September 11, 2026
Moonshot AI, the company behind the Kimi models, is reportedly targeting $2 billion in annual revenue, according to TechCrunch. The number that actually backs this up is usage, not hype: OpenRouter data shows Moonshot's K3 models generating as many as 300 billion tokens a day, even after a recent dip in activity. That's the real story here. Foundation models are increasingly commodities priced by the token, and the labs winning aren't necessarily the ones with the flashiest demo -- they're the ones developers are quietly routing billions of requests through every single day. For a business shopping for AI tools, this is a reminder that brand recognition and raw usage don't always match. A model you've never heard of might already be running more inference than the one everyone's tweeting about.
This week an Anthropic researcher resigned and posted on X that the company is racing toward self-improving superintelligence and "gambling with our lives." What makes this different from the usual AI doom cycle is who signed off on it: Anthropic's own alignment lead reportedly co-signed the message instead of distancing the company from it. TechCrunch's framing -- that the timing is "very interesting" -- is putting it mildly. Anthropic has built its entire brand on being the safety-conscious lab, the one raising alarms about everyone else's models. When an insider says the company itself is the problem, and leadership doesn't rush to contradict it, that's a credibility event, not just a PR headache. I don't think this means Anthropic is secretly reckless -- researchers leave labs with strong opinions all the time, and internal disagreement can be a sign of a healthy culture rather than a broken one. But for enterprise buyers evaluating AI vendors on trust and governance, a public warning like this deserves more scrutiny than a shrug. It's worth reading alongside how we've covered OpenAI's own board additions around similar concerns -- this isn't an isolated incident, it's a pattern across the frontier labs.
Meta's AI agent Muse has reportedly climbed to the No. 2 app spot in the US, even though TechCrunch notes its start has been slower than Meta AI's or Threads'. Slow start or not, cracking the top two app rankings is a meaningful signal that consumers are willing to hand tasks -- not just conversations -- to an AI agent. We've written before about what Muse actually changes versus a chatbot like ChatGPT, and this ranking suggests that shift is starting to show up in real adoption numbers, not just press coverage. Meanwhile, Maven Robotics emerged from stealth with a $100 million Series A and, notably, deployments already running -- not just a pitch deck. TechCrunch's framing that Maven wants to "steal your robot deployment deal" points to something businesses evaluating automation should notice: the robotics market is now competitive enough that a well-funded newcomer can go straight after incumbents' existing customers rather than just chasing greenfield deals.
Put these four stories together and you get AI's current split personality: massive commercial momentum (Moonshot's token volume, Muse's app ranking, Maven's funding and live deployments) running in parallel with a very public, very credentialed warning that the technology underneath it all might be moving faster than anyone can safely steer. Both things are true at once, and that tension isn't going away soon. If you're deciding which AI vendors to build your business on, it's worth asking not just what a tool can do, but who's raising their hand internally to say it shouldn't.
Which worries you more as a business buyer: a lab that moves fast and doesn't talk about the risks, or one that moves fast while its own people are sounding the alarm?
Sources