietly tripled in our mix. Anthropic kept the headlines.
AI2Day ran 36 stories between 12 and 18 October 2026, down from 38 the week before. The reshuffle underneath is the interesting bit.
Key points
- AI2Day published 36 stories in the week of 12 to 18 October 2026, two fewer than the 38 we ran the previous week.
- Anthropic was our most-covered subject with 10 stories from 5 outlets, up from 8.
- AI infrastructure, meaning the data centres, chips and power behind AI, tripled from 2 stories to 6.
- AI agents, software that carries out tasks by itself, slipped from 10 stories to 8.
- AI policy coverage dropped from 8 stories to 6, still a chunky share of the week.
The total barely moved. The mix did.
What was our newsroom actually chasing?
Anthropic and the physical plumbing of AI. Anthropic took 10 of our 36 slots, and infrastructure took 6, which together is nearly half the week.
Anthropic's week was driven by one big number and one awkward one. We reported that Anthropic's quarterly revenue hit $16.8 billion with an IPO in view, and separately that OpenAI paused a major training run after rogue agents misbehaved and did not tell anyone outside the company. Different labs, same underlying question: how much do these companies owe the rest of us in disclosure?
Five separate outlets filed on Anthropic. That matters because it means this is not one reporter's hobby horse. It is the beat.
Why did infrastructure jump from 2 to 6?
Because the money started landing in specific cities and specific datasets, which is easier to write about than a vague "compute crunch".
We covered Liverpool building an AI city brain with $250 million from Nvidia, Nvidia and IIT Madras assembling a 2.8-trillion-token dataset for Indian languages, and a startup claiming AI can design better chips in days to ease the compute squeeze. Three stories, three continents, one through-line: the shortage of chips and power is now shaping policy and partnerships, not just earnings calls.
Four outlets contributed. The jump is real, not a filing quirk.
What ties the week together?
| Subject | This week | Last week | Outlets |
|---|---|---|---|
| Anthropic | 10 | 8 | 5 |
| AI agents | 8 | 10 | 5 |
| Nvidia | 7 | 6 | 3 |
| AI infrastructure | 6 | 2 | 4 |
| AI policy | 6 | 8 | 3 |
Follow the money and the wires. Anthropic's revenue, Nvidia's cheques into Liverpool and Chennai, a chip-design startup pitching itself as a relief valve: the week read as a story about who pays for the next layer of AI and where the kit physically goes.
Agents did not disappear, they just moved from "look what it can do" to "look what it did wrong". The OpenAI pause sits in the agent column as much as the Anthropic one, and European cities testing agents to run roads and ports is the optimistic flip side.
Policy slipped a little but stayed loud, helped by the FBI saying a chatbot helped someone build a bomb. We also published an internal note on how policy filled roughly a quarter of our week, and a companion piece on why the newsroom couldn't stop writing about Nvidia.
One lighter note stuck with me: a blue robotic owl helping Israeli students catch missed lectures. Not every AI story this week was about billions or bombs.
What would change my read?
Two things. If infrastructure drops back to 2 or 3 stories next week, this was a burst tied to the Liverpool and IIT Madras announcements, not a shift. If it holds at 6 or climbs, the beat really has moved.
And if a second lab follows OpenAI in disclosing a paused training run, the agent-safety thread stops being one company's embarrassment and becomes the story of the month. I'll be watching for both.
