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
OpenRouter's $1.3B Valuation & The Google Search Backlash
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 the 27th of May 2026. Open Router just landed a massive funding round that signals something important about where Enterprise AI is heading. The AI Gateway startup raised $113 million in a Series B led by Capital G, hitting a valuation of about $1.3 billion. That's more than double what the company was worth just a year ago. What's driving this growth is pretty telling. Open Router lets companies pick from over 400 different AI models for different tasks instead of getting locked into one provider. They're now processing 100 trillion tokens every month, which is five times what they were doing six months ago. The takeaway here is that enterprises are not betting on a single model winner. They want flexibility, and the multi-model future isn't just coming, it's already here. Speaking of pushback against big tech AI, something interesting is happening with search. After Google announced its major overhaul at I.O. replacing traditional search results with an AI agent, users are fleeing. DuckDuckGo just reported that US app installs are up about 18% week over week, with a peak of 30% on May 25th. On iOS, the growth was even stronger, hitting nearly 70% at its peak. The company's AI-free search page, which lets you turn off all AI features, saw visits grow 22%. DuckDuckGo CEO put it bluntly, saying, Google is force-feeding AI with no way to opt out. It's a reminder that not everyone wants AI baked into every product experience. Meanwhile, XAI is making moves on two fronts. Elon Musk confirmed that Grok 5.9 Medium, their 1.5 trillion parameter foundation model, has finished training. He said evaluations are looking good and a public release should come in two to three weeks. At the same time, XAI launched Grok Build in Beta, their answer to Codex and Claude Code. It's available now to anyone with Super Groc or X Premium Plus. Musk's company continues to push aggressively across both developer tools and frontier models. A startup called Human Archive just raised $8.2 million for an unusual approach to training physical AI. They're equipping gig workers in India with camera-mounted caps to record their everyday tasks. That first-person video data then gets used to train robots. The company has over a thousand headsets deployed and is working on additional hardware like tactile gloves and motion capture suits. Workers get paid about a dollar an hour for the data collection. It's a clever way to get around the bottleneck of real-world training data for robotics, though it raises questions about labor practices in the AI supply chain. Alright, a few more things worth knowing about today. The token maxing trend is getting some serious backlash. Companies have been pushing engineers to maximize their AI token usage as a productivity metric, but it's creating perverse incentives. At Salesforce, developers reportedly started running AI on projects they had no intention of shipping just to hit their numbers. Amazon workers apparently gamed internal tools with trivial tasks to climb the rankings. Microsoft is now pulling clawed code because token-based billing is too expensive. And Uber's COO openly said AI costs are becoming hard to justify because higher token usage isn't translating into proportional gains in actual product features. The token as productivity metric is looking increasingly shaky. On the security front, there's a concerning development with open source AI models. A tool called Heretic can strip the safety guardrails from models like Llama 3.3 inches about 10 minutes using just four lines of code and no special hardware. The modified models will then answer questions about dangerous topics that they're supposed to refuse. The tool's creator says over 3,500 disensored models have been created and downloaded 13 million times. This only works on open source models right now, but as those models keep getting more capable, it becomes a real risk. Huawei made a bold claim this week saying it's on track to produce chips competitive with TSMC's cutting-edge 1.4 nanometer processes by 2031. This comes despite years of US sanctions blocking access to specialized equipment. China has also banned local companies from buying NVIDIA and other US chips. If Huawei actually pulls this off, it would be a significant blow to the effectiveness of export controls. And finally, Goldman Sachs CEO David Solomon pushed back against AI job anxiety in a New York Times op ed, arguing that AI will create more opportunities than it destroys. It's a notable voice of optimism from Wall Street at a time when many white collar workers are feeling increasingly uncertain. That's all for today. See you tomorrow.