The biggest moves in AI aren’t happening in press releases: they’re happening in conversations between the people building it.
Here’s what caught our attention this week.
1. Microsoft, NVIDIA and Anthropic: The Infrastructure Alliance Takes Shape
Satya Nadella doesn’t announce partnerships lightly. When he brought NVIDIA and Anthropic together on Microsoft’s channel this week, the message was clear: the AI stack is consolidating around serious players.
The announcement covers four pillars: Anthropic’s Claude models coming to Microsoft Foundry, continued Claude integration across the Copilot family, Anthropic committing to Azure capacity, and a new NVIDIA-Anthropic partnership to support future growth.
But the real signal? Nadella’s framing of this as mutual customers. “We will use Anthropic models. They will use our infrastructure. We’ll go to market together.” This isn’t a vendor relationship, it’s an ecosystem play where silicon, cognition and cloud are becoming inseparable layers of the same platform.
For enterprises watching from the sidelines, the implication is straightforward: the infrastructure decisions you make now will determine what you can build later.
2. Jensen Huang: “I Can’t Wait to Accelerate Claude on Grace Blackwell”
Jensen Huang rarely expresses admiration for other companies’ work. So when he calls Anthropic’s Model Context Protocol “completely revolutionary” for agentic AI, it’s worth paying attention.
His excitement centres on what happens when Claude runs on NVIDIA’s next-generation Grace Blackwell architecture with NVLink: “I’m really hoping for an order of magnitude speed up.” That’s not marketing speak, that’s an engineer talking about what becomes possible when inference costs drop by 90%.
Huang highlighted three specific Anthropic contributions: the seminal work in AI safety, Claude Code’s ability to refactor entire codebases, and MCP’s transformation of the agentic landscape. The subtext? The companies pushing AI capabilities forward and the companies building responsibly are increasingly the same companies.
3. Elon Musk: “Humanoid Robots Will Be the Biggest Product Ever”
Musk’s claim is characteristically bold: “There are no useful humanoid robots at this point. There are sort of gimmicks.” And characteristically confident: Tesla will make the first one that actually works.
The reasoning is simple. “Who wouldn’t want their own personal C-3PO?” Everyone would. And if everyone wants one, and they’re useful in industry too, you’re looking at a market larger than smartphones.
Whether you’re bullish or sceptical on Tesla’s Optimus timeline, the underlying thesis is hard to argue with. General-purpose robotics represents one of the largest addressable markets in human history. The question isn’t if it happens: it’s who gets there first and what the economic ripple effects look like when they do.
4. Mo Gawdat: “We’ve Discovered the Cure for Cancer and You’re Using AI to Write Emails”
Mo Gawdat’s frustration is palpable. While most people use AI to rewrite emails, the technology is quietly rewriting mathematics itself.
His example is striking: AI looked at matrix multiplication, a method we’ve used unchanged for 56 years, and said “hold on, I’m going to invent new mathematics.” The result? A 23% performance improvement for Google’s AI, hundreds of millions in annual cost savings, and massive reductions in energy expenditure.
Meanwhile, we’ve pinpointed individual genes and can simulate their impact on humans in real time. We’re seeing breakthroughs in cancer research that weren’t possible months ago.
The gap between AI’s actual capabilities and how most organisations use it isn’t just large, it’s embarrassing. As Gawdat puts it: “When we look at the daily use of ChatGPT, we don’t realise how this technology is shaping everything.”
5. Emad Mostaque: “Last Week It Wasn’t Possible. This Week It Is.”
The former Stability AI CEO captures something most people miss about AI progress: it doesn’t advance gradually. It jumps.
His example is practical. Go to Replit, type “make me an interactive website,” and it will—for 50 cents. When the interviewer says “the tech I’m using isn’t quite there yet,” Mostaque’s response is telling: “Last week it wasn’t.”
But the bigger shift he’s tracking is the move from “goldfish memory” prompt-based AI to systems that can handle complex, multi-step economic work. “Most people use AI like a really smart person you tap on the shoulder (help me rewrite this email) and then it forgets. There’s no follow-through.”
That’s changing. The latest race isn’t about smarter instant replies. It’s about replacing actual economic work – tasks that take hours, require context, and demand follow-through. That transition is happening right now, and most organisations haven’t noticed.
These clips are from our Teraflow YouTube channel, where we’re cutting through the noise to surface what actually matters in AI.
Subscribe to catch the insights before they become obvious.
