🔮Sunday edition #527: The AI race; vibe hacking; DeepSeek & inference; Ukraine's drones, AI cures & baby longevity choice++
An insider's guide to AI and exponential technologies
Hi, it’s Azeem.
The data shows AI-designed medicines could already save lives. Insilico's AI-created pill just proved it can slow lung disease progression in patients – the strongest clinical proof yet that AI can collapse drug discovery timelines. Meanwhile, Meta’s taking entire nuclear reactors offline and 3.5 million chip orders reveal the infrastructure war raging behind the scenes. We’re at the inflection point.
Lucky for you, you’re in exactly the right place to understand what comes next. Let’s go!
Playing to win
Future competition in AI won’t be decided by innovation alone, but by the ability to shift and absorb constraints, especially across compute, energy, talent, and what
calls “absorption capacity.” Nvidia’s Blackwell chips, which halve AI training time, are being snapped up by U.S. hyperscalers – over 3.5 million orders to date – offering early movers a critical edge in velocity. Velocity is the core advantage in AI but it depends on controlling both compute and power. That’s why companies like Meta, Microsoft and Google are locking up long-term power deals with nuclear facilities, taking entire gigawatt-scale plants off-grid to secure reliable, emissions-free energy for AI workloads. Meta’s 20-year contract for the full output of Illinois’ Clinton reactor is taking an operating reactor fully off the public grid, the first such deal of its kind. Europe, on the other hand, is currently playing a losing game. I wrote about the consequences of this yesterday:If Europe fails to build an energy system fit for an exponential economy, it will quietly deindustrialise by default, even while celebrating its climate goals on paper.
See also:
DeepSeek-R1 has caused inference workloads to explode. According to Nvidia’s VP of Accelerated Computing, Ian Buck, the new large-language model caused token generation to grow 13x higher than before. That surge is driving up demand for Nvidia GPUs and boosting profit margins across AI infrastructure.
Asymmetry, asymmetry everywhere
This week, the exponential transformation of warfare came into stark relief. Ukraine’s “Spider’s Web” operation, deploying 117 cheap, commercially available drones to disable one-third of Russia’s strategic bomber fleet, delivered a $7 billion shock to the nuclear status quo. Simultaneously, we’re seeing cyberweapons evolve at similar velocity – call it vibe hacking.
Tools like XBOW – AI-powered penetration testers – can now breach most web apps autonomously, shrinking exploit cycles from months to hours. In this environment, power no longer rests with the most resourced actor, but with the one best adapted to high-speed, software-first conflict. Cloudflare’s Matthew Prince told me that defenders still hold the advantage, for now. They operate across broad attack surfaces with rich telemetry. But the edge is shrinking.