The dominant story today is the widening split between the brute-force scale-up camp and the efficiency camp, and the money is still overwhelmingly betting on scale even as the evidence piles up that it doesn’t have to. That’s a mistake worth calling out plainly: infrastructure spending is running years ahead of proof that more compute buys proportionally more value, and today’s stories quietly make the case for the other side.
Exhibit A for team scale: a consortium led by Abu Dhabi’s MGX, the AI Infrastructure Partnership, and BlackRock’s GIP just closed a $40 billion buyout of Aligned Data Centers, reportedly the largest data center deal ever done. Meanwhile Google just raised its 2026 capex guidance to $195–205 billion, and investors are visibly nervous about it, record cloud revenue notwithstanding. And AMD just locked in a gigawatt-scale chip partnership with Anthropic — a compute commitment measured in power-plant units, not GPUs. That’s roughly a quarter-trillion dollars of conviction, in one day’s headlines, that the winning move is to build more.
Exhibit B, the skeptics, is smaller but sharper. NASA’s JPL flew a 4-bit, 4-billion-parameter Gemma 3 model aboard a Loft Orbital satellite and hit 88% accuracy on a 7,960-image benchmark with zero space-specific training data. No fine-tuning, no bespoke dataset, no giant cluster — just a well-built small model doing real work in orbit. European startup kausable just raised a €12 million seed to build models that adapt without constant retraining cycles, which is a direct bet that the retrain-everything treadmill Google and Anthropic are funding is a cost center, not a moat. And Mistral, per Sifted, is reportedly abandoning the frontier-scale race with Anthropic in favor of a Palantir-style enterprise and government playbook — another signal that even well-funded labs are hedging against the idea that bigger models win the whole market.
Put those together and the picture is: the capital is scaling, but the smart operators are increasingly betting on distribution, efficiency, and deployment rather than raw parameter counts. Somebody’s going to be very right and somebody’s going to be writing down a lot of data center debt.
Elsewhere, the alt-data and legal beats offered a reminder that the underlying data itself is getting messier to both use and protect. Consumer Edge’s release shows BNPL usage nearing record highs while lower-income households show growing strain — a genuinely useful signal for anyone pricing consumer credit risk, and a reminder that alt data’s best use case is still spotting stress before the balance-sheet data catches up. Anna Money’s acquisition of Business Data Group and UK Business Forums (terms undisclosed) is a small, sensible bolt-on for SME data. And the Fourth Circuit’s ruling that border agents can hand-search phones with no suspicion required is a reminder that data-rights law is being written case by case, far behind the pace at which data itself is being bought, sold, and trained on.
Tomorrow, watch whether Mistral’s Palantir pivot produces an actual named government contract — that’s the number that will tell us if the efficiency bet is paying off or just a hedge.
Stories covered
- kausable Bags €12M Seed to Build ‘Reasoning-First’ AI That Skips Retraining (Deals & Funding)
- Consumer Edge: BNPL Usage Nears Record High, Strain Hits Lower-Income Cohorts (Alt Data)
- NASA Flies Google’s Gemma 3 in Orbit With Zero Space-Specific Training Data (AI Training Data)
- MGX, AIP, BlackRock’s GIP Close $40B Buyout of Aligned Data Centers (Deals & Funding)
- Google Lifts 2026 AI Capex to $195–205B, Even as Investors Balk (AI Training Data)
- Fourth Circuit: Border Agents Can Hand-Search Phones With No Suspicion (Licensing & Legal)
- Sifted: Mistral Pivots From Anthropic Rivalry to Palantir Playbook (Deals & Funding)
- AMD Locks In Gigawatt-Scale AI Chip Deal With Anthropic (AI Training Data)
- Anna Money Buys Business Data Group and UK Business Forums, Price Undisclosed (Deals & Funding)
Rhea Rundown is an AI-assisted column persona of The Data Commenter; every column is reviewed by an editor before publication. Nothing here is investment advice.
