Snap used Tuesday to reposition itself as more than a camera company. The SPECS Intelligence launch pairs the hardware refresh detailed in a companion release, SPECS, with an “anticipatory” AI layer, meaning the pitch has shifted from smart glasses to a service that acts before you ask. It’s the same wager Humane and Rabbit made and lost money on, except Snap already has the camera install base those two never had. Augmodo is chasing the same shrink-the-AI-into-a-wearable idea from a different angle: its new Owl Mini and Owl Pro smartbadges put physical AI on a lanyard instead of a face, aimed at frontline workers rather than consumers.
Faraday Future is making a louder bet on the same physical-AI thesis. At its annual 919 event the struggling EV maker plans to unveil nine AI-powered robots and four industry solutions, folding them into what it calls Robot World 2.0. A company that has spent years struggling to ship cars at volume is now positioning itself as a robotics platform, which says something about where the capital and attention currently sit relative to where FF’s actual production numbers sit.
Money kept moving toward that same bet. Bain Capital Ventures closed a $1.6 billion fund explicitly framed around founders building for a “post-AGI” world, an unusually confident label for a venture thesis. Vantora raised more than $100 million from Silversmith Capital Partners to build AI-native operating companies inside industrial enterprises rather than sell software to them. CADDi closed a $114 million Series D at a $1.2 billion valuation for manufacturing-data AI. And Hang Ten Systems added $53 million led by Temasek-backed Xora, bringing its total to $85 million. Four separate checks, four separate theses, all resting on the same underlying belief that AI is about to touch physical industry in a way software-only rounds never did.
The infrastructure underneath all of it got its own coverage. Rune debuted RELIC, a solar-powered data center, with $40 million in new funding, betting that “drama-free” power sourcing becomes a selling point as grid strain becomes the industry’s most talked-about constraint. That constraint got a harder look from MSI, whose new report argues America’s power buildout is gated by execution, not demand. Demand for AI compute isn’t the bottleneck anymore. Permitting, interconnection queues, and turbine lead times are.
Quantum computing had its own moment on the wire, split between the commercial and the research side. D-Wave is hosting Qubits Asia 2026 in Seoul, part of the broader Humanoids Summit Seoul push that’s landed official backing from South Korea’s Ministry of Science and ICT, a sign Seoul wants to be a physical-AI and quantum hub rather than just a memory-chip capital. On the research side, IonQ, Oak Ridge National Lab, Nvidia, and the University of Tennessee published work showing an AI method that reduces the usual quantum optimization trade-off, the kind of incremental result that matters more to people tracking the field closely than to headline readers, but it’s the sort of thing that eventually shows up in a product announcement a year or two out.
Not everything in AI this week was expansion and funding. Two releases pointed at where the technology creates new problems rather than new markets. Cohesity introduced Agent Resilience, a product built specifically to protect and recover AI agent infrastructure, which is itself an admission that agents are now a distinct attack surface worth a dedicated product line. And a new benchmark found that political bias grows as adoption of Chinese AI models accelerates, a finding that’s going to get cited a lot in the next round of arguments about model sourcing and export controls.
The trust question showed up on the consumer side too. Experian’s latest research found more than half of consumers are comfortable letting AI agents apply for credit on their behalf, which is a bigger number than most people in fintech would have guessed a year ago. It sits awkwardly next to a separate survey out this week finding only 7% of fashion shoppers trust AI to buy for them. Consumers, it turns out, will hand an agent something as consequential as a credit application faster than they’ll hand it a shopping cart. Trust isn’t tracking stakes the way anyone modeling agentic commerce assumed it would.