AI News Roundup: The Week the Money Went to Chips, Not Chatbots
This week the funding didn't chase a smarter chatbot. It chased the hardware that serves the ones we already have. Positron AI closed $875 million at a $5 billion valuation on October 19 to build its Asimov inference chip, the kind of silicon that turns a trained model into a fast, cheap answer. Tenstorrent followed on October 20 with a $1.5 billion Series E, and Opsin raised $21 million to watch enterprise agents the way a security camera watches a door. The story of late October is that inference, not training and not the next flagship, is where the money thinks the durable business lives.
That shift makes sense once you look at the bills. Every chatbot query, every agent step, every image rendered burns compute someone pays for by the token. The labs that win the long game may not be the ones with the best model. They may be the ones who serve it for less. Positron and Tenstorrent are betting the moat is silicon, and a pile of investors just agreed.
Meanwhile the model side kept its own quiet rhythm. OpenAI is rushing a $30 billion round at a valuation above $1.4 trillion ahead of Anthropic's expected November IPO, according to Daily AI Brief, and Eric Schmidt was reported to hold about 1% of Anthropic from a years-old Series A. Apple, meanwhile, picked up the assets and some staff of Swiss audio startup Samplab on October 19, a small grab that tells you where the next on-device features are coming from.
The throughline: October split the AI economy into two stories. One is the flashy model race everyone watches. The other is the boring, enormous business of serving intelligence cheaply and safely, and that's where the biggest checks landed this week.
My read: the labs that survive won't be the ones with the highest benchmark. They'll be the ones who turned inference from a cost center into a margin, and this week's funding says the market agrees.
Updated October 28, 2026: funding figures below come from the original announcements.
Frequently asked questions:
Q: Which AI chip startups raised this week?
A: Positron AI closed $875 million at a $5 billion valuation on October 19 for its Asimov inference chip; Tenstorrent followed with a $1.5 billion Series E; Opsin raised $21 million for agent monitoring.
Q: Why is inference silicon the hot bet?
A: Every chatbot query, agent step and image burns compute someone pays for by the token. The labs that win the long game may be the ones who serve models for less.
Q: What does this mean for users?
A: Inference cost is most of what you pay per query. Better silicon pushes per-token prices down - we already saw Haiku 5.5 come in about 90 percent cheaper than its predecessor.