THE SWELL #18
SAIL Weekly Digest | May 18 – May 22, 2026 | Issue #18
This week, SAIL authors tracked the shifting financial fault lines of the AI boom. A blockbuster Cerebras IPO, the creeping “token anxiety” inside every AI budget, and Anthropic’s march into mainstream business all point to a market finally putting a price on AI scale — while the contest with China played out over talent, prestige, and open models. The throughline: intelligence is getting cheaper to run and more expensive to win.
The Week in AI
Google I/O 2026 — Google launched Gemini 3.5 Flash on May 19, immediately making it the default model in the Gemini app and Search worldwide, alongside Gemini Omni Flash, the Spark agent, and Android XR glasses — while Gemini 3.5 Pro slipped to June.
OpenAI files to go public — In a frantic May 21 stretch, OpenAI filed for an IPO (~$25B ARR against an $852B valuation) while Anthropic circulated prospectus-style numbers showing a projected $10.9B in Q2 revenue, up 130% quarter over quarter, and its first quarterly operating profit.
Anthropic locks in compute — Anthropic agreed to pay SpaceX roughly $1.25B per month through May 2029 — about $45B in total — as its raise of $30B+ at a $900B-plus valuation nears close, a figure that would eclipse OpenAI’s valuation for the first time.
This Week from SAIL Authors
The Trump–Xi Moment & the Policy Game
Catching up? We covered the lead-up last week in ChinaTalk's Xi-Trump to talk AI Safety, Huh? and Macartney to Mar-a-Lago.
Watching the Super Bowl of US-China Nerds — Kevin Xu frames the Trump–Xi event as the championship game for US-China AI watchers: a lot of noise, with the long game still ahead. — Interconnected. Read more
Prestige on the Cheap — ChinaTalk reads the Trump–Xi summit through Cold War history, asking what real leverage looks like beneath the pageantry. ChinaTalk. Read more
Doing Big Things in Policy — Jordan Schneider talks with Kumar Garg and Remco Zwetsloot about finding the “white space” in policy and steering scarce talent toward neglected, high-impact work. — ChinaTalk. Read more
The Economics of AI Scale
Cerebras and the IPO pop — Cerebras’s 107% first-day jump is less a hype signal than evidence the market is finally pricing in real demand for AI inference. — Exponential View. Read more
Data to start your week: The cost of tokenmaxxing — Tokens are now one of the fastest-growing, least predictable line items in the AI stack — goodbye fixed costs, hello token anxiety. — Exponential View. Read more
Exponential View #574 — Anthropic pushes into Main Street as Microsoft and OpenAI consciously uncouple, plus the case for AI pluralism against a sycophantic model monoculture. — Exponential View. Read more
Open Models & Efficient Architectures
Latest open artifacts (#21): Open model bonanza — Gemma 4, DeepSeek V4, Kimi K2.6, MiMo 2.5, and GLM-5.1 all landed, and a CAISI evaluation weighs how far open models still trail the American frontier. — Interconnects. Read more
Recent Developments in LLM Architectures — A tour of how new open-weight models cut long-context costs through KV sharing, multi-head compression, and compressed attention. — Ahead of AI. Read more
AI Hits the Real World
Why it might not make sense for you to own a self-driving car — A ride in Tensor’s prototype makes the case that full-autonomy hardware may be too expensive to personally own, leaving robotaxis as the likelier future. — Understanding AI. Read more
AI-Native Healthcare: 100M Doctor Visits, 10–20 Hours Saved — Abridge’s Janie Lee and Chai Asawa on scaling clinical-documentation AI across 250 health systems, and why healthcare’s fatal downside risk shapes how you build. — Latent Space Read more
Notes on AI, labor, and China — An expansion pack to Jasmine Sun’s New York Times essay on AI, work, and Silicon Valley’s growing fear of a “permanent underclass.” — Jasmi.News. Read more
Full Library
Access the complete, searchable archive of SAIL Media in our Sitemap.
Doing Big Things in Policy — Jordan Schneider & Phoebe Chow, ChinaTalk
AI-Native Healthcare: 100M Doctor Visits, 10–20 Hours Saved, Prior Auth in Minutes — Swyx, Latent Space
Exponential View #574: Inside Anthropic’s rocket ship — Azeem Azhar, Exponential View
Why it might not make sense for you to own a self-driving car — Timothy B. Lee, Understanding AI
Recent Developments in LLM Architectures: KV Sharing, mHC, and Compressed Attention — Sebastian Raschka, Ahead of AI
Latest open artifacts (#21): Open model bonanza! — Nathan Lambert & Florian Brand, Interconnects
Cerebras and the IPO pop — Azeem Azhar, Exponential View
Notes on AI, labor, and China — Jasmine Sun, Jasmi.News
Data to start your week: The cost of tokenmaxxing — Azeem Azhar & Hannah Petrovic, Exponential View
Additional content on the SAIL Substack — supplementary coverage of the Trump–Xi event:
Watching the Super Bowl of US-China Nerds — Kevin Xu, Interconnected
Prestige on the Cheap — Jordan Schneider, Kevin Xu, Sergey Radchenko & Lily Ottinger, ChinaTalk
