Mistral's $830M Data Center & Space-Based AI Infrastructure

Mistral AI raises $830M for Paris data center, Rebellions AI chip startup hits $2.3B valuation, Starcloud gets $170M to build data centers in space, plus Americans' growing distrust of AI and Anthropic's new peak-hour surge pricing.

Transcript

SPEAKER_00 0:01

Hey, welcome to AI Mornings with Andreas Vig. It's Monday, March 31st, 2026. Mistral AI, the French AI lab, just secured 830 million US dollars in debt financing to build a data center near Paris. The facility will run on NVIDIA chips and is slated to be operational by the second quarter of this year. CEO Arthur Mensch says the goal is deploying 200 megawatts of compute capacity across Europe by 2027, emphasizing that European AI infrastructure autonomy is critical. Mistral has now raised over$3.1 billion total, making it one of the best funded AI companies on the continent. Speaking of AI infrastructure, a South Korean chip startup called Rebellions just raised$400 million at a$2.3 billion valuation. The company designs chips specifically for inference, which is the compute needed when AI models actually respond to user queries. They also announced two new products, Rebel Rack and Rebel Pod, which are infrastructure platforms for large-scale AI deployment. Rebellions has now raised$850 million total, with$650 million of that coming in just the last six months. They're expanding into the US, Japan, Saudi Arabia, and Taiwan. Here's one of the more ambitious ideas I've seen lately. A startup called StarCloud just raised$170 million to build data centers in space. They've already launched a satellite with an Nvidia H100 GPU and ran an AI model on orbit, which is apparently a first. The company says space-based data centers could become cost competitive with terrestrial ones by 2028 or 2029, assuming SpaceX's Starship brings launch costs down to around$500 per kilogram. The Series A valued StarCloud at$1.1 billion, making it a unicorn just 17 months after Y Combinator Demo Day. On the infrastructure management side, a company called Scaleops raised$130 million at an$800 million valuation. Their software automatically manages and reallocates computing resources in real time, and they claim it can reduce cloud and AI infrastructure costs by up to 80%. Customers include Adobe, Wiz, DocuSign, and Salesforce. The company has seen 450% year-over-year growth. Alright, a few more things worth knowing about today. A startup called Kodo raised$70 million for AI-powered code verification. As AI coding tools generate billions of lines of code monthly, the next bottleneck is making sure that code actually works. Kodo ranked number one on Martian's code review bench at 64.3%, beating Claude's code review by 25 points. The founder makes an interesting point. Code generation and code verification require fundamentally different approaches, and LLMs alone can't understand organizational standards or tribal knowledge. Enterprise customers already include NVIDIA, Walmart, and Texas instruments. A new Quinnipiac poll reveals something of a paradox about AI adoption in the US. AI use is up, with only 27% saying they've never used AI tools, down from 33% a year ago, but trust is way down. 76% say they trust AI rarely or only sometimes, and just 21% trust AI-generated information most of the time. 55% think AI will do more harm than good in their daily lives. 70% believe AI will lead to fewer job opportunities, up from 56% last year, and 65% don't want data centers built in their communities. From that same poll, 15% of Americans said they'd be willing to work for an AI boss, meaning a program that assigns tasks and sets schedules. That's not a huge number, but it's not zero either. The trend some are calling the Great Flattening is already happening, with companies like Amazon deploying AI workflows that replaced middle management layers. Finally, Anthropic has started implementing peak hour usage limits for Claude. During weekdays between 8 a.m. and 2 p.m. Eastern, users on free, pro, and max tiers hit their session limits faster, with each prompt counting as 1.5 to 2 units instead of one. Anthropic frames this as demand management. Users are calling it surge pricing. Either way, it affects about 7% of users, and the recommendation is to shift heavy background tasks outside those peak hours. That's it for today. See you tomorrow.