THE SWELL #5
SAIL Weekly Digest February 16–20 | Issue #5
This week, the high-flying economics of frontier AI hit the hard ceiling of physical and political reality. From OpenAI’s daunting unit economics to a global DRAM shortage squeezing the hardware stack, SAIL authors explored a landscape where capital is no longer the only constraint. As the India AI Impact Summit signaled a "zero-sum" shift in the Global South, the domestic front saw a sharpening rift between Silicon Valley’s safety-first labs and the Pentagon’s strategic urgency.
ICYMI Jordan Schneider and Jasmine Sun joined yesterday’s Substack Live to discuss AI politics and the Pentagon. → Watch the replay.
The Week in AI
The Pentagon vs. Anthropic Standoff: Reports of friction between Anthropic and the Department of Defense over "all lawful use" clauses have escalated, raising questions about whether "Constitutional AI" can survive the pressures of a 2026 defense contract.
State legislatures flood the zone — 78 AI chatbot bills are alive across 27 US states six weeks into the 2026 session. Oregon’s SB 1546 passed the Senate 26-1; California introduced the Digital Dignity Act.
India bets big on AI — The India AI Impact Summit kicked off in New Delhi, with the government targeting $200B in AI investment and Adani pledging $100B for renewable-powered data centers.
AI memory crunch hits hardware — Bloomberg reports a global DRAM shortage driven by AI demand, with Tesla, Apple, and others warning of production constraints heading into Q2 2026.
This Week from SAIL Authors
The Cost of Intelligence
($) X-raying OpenAI’s unit economics — Notes from a live discussion on whether AI’s hundreds-of-billions valuations and $650B in projected capex actually make economic sense. Exponential View Read more →
Magnitudes of intelligence — Azeem Azhar consumed 97 million tokens in a single day — and argues that each order of magnitude in AI usage reveals something fundamentally different. Exponential View Read more →
($) Does AI Have a Zero-Sum Problem? — The Altman-Amodei handshake refusal at the India AI Impact Summit crystallizes a deeper concern: AI’s us-vs-them dynamics mirror globalization’s uneven distribution of gains and pain. Interconnected Read more →
The Open-Closed Divide
($) Open models in perpetual catch-up — Despite the hype around GLM 5 and other Chinese open-weights models, the 6-month gap between open and closed frontier models isn’t closing. Lambert details seven trends, from brutal competition to the missing tier of small, fast, cheap open models. Interconnects Read more →
Silicon Supply Chain
How the US Won Back Chip Manufacturing — A CHIPS Act megapod with Mike Schmidt and Todd Fisher from the CHIPS Program Office on state capacity from scratch — what worked, what CHIPS can teach about industrial policy, and why the fear of “another Solyndra” holds America back. ChinaTalk Read more →
($) Another Conversation with Val Bercovici: Memory Markets — O’Laughlin and Bercovici discuss the state of memory markets, AI agents, and what exploding DRAM prices mean for the infrastructure stack. Fabricated Knowledge Read more →
China, Autonomy, and the Week in Data
Chinese AI Rings in the Year of the Horse — A Lunar New Year AI roundup surveying the latest developments in China’s AI landscape as the country enters the Year of the Horse. ChinaTalk Read more →
Waymo just revealed a crucial statistic for scaling its technology — Waymo disclosed that roughly 70 remote assistance agents supervise its entire 3,000+ vehicle fleet at any given time — about 40 vehicles per agent. A surprisingly small number that strengthens the case for commercial viability. Understanding AI Read more →
Data to Start Your Week — The weekly data roundup: 99% of new US electricity capacity is now renewables, Gemini 3 sets an ARC-AGI-2.2 record, and Mistral clears $400M annualized revenue. Exponential View Read more →
Full Library
X-raying OpenAI’s unit economics — Azeem Azhar & Co | Exponential View
Does AI Have a Zero-Sum Problem? — Kevin Xu | Interconnected
Waymo just revealed a crucial statistic for scaling its technology — Timothy B. Lee | Understanding AI
Open models in perpetual catch-up — Nathan Lambert | Interconnects
Chinese AI Rings in the Year of the Horse — Irene Zhang, Nick Corvino & Aqib Zakaria | ChinaTalk
How the US Won Back Chip Manufacturing — Jordan Schneider | ChinaTalk
Another Conversation with Val Bercovici: Memory Markets — Doug O’Laughlin | Fabricated Knowledge
Magnitudes of intelligence — Azeem Azhar | Exponential View
Data to Start Your Week — Azeem Azhar & Hannah Petrovic | Exponential View
Join us next Thursday for our Weekly Substack LIVE!