The trial wrapped Friday. Anthropic put a trillion on the table.

news · anthropic / capital · may 12

Why for you: The round would put Anthropic above OpenAI's $825B for the first time. The specific number matters: $30B is equity capital going directly to Anthropic — separate from Google's $40B compute commitment and Amazon's $25B infrastructure pledge. If it closes at or near $950B, Anthropic becomes the most valuable private AI company, a position OpenAI has held since the beginning. Sacra puts Claude's ARR at $30B as of April 2026, up from $9B at year-end 2025, with eight of the Fortune 10 as paying customers. Whether the deal closes at $900B or $950B is less interesting than the signal: the market believes Claude's enterprise trajectory justifies a valuation that implies the next phase of growth is still ahead, not behind.

news · openai / strategy · may 16

Why for you: Merging ChatGPT (consumer), Codex (developer tooling), and the API under a single product org led by Greg Brockman is a consolidation move: one coherent strategy rather than three teams competing for resources and roadmap priority. The timing is pointed — the announcement lands four days before Google I/O, signaling that OpenAI has its organizational house in order just as Google is about to show the world what Gemini can do in production. The long game here is the desktop superapp: ChatGPT, the Atlas browser, and Codex unified into a single experience that handles research, code, and agentic tasks without switching contexts. Unified product leadership is a prerequisite for that to actually ship.

interview · stripe / agentic commerce · may 16

Why for you: Collison on the record, speaking slowly: why keyword search is "ridiculous," how AI agents change the purchasing intent signal Stripe reads at checkout, and whether AI can actually replicate human taste. Not a product announcement — this is him explaining how Stripe thinks the next decade of commerce works. The argument that matters: agentic commerce is not just "AI buys things faster" but a wholesale change in who initiates purchasing intent. Where today's payment flow starts with a human decision, tomorrow's starts with an agent inference. Stripe's entire Sessions suite — Agentic Commerce Suite, Link Agent Wallet, Machine Payments Protocol — is built on that premise. Worth reading as the architectural thesis behind last Thursday's launches.

Also from this session: The Block on Collison's stablecoin prediction for agent-to-agent settlement

preview · google / ai strategy · may 17

Why for you: Google I/O opens Tuesday at 10am PT. What's coming: a new Gemini model, Android 17 with Gemini Intelligence embedded at the OS layer (camera, screen context, third-party app integration), first public preview of Android XR glasses, and Googlebook software announcements. The strategic frame worth tracking: Google's bet is that Gemini so embedded in Android's daily operation creates a structural moat, not just a feature advantage — when the intelligence layer becomes the operating system, switching costs compound. Everything OpenAI and Brockman's newly unified team does this week will be interpreted against whatever Google shows Tuesday.

analysis · openai / legal · may 15

Why for you: TR's weekly trial recap on the closing arguments: Musk's team arguing Altman misused charitable assets to build a for-profit juggernaut; OpenAI's lawyers arguing the nonprofit structure was never viable at the capital intensity AI actually requires. The framing buried in the piece is the useful one: AI spending from Google, Meta, Microsoft, and Amazon in Q1 2026 exceeded three times the total cost of the Manhattan Project. Musk's original vision of a scrappy open-source nonprofit was not financially plausible at that scale. Which doesn't resolve the legal question, but does explain why Altman's fundraising argument keeps landing with the defense — and why the framing of "stolen charity" struggles against the evidence of what frontier AI actually costs.

update · stratechery · may 2026

Why for you: Thompson in one update: Apple's hardware succession story — Ternus as the continuity candidate for the post-Cook era, with a hardware-first strategy that bets the AI moment ultimately accrues value to whoever controls the device; the SpaceXAI deal structure and what it means for xAI's compute position relative to OpenAI; and what the Cursor acquisition implies for who controls the coding agent layer. The through-line: all three are about who owns the stack that AI runs on — Apple through device hardware, xAI through inference at scale, Cursor through developer tooling. Dense even by Thompson's standards, and most of the value is in the juxtaposition.

product · xai / ai coding · may 2026

Why for you: xAI enters the agentic coding CLI space with a terminal-native tool targeting professional software engineering, app development, and workflow automation — initially available on the SuperGrok Heavy tier. The competitive context: Claude Code leads SWE-bench Verified at 87.6%; Cursor leads IDE integration; Codex CLI has Microsoft and OpenAI behind it. Grok Build is xAI's opening bid on the part of the stack where engineers actually live — the terminal, not the chat window. Worth watching whether Musk's infrastructure advantage translates to benchmark performance, and whether it ships fast enough to matter before the field consolidates around two or three dominant tools.

essay · sean goedecke · may 2026

Why for you: Goedecke's argument: until recently, experience in software engineering compounded — the longer you worked, the better you got, because the craft rewarded depth and accumulated judgment. AI coding tools may break this model. If agents handle increasing fractions of implementation, the thing that got better with experience — writing code, debugging, holding mental models of large systems — is being automated away before the next generation can build those skills at depth. His comparison to professional athletes whose physical peak has a ceiling is apt: software engineering may be shifting from a compounding career to one with a performance cliff. Short piece, worth reading if you're responsible for hiring or developing early-career engineers right now.

essay · noahpinion · may 2026

Why for you: Smith runs down the empirical case against the Washington assumption that China operates with a centuries-long strategic horizon. The evidence: Chinese policy is frequently reactive, shaped by factional politics and short-term pressures, not a grand master plan executed over generations. The AI policy angle: a significant portion of U.S. export controls and AI governance decisions are calibrated against China-as-perfectly-coordinated-long-term-adversary. If Smith's read is right, some of those decisions are solving for a threat model that does not actually exist — which changes the analysis of how tight the controls need to be, and who bears the cost of getting that wrong.