AI Mornings with Andreas Vig
Your daily AI news briefing in under 10 minutes. New models, product launches, research breakthroughs, and industry shifts, explained clearly, no hype.
AI Mornings with Andreas Vig
Amazon's $25B Anthropic Bet & Kimi's 300-Agent Swarms
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hey, welcome to AI Mornings with Andreas Vig. It's April 21st, 2026. Amazon just dropped one of the biggest AI infrastructure deals we've ever seen. The company announced it's investing another$5 billion into Anthropic immediately, with the potential to add$20 billion more down the line if Anthropic hits certain milestones. That would bring Amazon's total investment in the CloudMaker to a staggering$25 billion. In exchange, Anthropic has committed to spending over$100 billion on AWS over the next decade. This isn't just about cloud credits though. Anthropic is securing up to 5 gigawatts of computing capacity specifically on Amazon's trainium chips, including future Trainium 4 chips that don't even exist yet. The structure mirrors Amazon's recent$50 billion commitment to OpenAI's funding round earlier this year. It's a circular flow of capital. Amazon invests in the AI company, which then commits to spend that money right back on Amazon's infrastructure. And these compute deals are becoming the currency of the AI arms race. Kimi just released K26, their latest open source model, and it's pushing the boundaries of what agent swarms can do. We're not talking about a single AI assistant anymore. K2. 6 can coordinate up to 300 subagents executing across 4,000 coordinated steps simultaneously. That's three times the scale of their previous model. In one demo, the model ran autonomously for 13 hours, optimizing an 8-year-old financial trading engine and achieving a 185% throughput improvement. It also introduces something called claw groups, essentially letting agents from different devices and even different models work together under Kimmy's orchestration. The benchmarks are strong too. 66.7% on Terminal Bench 2 and over 70% on Suiebench Pro. This is open source by the way, available now. Alibaba's Quen team released an early preview of their next proprietary model, Quen 3 6 Max. It's topping six major coding benchmarks, including Suibench Pro, Terminal Bench 2, and Skills Bench. Compared to their previous model, the Max Preview shows significant gains in agentic coding tasks, nearly 10 points better on Skillsbench. The model supports a preserve thinking feature that's recommended for agenc workflows, keeping the reasoning trace across multiple turns. It's available through Alibaba Cloud Model Studio. Alright, a few more things worth knowing about today. Allbirds, the sustainable sneaker company, just pulled one of the wildest corporate pivots of the year. After selling off its brand assets for$39 million in March, down from a$4 billion IPO peak, the company announced a$50 million financing deal to reinvent itself as New Bird, AI, a GPU rental business. The stock shot up over 600% in a single day. Shareholders will vote next month to strip the company's public benefit status, formally ending its sustainable footwear mission. It's the kind of pivot we used to see with blockchain rebrands, but now it's AI Compute giving dying tickers a second life. SNAP is cutting a thousand jobs, about 16% of its workforce, and CEO Evan Spiegel is directly crediting AI productivity gains for the decision. The company says AI now writes 65% of its new code and handles over a million queries a month internally. They're reorganizing traditional teams into small AI-augmented pods, targeting$500 million in annual savings by the end of next year. The stock actually rose on the news, but the disconnect between what markets reward and what workers experience is getting pretty stark. OpenAI is quietly moving into advertising. A leaked pitch deck from ad tech company Stackadapt reveals they're selling ChatGPT ad placements based on what they're calling prompt relevance, targeting users while they're researching or comparing products. CPMs are reportedly as low as$15 during the pilot phase. It's the first formal step toward monetizing ChatGPT through ads, and it's happening through this partnership rather than being built in-house. Finally, Deezer dropped some striking numbers on AI generated music. 44% of all songs now uploaded to the platform are AI generated. That's about 75,000 tracks per day, over 2 million per month. But here's the catch. Actual consumption of AI music is just 1 to 3% of total streams, and 85% of those streams are flagged as fraudulent and demonetized. The platform has tagged over 13 million AI tracks since June of last year. An AI generated track also topped the iTunes charts in five countries last week, so this isn't going away. That's all for today. See you tomorrow.